View Full Version : Next NV Infomania
Again, not any actual info, just mania about info. :lol:
This one has been even more quiet than R520.
What do we think we know, other than 110nm? Feeling confident it is a straight quad bolt-on, 16==>24? Anyone heard anything to lead them to believe they did a little tweaking here and there? Any new features?
Somewhere I got the idea that they'd probably be announcing before ATI, anyone care to put a date range on that? Does the fact that it is a refresh rather than a new gen mean we'd expect to see little-or-no leadup at nvidia.com?
Is there any evidence that anyone has actually seen one of these critters? For instance, were there some at CeBit?
digitalwanderer
14-Apr-2005, 14:57
I was under the impression that it wasn't coming until the Fall. :?
Skinner
14-Apr-2005, 15:23
grgrgrgrgr :evil:
I was under the impression that it wasn't coming until the Fall. :?
Really? For the 110nm part? And thus we won't see 90nm until spring 2006? The problem is, they've talked about seeing 90nm in 2H 2005, so if you back up 6 mos from there. . .
I found this from Wavey's CeBit article interesting:
both ATI and NVIDIA are dancing around each other a little at the moment, trying to catch each other by surprise
. . .as to whether it might mean something more than timing.
MuFu's rolling a "super NV43" possiblity out there was also intriguing.
At the moment, I'm not sure this 110nm part rumor isn't an elaborate ruse to mess with ATI's mind. . .until someone can say, "Yeah, I've seen one, it's real" at least.
For instance, I didn't hear anything in the call with the analysts in February that committed NV to having this part, or them even mentioning it. Indeed, when the subject of the refresh came up, what I got out of it was that NV felt they were very well positioned as is, and that ATI was the one with the need to do something because of their ancient overclocked (paraphrasing) line-up.
digitalwanderer
14-Apr-2005, 16:01
Oh no, I ain't basing that on nothing...it's just kind of what I thought.
I haven't kept up on nVidia's upcoming part because I didn't think it was on the horizon yet. :oops:
Hellbinder
14-Apr-2005, 17:51
Its got 64 "pipelines"
The way Nvidia counts them anyway.
The way Nvidia counts them anyway.
AFAIK nvidia counts pipelines the same way as ati does,
are you being sarcastic here? :)
Tim Murray
14-Apr-2005, 19:11
Unless he means 64 pipelines in the same way NV30 had 8.
Demirug
14-Apr-2005, 19:20
I am vote for (x*4)+1 Pipes. ;)
Are you implying NVIDIA is going to use a "one pipeline" approach for Vertex Shading, Demirug?
;)
Demirug
14-Apr-2005, 19:44
No I am talk only about the pixelpart of the chip. Maybe I should not call it one pipe. 0.9 Pipe or something like this should more correct.
Are you implying NVIDIA is going to use a "one pipeline" approach for Vertex Shading, Demirug?
;)
Or does the one pipe have water in it and is the one we're all passing around right now. Duuude, don't Bogart the pipeline!
I heard they were going to have "Double Stuffed" pipes and have a marketing deal with Oreos! :lol:
Nvidia will claim their graphics cards are way tastier than ATI's.
No I am talk only about the pixelpart of the chip. Maybe I should not call it one pipe. 0.9 Pipe or something like this should more correct.
Hmm. Has anyone seen Demirug and Dave in the same place at the same time? :lol:
wireframe
14-Apr-2005, 19:57
No I am talk only about the pixelpart of the chip. Maybe I should not call it one pipe. 0.9 Pipe or something like this should more correct.
Hmm. Has anyone seen Demirug and Dave in the same place at the same time? :lol:
I caugh....saw them together once, but I am sworn to secrecy!
(this is certainly a lie and hopefully also a joke)
Are you implying NVIDIA is going to use a "one pipeline" approach for Vertex Shading, Demirug?
;) Yeah, the "+1" stands for Cell. :)
wireframe
14-Apr-2005, 20:09
Ok, I am getting a headache now. Someone explain the notation in (x*4)+1. This is obviously not normal math because then there would be no reason for the parentheses.
Ok, I am getting a headache now. Someone explain the notation in (x*4)+1. This is obviously not normal math because then there would be no reason for the parentheses.
Sez you. I always use parenthesis in those situations. God forbid someone might think I mean x*(4+1).
digitalwanderer
14-Apr-2005, 20:26
Ok, I am getting a headache now. Someone explain the notation in (x*4)+1. This is obviously not normal math because then there would be no reason for the parentheses.
It's a joke from how nVidia counted the FX's pipes. ;)
wireframe
14-Apr-2005, 20:30
Ok, I am getting a headache now. Someone explain the notation in (x*4)+1. This is obviously not normal math because then there would be no reason for the parentheses.
Sez you. I always use parenthesis in those situations. God forbid someone might think I mean x*(4+1).
At first I thought I understood what was meant by (x*4)+1, but then it didn't add up for me and then the headache took over as I tried other permutations. I need help here!
If it was plain-jane math it would be traditional to write it as: 4x+1.
Thing is, I am not even sure what X is supposed to be. Unknown number of what? The 4 had me thinking "quad" but then I want to see 1/4 or something like that, thinking X is the number of pixel pipelines. So that each quad gets some shared resource that is not a pipeline, but almost exactly 0.9 of one. Gahh... I really am getting dizzy. Think I will avoid looking at it further and go for a walk and listen to some Keane.
wireframe
14-Apr-2005, 20:32
Ok, I am getting a headache now. Someone explain the notation in (x*4)+1. This is obviously not normal math because then there would be no reason for the parentheses.
It's a joke from how nVidia counted the FX's pipes. ;)
Ok, but I thought that was a 4*2 magically becoming a 8*1 under certain conditions. Forget it. I will just get dizzier. :?
Edit: argh! 8*0 even...no!!! make it stop!
Demirug
14-Apr-2005, 20:35
digitalwanderer, I don't make bad jokes.
wireframe, don't get confused from the parentheses. As geo I tend to use more parentheses than needed.
Maybe you all need a hint:
http://img3.picsplace.to/img3/1642/thumbs/nVidiaPipe.PNG (http://img3.picsplace.to/img3/1642/nVidiaPipe.PNG)
At first I thought I understood what was meant by (x*4)+1, but then it didn't add up for me and then the headache took over as I tried other permutations. I need help here!
If it was plain-jane math it would be traditional to write it as: 4x+1.
Gosh, you really think he was joking, Digi?
What I took from the way he wrote it was that he had three variables in the equation of "how many pipes", but two of them were known to him (or at least willing to be disclosed by him) so he wrote out the equation and filled in where the two variables were he knew.
So (A*B) + C = no# of pipelines. A is the number of quads, and he either doesn't know or is being coy. B is the number of pipes per quad. C is his mystery ingredient but there are only 1 of them, later changed to .9
Or Demi's sense of humor is just too dry for me to get. :lol:
What about that weird thing someone posted? NV_ENGR1 or something like that... what was that thing?
http://www.google.com/search?q=+NV_ENGR1&sourceid=mozilla-search&start=0&start=0&ie=utf-8&oe=utf-8&client=firefox-a&rls=org.mozilla:en-US:official
digitalwanderer, I don't make bad jokes.
wireframe, don't get confused from the parentheses. As geo I tend to use more parentheses than needed.
Maybe you all need a hint:
http://img3.picsplace.to/img3/1642/thumbs/nVidiaPipe.PNG (http://img3.picsplace.to/img3/1642/nVidiaPipe.PNG)
Gee, Demi, 8/9 is only .88888 (etc)!, not .9! :)
Someone PLEASE invent language/organizing principles that are user-friendly to replace our current pipeline paradigms that are becoming increasingly obfuscating rather than clarifying! It might not make you rich, but it could make you famous!
Ah, finally they implemented the built-in heatpipe :lol:
6,879,207
Demirug
14-Apr-2005, 20:56
Ah, finally they implemented the built-in heatpipe :lol:
6,879,207
Do you have to betray everything? :twisted:
Hellbinder
14-Apr-2005, 20:59
its more like ((1+1(+1/2))*8 + ((1+1(+1/2))*8
(of course you have to then divide that by the contents at the bottom Grouches trash can)
Vegetto
14-Apr-2005, 20:59
:o u really don't have any clues do you? :twisted:
I Hate you guys (http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=/netahtml/search-bool.html&r=1&f=G&l=50&co1=AND&d=ptxt&s1=nvidia&OS =nvidia&RS=nvidia) ;)
I Hate you guys (http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=/netahtml/search-bool.html&r=1&f=G&l=50&co1=AND&d=ptxt&s1=nvidia&OS =nvidia&RS=nvidia) ;)
Oh, thank goodness! Imagine what I was making from this one!: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=687920 7&dopt=Abstract
:lol:
I Hate you guys (http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=/netahtml/search-bool.html&r=1&f=G&l=50&co1=AND&d=ptxt&s1=nvidia&OS =nvidia&RS=nvidia) ;)
But a little more seriously. . .only one extra pipe per chip, rather than one per quad? Damn, the routing on that must be. . .uh, creative? Or did quads qua quads (say that four times fast!) just get flushed in this paradigm? Wavey said he isn't expecting quads to go away this time around.
Vegetto
14-Apr-2005, 21:07
anyway what are we talking about here? G70 or G80? :?
anyway what are we talking about here? G70 or G80? :?
Heh. Tell us what G70 and G80 are and then maybe we can tell you. :D At one time I thot it was settled that one was a notebook chip, but even that seems up in the air now. I saw Rys recently say no one has given a straight answer that he's willing to rely on yet. This is why the thread title is "Next NV".
Demirug
14-Apr-2005, 21:12
I Hate you guys (http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=/netahtml/search-bool.html&r=1&f=G&l=50&co1=AND&d=ptxt&s1=nvidia&OS =nvidia&RS=nvidia) ;)
But a little more seriously. . .only one extra pipe per chip, rather than one per quad? Damn, the routing on that must be. . .uh, creative?
You only have to make sure that the units are placed near together. A good floorbuilder should solve this. Anyway there is a second solution with two extra units. Maybe this is for the case when you have very bad yields.
Demirug
14-Apr-2005, 21:16
anyway what are we talking about here? G70 or G80? :?
Heh. Tell us what G70 and G80 are and then maybe we can tell you. :D At one time I thot it was settled that one was a notebook chip, but even that seems up in the air now. I saw Rys recently say no one has given a straight answer that he's willing to rely on yet. This is why the thread title is "Next NV".
G70 is IMHO a desktop chip. There are already 7 versions (2 for profi boards). IMHO to much for a mobil.
A side question: does someone know why nvidia abruptly changed its own GPUs naming scheme?
Vegetto
14-Apr-2005, 21:21
anyway what are we talking about here? G70 or G80? :?
Heh. Tell us what G70 and G80 are and then maybe we can tell you. :D At one time I thot it was settled that one was a notebook chip, but even that seems up in the air now. I saw Rys recently say no one has given a straight answer that he's willing to rely on yet. This is why the thread title is "Next NV". The big problem with nvidia is that in this moment they don't have a single core waiting to get released ( u may call it g70 or whatever ) as DB said ati and nvda are just dancing with each other and waiting. If there is something i know is that nvidia is waiting, the when ati shows the R520 they will choose what path is the best, it's only a matter of time right now :x
pd: excuse my lame english, it's not my primary language :oops:
Methinks:
G70 - 90nm @ either 4*8, 400-450 MHz or 4*4, 500-600 MHz + improved branch prediction and support for 3Dc (?)
G80 - 90/65nm @ either 4*8 or 4*10, 600+ MHz + SM4/WGF2
Dave Baumann
14-Apr-2005, 21:23
G70 is IMHO a desktop chip. There are already 7 versions (2 for profi boards). IMHO to much for a mobil.
Think DTR.
G70 is IMHO a desktop chip. There are already 7 versions (2 for profi boards). IMHO to much for a mobil.
Think DTR.
I thought we had concluded some time back G70 was the 6800 Ultra in Dell laptops. Is this correct?
Vegetto
14-Apr-2005, 21:28
Methinks:
G70 - 90nm @ either 4*8, 400-450 MHz or 4*4, 500-600 MHz + improved branch prediction and support for 3Dc (?)
G80 - 90/65nm @ either 4*8 or 4*10, 600+ MHz + SM4/WGF2 will you care to bet that it is 90nm indeed? or maybe 110nm?? that's a risky bet :P. Anyway u can bet that the branching speed will be improved, and who knows, maybe a few surprises :)
Do you have to betray everything? :twisted:
The style of the drawing was just too revealing ;)
Are you sure this isn't what they're already using in the NV4x, but on a quad base? I didn't read much, but is there anything that shows they're not talking about redundant quads?
btw, some very interesting stuff in there...
FIG. 2 is a block diagram of a graphics processing unit that benefits by incorporation of embodiments of the present invention. This graphics processing unit 220 is connected to a CPU/Northbridge 210 and a frame buffer 240. The GPU 220 includes a front-end or host 230 and several texture units, each coupled to a pair of geometry/shading processors 260. The geometry/shading processors 260 are coupled through an interconnection network 270 to several raster operation (ROP) units which are in turn connected to memory interface circuits 290. The memory interface circuits 290 write and read data to and from the frame buffer 240.
G70 is IMHO a desktop chip. There are already 7 versions (2 for profi boards). IMHO to much for a mobil.
Think DTR.
I thought we had concluded some time back G70 was the 6800 Ultra in Dell laptops. Is this correct?
That is NV42. Which is already plenty fast for DTR, IMO. They need a new desktop high-end part.
I thought we had concluded some time back G70 was the 6800 Ultra in Dell laptops. Is this correct?
I don't think so. It would make more sense that G70 stands for Geforce 7. The Go 6800U wasn't special enough to warrant a drastic name change imo.
will you care to bet that it is 90nm indeed? or maybe 110nm?? that's a risky bet . Anyway u can bet that the branching speed will be improved, and who knows, maybe a few surprises
I believe the 110nm rumor to be something nvidia purposefully invented so that when they annouce their next 90nm chip after ATI launches the R520 it will turn more heads.
Probably more surprises as well, like gamma corrected AA and SM3b. :wink: :lol:
Demirug
14-Apr-2005, 21:38
G70 is IMHO a desktop chip. There are already 7 versions (2 for profi boards). IMHO to much for a mobil.
Think DTR.
Data Terminal Read?
Data Transfer Rate?
Dedicated Token Ring?
Digital Trunk?
Desktop Replacement?
Dave Baumann
14-Apr-2005, 21:39
Desktop replacement. What reason do you think that it wouldn't end up there?
he GPU 220 includes a front-end or host 230 and several texture units, each coupled to a pair of geometry/shading processors 260
Umh..a vertex engine per quad?
This doesn't not represent current pixel pipelines (or ALU blocks) and vertex engines ratio on current nvidia GPUs.
Demirug
14-Apr-2005, 21:44
Xmas, the style of this is to different from the NV4X pipes in the patents.
Dave, maybe one of this 7 version are for DTRs but 7 Chips for this segment? As Xmas already wrote the need a new desktop chip and NV48 is it not.
G70 is IMHO a desktop chip. There are already 7 versions (2 for profi boards). IMHO to much for a mobil.
Think DTR.
I thought we had concluded some time back G70 was the 6800 Ultra in Dell laptops. Is this correct?
That chip thats in that notebook which I have now came in 3 weeks ago is the 42m, the G70 is the next gen desktop should be out at E3
Demirug
14-Apr-2005, 21:47
he GPU 220 includes a front-end or host 230 and several texture units, each coupled to a pair of geometry/shading processors 260
Umh..a vertex engine per quad?
This doesn't not represent current pixel pipelines (or ALU blocks) and vertex engines ratio on current nvidia GPUs.
This sounds more like a unified shader core. If you look at the page with the whole GPU you will see that there are no vertex and pixelshader blocks. There is only one shader block.
Dave Baumann
14-Apr-2005, 21:48
Dave, maybe one of this 7 version are for DTRs but 7 Chips for this segment?
OK, I thought that would be a given - when has a mobile chip been not been derived from a desktop chip these days?
This sounds more like a unified shader core. If you look at the page with the whole GPU you will see that there are no vertex and pixelshader blocks. There is only one shader block.
I had the same idea the very moment I browsed that patent.
As it doesn't obviously represent any current product by nvidia, have you any reason to believe it represents some future products? ;)
(p.s. you have a pm since years.. :) )
Demirug
14-Apr-2005, 21:58
I had the same idea the very moment I browsed that patent.
As it doesn't obviously represent any current product by nvidia, have you any reason to believe it represents some future products? ;)
(p.s. you have a pm since years.. :) )
I have read many patents from nVidia, ATI and others. IMHO this one contains to much details for something that is only a big theorie. But maybe this is a solution for the PS3.
Dave, maybe one of this 7 version are for DTRs but 7 Chips for this segment?
OK, I thought that would be a given - when has a mobile chip been not been derived from a desktop chip these days?
Somebody somewhere (Josh, mebbee?) was going down this line too in a recent article. But leading with the laptop part --actually shipping!-- before you even announce for the desktop?
And what about Razor upstream reporting it is a 42m in that laptop, in his actual possession?
G70 is IMHO a desktop chip. There are already 7 versions (2 for profi boards). IMHO to much for a mobil.
Think DTR.
I thought we had concluded some time back G70 was the 6800 Ultra in Dell laptops. Is this correct?
You aren't imagining it, we *did* conclude that, with Wavey's help and the report that the info came from inside NV. Others have been calling it into question since.
Dave Baumann
14-Apr-2005, 22:02
Somebody somewhere (Josh, mebbee?) was going down this line too in a recent article. But leading with the laptop part --actually shipping!-- before you even announce for the desktop?
Oh, I doubt they will, but the thoughts come out dependant on where indications were first seen ;)
And what about Razor upstream reporting it is a 42m in that laptop, in his actual possession?
NV42 (http://www.beyond3d.com/misc/chipcomp/?view=chipdetails&id=99)(M) is a 110nm TSMC version of NV41 and completely interchangable (except NV42 can scale higher) - desktop 6800's are already using it in some cases (in fact, dependnant on capacities I would suspect that NV41 will go soon as NV42 is very likely to be cheaper).
NV42 (http://www.beyond3d.com/misc/chipcomp/?view=chipdetails&id=99)(M) is a 110nm TSMC version of NV41 and completely interchangable (except NV42 can scale higher) - desktop 6800's are already using it in some cases (in fact, dependnant on capacities I would suspect that NV41 will go soon as NV42 is very likely to be cheaper).
I'm sure that was clear to you, but I'm stuck in language normalization hell right now.
So.
1). You're sticking with G70 is in that Dell laptop?
2). Razor is right/wrong about 42m being in that laptop? Because one way I have to conclude that 42m = G70, and the other I have to conclude that Razor doesn't know what is in his own machine. :)
3). And thus the next high-end desktop part from NV is. . .G80?
Dave Baumann
14-Apr-2005, 22:22
NV42 is NV42. G70 isn't NV42 AFAIK.
wireframe
14-Apr-2005, 22:25
A side question: does someone know why nvidia abruptly changed its own GPUs naming scheme?
I think it goes something like this:
Nvidia started out only thinking about graphics and making chips for them. It was natural to name the chips after the company, where NV=NVidia. They obviously need some numeric designator to increment so you get NV10, NV20, etc.
Then Nvidia started making other logic solutions, like chipsets for AMD's Athlon. These are nowadays designated Cn.
As the portfolio grows and markets are more finely segmented, it becomes more logical to rename all lines. So C, for chipset stays, the classic pure NV becomes G for graphics, etc.
I am not sure if we will be seeing GM or MG for mobile graphics, for example, or if these will be simply M or something else. I do think they will have their own designator because it become very cluttery having names like NV4c when you run out of decimal digits.
Oh, and it makes sense for them to not call it G50 instead of NV50 because...this way you align the chip name with the product name, Geforce 7 at the same time as you are cleaning up your paperwork. Makes it all neat and tidy.
Demirug
14-Apr-2005, 22:40
NV42 is NV42. G70 isn't NV42 AFAIK.
If we look at the deviceIDs of this chips they should not the same.
GeForce Go 6800 Ultra = 0xC9 (GeForce Go 6800 = 0xC8)
G70 = 0x90-0x94 (G70GL = 0x9d-0x9e)
wireframe
14-Apr-2005, 22:48
wireframe, don't get confused from the parentheses. As geo I tend to use more parentheses than needed.
Maybe you all need a hint:
http://img3.picsplace.to/img3/1642/thumbs/nVidiaPipe.PNG (http://img3.picsplace.to/img3/1642/nVidiaPipe.PNG)
Ok, this is beginning to make a strange sort of sense to me now. This is probably the part that you tell me it is and elaborate joke and I am getting the wider end of it shoved up my backside...
NV42 is NV42. G70 isn't NV42 AFAIK.
Personally I think G70 is NV42, its too low end to be NV47/8.
IMO, G70 is whats in the new Dell Inspiron XPS.
Although thats the same configuration of chip as the original Go 6800 I know its not - NV41 was a 130nm IBM based part whereas this is a 110nm TSMC part (hence the clock increase). Initially this was said to me to be NV42M, but I have reason to suspect that this is what G70 is now.
Voltron, NV40 is AGP, NV41M has always been PCI Express; all notebook configurations that have used Go 6800 so far are PCI Express based hence they have used N41M. The information on NV42M came from NVIDIA's mobile product manager.
G70 is supposed to be "2x NV41" and WGF1.0.
So you're evolving on better data, right? You're allowed, y'know. We won't think any less of you. :lol: I think the above is a pretty good synopsis of why many of the rest of us are chasing our tails on this one so far. :D Did you go back to mobile product manager guy and say "wassup?"
russo121
14-Apr-2005, 23:17
NV42 is NV42. G70 isn't NV42 AFAIK.
AFAYK what is G70? Describe it please! :D What characteristics?
IMO Nvidia will launch a new Pc part just after the launch, in the christmas period, of PS3. I think it will be a complete different design and based on the PS3 graphic chip.
Richard
14-Apr-2005, 23:24
wireframe, don't get confused from the parentheses. As geo I tend to use more parentheses than needed.
Maybe you all need a hint:
http://img3.picsplace.to/img3/1642/thumbs/nVidiaPipe.PNG (http://img3.picsplace.to/img3/1642/nVidiaPipe.PNG)
Ok, this is beginning to make a strange sort of sense to me now. This is probably the part that you tell me it is and elaborate joke and I am getting the wider end of it shoved up my backside...
I don't know but if some one could answer me these two questions I'd appreciate:
1) Is that ONE quad?
2) Is that a way to read back other pixels in flight?
NV42 is NV42. G70 isn't NV42 AFAIK.
AFAYK what is G70? Describe it please! :D What characteristics?
IMO Nvidia will launch a new Pc part just after the launch, in the christmas period, of PS3. I think it will be a complete different design and based on the PS3 graphic chip.
So you're joining ANova in the "110nm desktop high-end part is just a ruse!" camp? I'm not there, but I can see the attractiveness of that position --until someone tells me they've seen the beastie. Part of why I see the attractiveness is the lack of mention in the analysts meeting, tho possibly that was a tad early (mid-Feb). If there had been some indication there I think it would be very difficult for them to back away from it --you don't mess with those guys lightly.
Dave Baumann
14-Apr-2005, 23:37
Personally, I don't think there have been significant changes to NVIDIA's roadmap since they first mentioned the longevity of the NV4x lower end parts, other than the codenames have shifted.
russo121
14-Apr-2005, 23:48
Personally, I don't think there have been significant changes to NVIDIA's roadmap since they first mentioned the longevity of the NV4x lower end parts, other than the codenames have shifted.
And I have the impression Xbox2 will be fitted with a x800 variant :?:
Mordenkainen - don't know about quads and all that, but think of the
Ps as 9 processing units (P0 - P8), and 8 inputs/outputs (I0-I7, O0-O7).
If there is a defect in P3 (say), you shut it off and do things as follows :
I0 - P0 - O0
I1 - P1 - O1
I2 - P2 - O2
I3
\- P4 - O3
I4 - P5 - O4
I5 - P6 - O5
I6 - P7 - O6
I7 - P8 - O7
Since each input/output is connected to 2 pipes, there is a way to route the inputs to working "pipes" so long as 8 out of 9 of them are functional.
I have read many patents from nVidia, ATI and others. IMHO this one contains to much details for something that is only a big theorie. But maybe this is a solution for the PS3.
...which is supposed to come to the PC space "a little bit later this year".
However, it all seems weird assuming NVidia wants to have a WGF2.0 part out in 2006. NV40 was announced about one year ago. Development of this architecture is very expensive, and you'd think they want to do a refresh of it, and soon. After that, it'll be at least six months before they present their next big chip, presumably a "PS3 GPU for PC" (let's call it PGPU). If that one is not WGF2.0, they only have about a year until they present their next architecture.
I think that leaves only a few options:
- PGPU is just an NV4x architecture refresh
- PGPU is halfway to WGF2.0, so their first WGF2.0 part will actually be a refresh of this
- PGPU is WGF2.0 and won't come out this year
- PGPU is halfway to WGF2.0, so their first WGF2.0 part will actually be a refresh of this
This approach doesn't seem very popular in the R520 thread (other than cranks like me :D ), but then maybe there are enough differences between the two IHV's approaches to "WGF2.0" implementation to make it more viable for NV. Anyone care to comment?
NV42 is NV42. G70 isn't NV42 AFAIK.
AFAYK what is G70? Describe it please! :D What characteristics?
IMO Nvidia will launch a new Pc part just after the launch, in the christmas period, of PS3. I think it will be a complete different design and based on the PS3 graphic chip.
You can't be told what the G70 is, you have to see it for yourself :)
You can't be told what the G70 is, you have to see it for yourself :)
Apparently, "The first rule of G70. . ." :D
Also the second rule. :lol:
Personally, I don't think there have been significant changes to NVIDIA's roadmap since they first mentioned the longevity of the NV4x lower end parts, other than the codenames have shifted.
Okay, boys, that's got whopper big hint written all over it --somebody hunt down the reference please.
Richard
15-Apr-2005, 02:31
Mordenkainen - don't know about quads and all that, but think of the
Ps as 9 processing units (P0 - P8), and 8 inputs/outputs (I0-I7, O0-O7).<snip>
Yes I understood that from reading the patent, I'm just trying to see if there are... hmm... how should I put this; potential beneficial performance side-effects along with higher yields.
You know there is probably an email circling around Nvidia's offices where they are pointing to this thread and laughing their arses off at us chasing our tails over the G70.
If I was in a position to do it, I'd make up a code name. :twisted:
Personally, I don't think there have been significant changes to NVIDIA's roadmap since they first mentioned the longevity of the NV4x lower end parts, other than the codenames have shifted.
Okay, boys, that's got whopper big hint written all over it --somebody hunt down the reference please.
Oh, this one again: http://www.beyond3d.com/forum/viewtopic.php?t=20937
Ruseies roll up your banners and try again another gen. . .
Oh, btw, it was Josh at Penstar who bit on the laptop theory and apparently is still sticking with it:
Now comes the hard part. NVIDIA’s future products have been kept under very strict wraps. Very little has been leaked about their upcoming refresh parts, but we can still look around and see where they are possibly going. It seems that NVIDIA is abandoning the classic NVxx codenames and going with G series. The G70 is the first indication of change, and I believe that we may very well have already seen the first G7x part slipped under our noses. Last month NVIDIA released the GeForce Go 6800 Ultra to the market, and it is a native 12 pipeline/5 vertex shader product that is created for the desktop replacement market. This chip is clocked at a very surprising 450 MHz, and does not require extreme cooling to do so. For reference, the NV40 based GeForce 6800 Ultra is clocked at 400 MHz and requires a hefty cooler. In the reviews that were released, nothing was said about what process it was manufactured on, what chip it was based on, or any other truly relevant feature that could have pointed to the parentage of this product. The fact that it runs at 450 MHz should be a very good indicator that it is not from the NV40, NV41, or NV42 family of products. This is very reminiscent of several years back when NVIDIA released its GeForce 4 Go (though at introduction it wasn’t called that) a full four months before they released the official GeForce 4 Ti/MX series of products. That mobile product was based on the same NV17 core that the upcoming GeForce 4 MX series of products were based on. From past experience, it appears as though NVIDIA is not shy about releasing the mobile version of future desktop products early. The initial results of the GeForce Go 6800 Ultra were stunning, and it easily matched the desktop GeForce 6800 GT in every category.
http://www.penstarsys.com/editor/so3d/1h_2005/1h_2005_4.htm
Reverend
15-Apr-2005, 05:41
Personally, I don't think there have been significant changes to NVIDIA's roadmap since they first mentioned the longevity of the NV4x lower end parts, other than the codenames have shifted.
Okay, boys, that's got whopper big hint written all over it --somebody hunt down the reference please.
Personally, I don't think any "lower end" parts by any of the IHVs hold much interest to anyone significantly interested in this particular B3D forum. Maybe in the graphics board forum but not here. Whopper big hint or not. NV4x is just about as interesting to those that only ever visit this particular B3D forum as those interested in what can be done more significantly by NVIDIA to help make NVIDIA more money by cutting a couple of pipelines to meet the "lower end" market. But that's just my opinion.
Oh, btw, it was Josh at Penstar who bit on the laptop theory and apparently is still sticking with it:
http://www.penstarsys.com/editor/so3d/1h_2005/1h_2005_4.htm
Taking this into account with regards to current product definitions (high end / mid-range) the G70 now available (6800MU) should be evidence of the next refresh's high mid-range performance.
Oh, btw, it was Josh at Penstar who bit on the laptop theory and apparently is still sticking with it:
http://www.penstarsys.com/editor/so3d/1h_2005/1h_2005_4.htm
Taking this into account with regards to current product definitions (high end / mid-range) the G70 now available (6800MU) should be evidence of the next refresh's high mid-range performance.
You mean G71? :D
Which really means, I would think, that Wavey's "2x NV41" ought really be "2x NV42" (aka "G71" --tho so far as I know I just made that up as a joke, not that NV refers to it that way!).
Why can't we keep all this speculation bull in one thread? :evil:
All the same in like 10 threads. F'in frustrating.
Why can't we keep all this speculation bull in one thread? :evil:
All the same in like 10 threads. F'in frustrating.
But this is the official NV infomania thread...
Anyway, seen the fact that NV put enormous investments in nv40 we don't see that architecture go away, this or next year (as explained by nv)
given the timeline of development of the ps3, taping out of that gpu should be happening really soon, since the launch is somewhere q1/q2 of '06.
thus, technology from g70 should've been able to be incorporated in the ps3 gpu and thus we could see a high efficiency nv40 bundled with xdr ... certainly not comparable to the gf2 in the xbox...
DegustatoR
15-Apr-2005, 12:46
I believe that G70 is a development of NV40 (more quads, new process, some tweaks and improvements in shader cores) and it will be out after ATI launches R520 since NV don't have any rush in the top-end segment at the moment. Whether it's 110 or 90nm is hard to say at the moment -- the closer to 2H05 we are the higher is 90nm probability.
G80 on the other hand may well be the WGF2-architecture which will be incorporated in PS3 too and launched in the PC space in spring 06 alongside Longhorn itself.
I think PS3 GPU technology is more related to G80 than to G70, but since Uttar hinted to the fact that we shouldn't expect much more than SM3.0 on PS3 it's hard to believe G80 could be their WGF2.0 architecture.
As a sidenote, when will WGF2.0 software and hardware should mark their appearance on the market?
I'm not aware of Longhorn roadmap..
I think PS3 GPU technology is more related to G80 than to G70, but since Uttar hinted to the fact that we shouldn't expect much more than SM3.0 on PS3 it's hard to believe G80 could be their WGF2.0 architecture.
As a sidenote, when will WGF2.0 software and hardware should mark their appearance on the market?
I'm not aware of Longhorn roadmap..
h2 2006 for the software, so probably the refresh/launch in q2 2006 should feature wgf 2.0
NV42 (http://www.beyond3d.com/misc/chipcomp/?view=chipdetails&id=99)(M) is a 110nm TSMC version of NV41 and completely interchangable (except NV42 can scale higher) - desktop 6800's are already using it in some cases (in fact, dependnant on capacities I would suspect that NV41 will go soon as NV42 is very likely to be cheaper).
I'm sure that was clear to you, but I'm stuck in language normalization hell right now.
So.
1). You're sticking with G70 is in that Dell laptop?
2). Razor is right/wrong about 42m being in that laptop? Because one way I have to conclude that 42m = G70, and the other I have to conclude that Razor doesn't know what is in his own machine. :)
3). And thus the next high-end desktop part from NV is. . .G80?
I just checked with rivaturner, its definitly a nv42
this is what it has under traget adapter
NVIDIA GeFroce Go 6800 Ultra
nv42(A1,12x1.5vp) with 256mb ddr
DegustatoR
15-Apr-2005, 18:12
I think PS3 GPU technology is more related to G80 than to G70, but since Uttar hinted to the fact that we shouldn't expect much more than SM3.0 on PS3 it's hard to believe G80 could be their WGF2.0 architecture.
As a sidenote, when will WGF2.0 software and hardware should mark their appearance on the market?
I'm not aware of Longhorn roadmap..
Maybe it'll be G70 version in PS3 who knows (or G7x to be more precise). But i don't see any point in launching non-WGF2 G80 just before Longhorn is released. So either G80 will be launched alongside G70 or G80 is WGF2 architecture.
Maybe it'll be G70 version in PS3 who knows (or G7x to be more precise). But i don't see any point in launching non-WGF2 G80 just before Longhorn is released. So either G80 will be launched alongside G70 or G80 is WGF2 architecture.
Your point is valid but imho it's unlikely PS3 is going to get a 2005 GPU since it will not arrive until next spring.
Maybe Uttar was wrong..:)
Or perphaps longhorn schedule is going to slip another time..
Your point is valid but imho it's unlikely PS3 is going to get a 2005 GPU since it will not arrive until next spring.
Maybe Uttar was wrong..:)
Or perphaps longhorn schedule is going to slip another time..
with the xbox360 slated for an october release, don't expect the ps3 to be too far behind, if it were to be released in may next year then one certain electronics giant will have a serious problem
Wasn`t nVidia more like a partner to Sony in terms of lending a hand with the graphics chip, rather than building one?Has the situation/understanding changed?
with the xbox360 slated for an october release, don't expect the ps3 to be too far behind, if it were to be released in may next year then one certain electronics giant will have a serious problem
Even if Xbox360 will be a successful and powerful console with a wonderful launch the earth will still revolving around the sun 6 months later..
We're talking about GPUs here anyway so let's keep console debates off from this forum.
JoshMST
15-Apr-2005, 21:25
Haha, unfortunately I have bit on a lot of things in my time speculating, and this Go 6800 Ultra part is probably one of the smaller nibbles. Go 6800 Ultra is a NV42M, built by TSMC. TSMC apparently has been adding to the secret mix that goes into their 110 nm process, and we are starting to see some really high performnace ASIC's coming off of that line. While they are not doing low-k on 110 nm, they are continually optimizing and improving that process.
Looking from far away, I would say that 110 nm gives about 93% of the transistor performance of TSMC's 130 nm LowK, depending on the design of course. Still, 110 nm is starting to look more and more enticing to manufacturers, mainly due to how relatively inexpensive it is, as well as how it performs against higher priced Low-K designs. I still believe that NVIDIA will release their next high end product on 110 nm, and it will probably clocked upwards of 500 Mhz (pure speculation on my part).
Haha, it is starting to look like a complete turnaround from 3 years ago. ATI is going with the smaller process for its new part, while NVIDIA looks to go with an older, tried-but-true process for its next generation. It will be very interesting how this turns out!
Haha, it is starting to look like a complete turnaround from 3 years ago. ATI is going with the smaller process for its new part, while NVIDIA looks to go with an older, tried-but-true process for its next generation. It will be very interesting how this turns out!
True dat. For all our sakes I hope it turns out much more competitively this time than that one did!
Tim Murray
15-Apr-2005, 21:33
Haha, it is starting to look like a complete turnaround from 3 years ago. ATI is going with the smaller process for its new part, while NVIDIA looks to go with an older, tried-but-true process for its next generation. It will be very interesting how this turns out!
True dat. For all our sakes I hope it turns out much more competitively this time than that one did!
well... I don't know if I'd call it a COMPLETE turnaround. last time, NV was shooting for adding low-k and a process shrink in the same generation (which failed miserably--if NV30 was capable of 600Mhz without a dustbuster... things might have turned out far more competitively), and this time, we're just looking at a process shrink--low-K's already there. so I don't think we're going to have a catastrophic failures like we had with NV3x.
Chalnoth
15-Apr-2005, 21:37
It was more than just process problems that really screwed up the NV3x. One huge problem was still that they didn't properly anticipate how hard it would be to produce a compiler for the VLIW architecture of the NV3x (which is why you have massive performance differences between hand-tuned shaders and compiled shaders for the NV3x).
After all, if nVidia had gotten that part right, and had gone for a 8x1 instead of 4x2 architecture, we might have had a GeForce 6600 GT (albeit without SM3) instead of the GeForce 5800 Ultra.
It was more than just process problems that really screwed up the NV3x. One huge problem was still that they didn't properly anticipate how hard it would be to produce a compiler for the VLIW architecture of the NV3x (which is why you have massive performance differences between hand-tuned shaders and compiled shaders for the NV3x).
After all, if nVidia had gotten that part right, and had gone for a 8x1 instead of 4x2 architecture, we might have had a GeForce 6600 GT (albeit without SM3) instead of the GeForce 5800 Ultra.
I agree with you in that it was more than just process problems. But I don't agree with the rest of your post. NVidia has introduced a lot of techniques in the past (e.g. T&L, SM1, SM3, FP blending), but the first generation of NVidia hardware which introduced new techniques was always slow when executing the new stuff. I think with NV3x they did the usual job, namely implemented the new technology SM2, which was working more or less correctly in NV3x, but as usual they aimed quite low with the performance of the new technology. They simply didn't expect to have such a good competition. Add all the process related problems and the delays and suddenly the NV3x was disappointing - when compared to the competition.
You can see that NVidia aimed for fixed function performance mainly in that they kept using fixed function hardware and just bolted SM2 hardware on top of that. The SM2 hardware was just meant to work, but NVidia concentrated on fixed function performance. Compare that to ATI's approach. ATI dropped all fixed function hardware and optimized the hell out of their SM2 pipelines. NVidia just underestimated the SM2 performance of their competition.
At least that's my view of NV3x.
Dave Baumann
15-Apr-2005, 22:57
support for 3Dc (?)
Yes.
Chalnoth
15-Apr-2005, 23:00
I don't remember any performance problems related to the T&L in the GeForce 256, nor any performance problems with PS 1.1 in the GeForce3.
Galduta
15-Apr-2005, 23:25
http://www.vr-zone.com/?i=2010&s=1
http://img226.echo.cx/img226/8720/grito5jz.jpg
:shock: :shock:
In fact I don't believe it :)
Going 90 nm and multichip at the same time? wow..it's a risky bet.
Moreover shoulnd't be feasible to have a 32 pipelines single chip on a 0.90 process.
NVIDIA SLI implementation is good, but it's not rock solid..
Vegetto
16-Apr-2005, 00:03
It's never too risky to beat the other :P
support for 3Dc (?)
Yes.
I spoke with Tony Tamassi out in San Fran, and apparently there is already support for 3Dc in NV4x. They just can't call it 3Dc as ATI owns the rights to the name. NVIDIA haven't bothered with the fancy name, it's just another improved normal map compression technique AFAIK.
Chalnoth
16-Apr-2005, 00:46
http://www.vr-zone.com/?i=2010&s=1
Okay, that post has some really stupid things in it:
1. A 32-pipe NV4x part would be roughly 450 million transistors. I seriously doubt this will be viable just yet. ATI may be able to squeeze out 32 pipelines with a unified design, but since unified is unlikely, 32 pipelines is similarly unlikely.
2. The next high-end part from nVidia will be a refresh of the NV4x lineup, not a whole new architecture.
3. Dual-core is just stupid when talked about in reference to GPU's.
The only part that really seems to be well thought-out was the equipping of all products with turbocache. This will probably be the case, but it doesn't mean the high-end parts won't still have lots of onboard memory.
If I were a betting man, I'd be laying down a large chunk of cash on it being called GeForce 7.
Chalnoth
16-Apr-2005, 01:36
GeForce 7800 and friends would be more likely :)
GeForce 7800 and friends would be more likely :)
Well, in the same way that a GeForce 6800 is actually a GeForce 6, yes, you might be right.
Chalnoth
16-Apr-2005, 01:48
GeForce 6800 is actually a GeForce 6
Um, I don't exactly understand your reasoning. nVidia calls its 6x00 range of products the "GeForce 6 series" with no individual part being labeled a GeForce 6.
I see what you mean. Here's another go:
If I were a betting man, I'd be laying down a large chunk of cash on the new generation of products being called the GeForce 7 series.
The cooler bit seemed awfully detailed compared to what you usually see for a wild-ass flyer. Not that I'm buying --just saying.
digitalwanderer
16-Apr-2005, 03:47
The cooler bit seemed awfully detailed compared to what you usually see for a wild-ass flyer. Not that I'm buying --just saying.
True, but it also sounds like the kind of thing I'd make up to add to the description if I was faking it too.
The cooler bit seemed awfully detailed compared to what you usually see for a wild-ass flyer. Not that I'm buying --just saying.
True, but it also sounds like the kind of thing I'd make up to add to the description if I was faking it too.
Or he fell for a "confidential" photochop as part of the package. The wording suggests he actually saw it (or thinks he did).
Our old friend pcpop is claiming NV50 is 5% faster than R520 in 3DM05.
Issues the date: On April 15, 2005 author: Pineapple edition: Pineapple
Comes from the eNet news, recently the R520 news really was all over the sky flies, although in which has many is the vacation news, but from in the news which already confirmed we to the R520 nature prefers to peep a spot, 14,000 many 3Dmark 05 scores points with ease sufficiently does 6800SLI. Facing the such strong match, main item eldest child NVIDIA certainly cannot wait.
Now has the news to indicate about nVIDIA NV50, its performance must be stronger than R520. Has FutureMark Corporation to disclose that, NV50 under 3Dmark05 scoring points is 15,394, this way NV50 has surpassed the R520 probably 5% performance. Looked like new turn reveals the card to strive for hegemony the war soon to perform.
There's more to the article that claims NV50 SLI is a terrorist. Heh.
http://document.pcpop.com/0/84/84367.shtml
Come to think of it, this is a milestone event --"dualing fictitious benchmarks of both unreleased/unannounced cards compared". Check!
ATI - My card can do 14000 in 3dmark05.
nvidia - Oh yeah, my card can do 15000.
ATI - Did I say 14000? I meant 16000.
nvidia - Pfff, ours can do 17000.
Chalnoth
16-Apr-2005, 07:35
NV50 isn't the next part out from nVidia. That, at least, seems quite clear. The NV4x architecture is just too well-positioned to replace entirely before WGF.
Ailuros
16-Apr-2005, 10:08
Since NVIDIA won't most likely have a 90nm part prior to late 2005, if there's going to be a NV4x spring refresh, the >4 quads @ 110nm makes sense. More of an alternative measure to match the competition's estimated fill-rates.
If the above should turn out to be true, the question for me would then be whether the possible fall 90nm part will have the same amount of quads as the spring refresh or not. More like a theoretical question quite frankly, since I'm mostly interested what comes out in the end; how it's achieved is less relevant as long as the price/performance ratio is more or less on the same level.
The Spring Refresh is looking awful Summerish at this point. At what point do you start pushing next parts into 2006?
Even if Xbox360 will be a successful and powerful console with a wonderful launch the earth will still revolving around the sun 6 months later..
We're talking about GPUs here anyway so let's keep console debates off from this forum.
*sigh*
am talking about the graphics processor inside the aforementioned console.
your selective snipping reeks of propaganda style..
The only part that really seems to be well thought-out was the equipping of all products with turbocache. This will probably be the case, but it doesn't mean the high-end parts won't still have lots of onboard memory.
I think all NV40's allready have "turbocache"
Why? since most r300's support this as well, there's a registry key to enable hypermemory.
Although there is no real performance gain (not everyone has P3200 running at 2-2-2-5) the feature is available.
am talking about the graphics processor inside the aforementioned console
your selective snipping reeks of propaganda style..
When you write "one certain electronics giant will have a serious problem" is crystal clear you're not talking about GPUs, in fact in your reply there wasn't anything about some GPU.
I'm not here to spread any kind of propaganda, if you have some problem I'm sure moderators can help you.
The only part that really seems to be well thought-out was the equipping of all products with turbocache. This will probably be the case, but it doesn't mean the high-end parts won't still have lots of onboard memory.
I think all NV40's allready have "turbocache"
Why? since most r300's support this as well, there's a registry key to enable hypermemory.
Although there is no real performance gain (not everyone has P3200 running at 2-2-2-5) the feature is available.
I don't believe that is true to be honest, NV44 had to be re-architected to remove as much of the latency as possible that is present when accessing system memory.
Dave Baumann
16-Apr-2005, 23:20
Its not the latency thats the issue for the hardware, its the fact that they physically can't render surfaces to main memory. Now that its done in NV44 its likely that this will follow into all subsequent chips whether its used or not, especially since this is a step towards full virtualisation as required by WGF2.0.
Chalnoth
17-Apr-2005, 03:38
I think all NV40's allready have "turbocache"
Not very likely, given that most aren't PCI-Express native.
Wunderchu
17-Apr-2005, 05:01
http://www.vr-zone.com/?i=2010&s=1
Chalnoth
17-Apr-2005, 06:21
http://www.vr-zone.com/?i=2010&s=1
Posted two pages back.
Wunderchu
17-Apr-2005, 09:51
http://www.vr-zone.com/?i=2010&s=1
Posted two pages back.o. oops.. hehe :oops: :)
overclocked
18-Apr-2005, 11:17
Has anyone a clue how many transistors a ROP in the NV4x takes, a guess?
kyetech
19-Apr-2005, 13:53
http://www.the-inquirer.com/?article=22645
G70 - next gen part tapes out... so thats what 2.5 - 3 months away from release?
http://www.the-inquirer.com/?article=22645
G70 - next gen part tapes out... so thats what 2.5 - 3 months away from release?
If all goes well.
DegustatoR
19-Apr-2005, 15:26
http://www.the-inquirer.com/?article=22645
G70 - next gen part tapes out... so thats what 2.5 - 3 months away from release?
For one respin. 6 for two. I'd say it's starting to look more like Autumn'05 and 90nm if this Inq story is correct.
http://www.the-inquirer.com/?article=22645
G70 - next gen part tapes out... so thats what 2.5 - 3 months away from release?
They don't say when it taped out, or when first silicon came back (in fact, you get the impression that they still haven't grasped the concept of a tape out properly). Not much to go on at all.
If first silicon was back only recently and it really is an R520 rival, that would put ATi almost 5 months ahead. They've already completed a minor respin, which would suggest (assuming all went well) that they're in/close to the production qualification stage right now.
Dave Baumann
19-Apr-2005, 15:33
I think he's talking about "tape out of final production silicon", not "initial tape out after RTL completetion" (which I what I think most of us take tapeout to mean). I believe the code on the PCI SIG list is for G70 and it make sense to validate an engineering sample if it is not going to require any changes from the production silicon.
Well that would certainly make more sense.
Well, criminy, then where's that 110nm high-end part? Did it just go poof! again?
DegustatoR
19-Apr-2005, 15:45
Well, criminy, then where's that 110nm high-end part? Did it just go poof! again?
Well they don't say it's gonna be 90nm don't they... I assume that G70 is 110nm part.
Well, criminy, then where's that 110nm high-end part? Did it just go poof! again?
Well they don't say it's gonna be 90nm don't they... I assume that G70 is 110nm part.
Oh. Right. Back to the diet coke bin for enuf caffeine to support rational thot. :oops:
The assumptions in this part weren't a comfort:
We don't have much information about specifics but we strongly believe that the G70 will be released very close to the R520's shipping date.
And we're told that this should happen sometime in the autumn. . .
trinibwoy
19-Apr-2005, 15:55
Well maybe that's a good thing. We only got NV40 and R420 last summer - maybe we should give these companies some time to make back the R&D $$$ they spent developing these parts :) I still don't understand how they make any money with these aggressive schedules. And if we do get these parts this summer what are we going to do? Keep playing Doom3 at 60fps and get HL2 up into the 200fps range. IMO current games do not demand new hardware anyway but it does keep the boards alive.
Oh, and if this critter isn't on the streets until September, does anyone here really see an NV 90nm on the streets before, say, March 2006? Maybe closer to May? At which point it would pretty much have to be their WGF2.0-capable part, wouldn't it?
Well, criminy, then where's that 110nm high-end part? Did it just go poof! again?
I've heard these pieces of information in reference to a forthcoming 110mn part:
Native PCIE high-end NV4x.
10% more transistors than NV40.
Target clock ~550MHz.
May be pitched (in SLI form?) against R520 for a while until next-gen is ready.
From separate sources, at different times - all very confusing. :?
mufu that is what i would expect from a 110nm part , 120mhz speed increase should give it a nice boost and it will most likely be slower than the r520 it will be very close , so close that it wont really matter either way
Well, criminy, then where's that 110nm high-end part? Did it just go poof! again?
I've heard these pieces of information in reference to a forthcoming 110mn part:
Native PCIE high-end NV4x.
10% more transistors than NV40.
Target clock ~550MHz.
May be pitched (in SLI form?) against R520 for a while until next-gen is ready.
From separate sources, at different times - all very confusing. :?
And would it be your understanding that is "G70"?
That strain of rumors fits in with the wait-and-see what R520 is and maybe-it's-cancelled-or-maybe-we'll-release-it crowd. The idea being that if such a part looks competitve they go forward, if it still get's its ass kicked then lick their wounds and rush the 90nm part.
Still, their ability to pull the 90nm part forward must be very limited, and you'd have to think that stuffing another high-end release in there must push the 90nm part back.
I still keep looking at how often they say "about every two years" and think that Spring 2006 looks more likely for 90nm for NV even tho they have also talked about 90nm "in the fall" of 2005.
digitalwanderer
19-Apr-2005, 16:49
Actually I see this G70 news as a portent of ATi launching R520 soon.... ;)
And would it be your understanding that is "G70"?
Possibly. Looks pretty weak on paper vs. R520 though (unless they've squeezed in an extra quad). I'd be more inclined to believe this is some kind of interim part that might have already disappeared from the roadmap (NV48?). nV are yet to counter R480/R481 and perhaps they never will.
JoshMST
19-Apr-2005, 19:28
Could Faud maybe have printed something a bit more substantial than this? I am really actually quite confused about what is going to be released in the next few months, but I guess since NVIDIA is pushing the 512 MB Ultras at the end of this month, that it probably signals that the next gen is not going to be out in quite some time. So the more we hear, the more it looks like ATI will be out first with the R520, and that NVIDIA will trail for a couple of months (possibly late July release?).
Who knows, but with 8 to 12 weeks for fabrication, that comes out about right. That is... assuming that final production tapeout did just occur recently. And that is a big assumption. So really, who the hell knows? Well... outside of NVIDIA that is.
nV are yet to counter R480/R481 and perhaps they never will.
From the information that I've got, I've been told that 'they don't need to' refresh. They've got products in stock on both AGP and PCI-Express, and the refresh would probably only run at 450/1200MHz - there are already products from EVGA and Gainward that operate at those clock speeds. :)
trinibwoy
19-Apr-2005, 20:22
mufu that is what i would expect from a 110nm part , 120mhz speed increase should give it a nice boost and it will most likely be slower than the r520 it will be very close , so close that it wont really matter either way
You mean 150Mhz right? Don't Ultras run at 400 now?
The situation is really pretty ridiculous --somebody call up Jen-Hsun and get to the bottom of it.
I can understand why some think the 110nm high-end part to just be a big ruse.
The weird thing is, the tapeout of the R520 and R500 happened months, almost half a year ago.
Ang the G70, taped out only recently? could it be that NV will launch a product in may 2004 and refresh it in november 2005?
Those graphics chips need testing, a LOT of testing, you don't just plug it out and straight into production, not if you want to be in a world of hurt when some things don't add up right..
If the G70 tapeout is true then it will be smooth sailing for the R520 for a couple of months up to half a year...
Would nV allow that? to have one card best their high-end SLI solution? even if they did bump the speed marginally, an SLI'ed Ultra512 setup would be "on-par" with an R520, at half the power consumption AND half the price...
The weird thing is, the tapeout of the R520 and R500 happened months, almost half a year ago.
Ang the G70, taped out only recently? could it be that NV will launch a product in may 2004 and refresh it in november 2005?
Those graphics chips need testing, a LOT of testing, you don't just plug it out and straight into production, not if you want to be in a world of hurt when some things don't add up right..
If the G70 tapeout is true then it will be smooth sailing for the R520 for a couple of months up to half a year...
Would nV allow that? to have one card best their high-end SLI solution? even if they did bump the speed marginally, an SLI'ed Ultra512 setup would be "on-par" with an R520, at half the power consumption AND half the price...
See Wavey/MuFu a few notes upstream. Dave seems to be suggesting this might be final silicon pointing at another indicator someone stumbled on recently. Also that Inq needs a course on industry-standard terminology. :D
See Wavey/MuFu a few notes upstream. Dave seems to be suggesting this might be final silicon pointing at another indicator someone stumbled on recently. Also that Inq needs a course on industry-standard terminology. :D
Didn't want to check the last two pages, lazy me...
Anyway, I'd hate it if hellbinder was giving fuad this info and we'd be discussing it as " inq confidential" right here...
I'm a bit boggled by all this refresh "news"...
I don't get why a refresh is currently needed, games played on todays high end hardware (EQ 2 aside) run pretty well.
Apart from ATi needing to put the lid over "nvidia has PS3" load of crap I really can't see a performance jump needed. Both companies have amazingly fast high end products mostly still held back by todays CPUs.
But I'm sure I'm alone taking this stance...
trinibwoy
19-Apr-2005, 22:16
But I'm sure I'm alone taking this stance...
You're not.
I'm a bit boggled by all this refresh "news"...
I don't get why a refresh is currently needed, games played on todays high end hardware (EQ 2 aside) run pretty well.
Apart from ATi needing to put the lid over "nvidia has PS3" load of crap I really can't see a performance jump needed. Both companies have amazingly fast high end products mostly still held back by todays CPUs.
But I'm sure I'm alone taking this stance...
So.. you wanted to play Quake4 on your tnt2 now did you?
With the multi core processor upon us, the road to massive processing power is ahead.
It would be no sense to get the latest and greatest and feel one of your components was inferior, for these people, the social impared but in love with their GPU the market is small but the margins are high.
Along with these high margins come free advertisement.
And you know what free advertisement does, it sells the low and mid-range categories, and it especially sells the mid-range *WELL*
You've seen it with the 9500Pro, which was the love child of anyone who could flash a bios.
And now the 6600GT, not really cheap, but the advertisement of SLI helps sell it, and it sells well because it is faster than it's direct opponent, the x700pro.
This really is all about making money, we fool ourselves into believing we're getting something in return (good graphics, GREAT graphics) but we're just giving the CEO's at NV and ATi a nice $ coloured rimjob....
At the very littlest least, for all our sakes, a transition to 90nm ought to be in everyone's interest, even if all they did was a straight shrink of the existing architectures. Yields=pricing. (Well, that does leave out demand, but still.)
At the very littlest least, for all our sakes, a transition to 90nm ought to be in everyone's interest, even if all they did was a straight shrink of the existing architectures. Yields=pricing. (Well, that does leave out demand, but still.)
I've seen what a shrink does for pricing.. but really, the pricedrop of a 3500+ from newcastle to winchester didn't get me excited!
MistaPi
20-Apr-2005, 00:53
I've heard these pieces of information in reference to a forthcoming 110mn part:
Native PCIE high-end NV4x.
10% more transistors than NV40.
Target clock ~550MHz.
May be pitched (in SLI form?) against R520 for a while until next-gen is ready.
From separate sources, at different times - all very confusing. :?
What do you think the 10% more transistors will be used for? Is it enough for a extra quad of pipelines, perhaps without ROPs?
Chalnoth
20-Apr-2005, 02:29
A product like that? Efficiency improvements.
Hellbinder
20-Apr-2005, 04:05
And would it be your understanding that is "G70"?
Possibly. Looks pretty weak on paper vs. R520 though (unless they've squeezed in an extra quad). I'd be more inclined to believe this is some kind of interim part that might have already disappeared from the roadmap (NV48?). nV are yet to counter R480/R481 and perhaps they never will.
aaah.. Now thats someone who has actually seen some stuff. ;)
Pretty soon all this "Dont have such high expectations of the R520" talk are going to be put to rest.
R520 = MONSTER.
Chalnoth
20-Apr-2005, 04:11
:roll:
Vegetto
20-Apr-2005, 04:23
And would it be your understanding that is "G70"?
Possibly. Looks pretty weak on paper vs. R520 though (unless they've squeezed in an extra quad). I'd be more inclined to believe this is some kind of interim part that might have already disappeared from the roadmap (NV48?). nV are yet to counter R480/R481 and perhaps they never will.
aaah.. Now thats someone who has actually seen some stuff. ;)
Pretty soon all this "Dont have such high expectations of the R520" talk are going to be put to rest.
R520 = MONSTER. It's always the same isn't it? :roll:
So.. you wanted to play Quake4 on your tnt2 now did you?
This really is all about making money, we fool ourselves into believing we're getting something in return (good graphics, GREAT graphics) but we're just giving the CEO's at NV and ATi a nice $ coloured rimjob....
Well....I'm awaiting a 6800 GT right now...since my 9700 pro burned out.
Suffice to say I was still very happy with it, and it supplied me with an immense gaming experience.
I do believe er are getting something in return but for now it seems that even a solid midrange card can provide an amazing gaming experience, so I'm a bit pensive about the need to make a bigger better card today.
It's not that I don't applaud the companise efforts, but I think we might be stupid consumers.... :roll:
Unknown Soldier
20-Apr-2005, 07:29
Why stupid. If you have the money and want the latest fastest h/w then you are entitled to buy it. If you don't have the money ... well then stay with the older cards and be happy.
As it was said earlier .. the main graphics cards are almost always used to sell the mainstream cards .. the GT6800, GT6600, X800XL, X800Pro, X800. Unfortuantely .. it's starting to look like even these cards can sometimes be out of the mainstream price range for some people and cheaper alternatives are always welcome. As it is .. the GT6600 and X800 are the cheap alternatives atm .. and they do a decent job.
US
dizietsma
20-Apr-2005, 07:32
How does assumed change of manufacturing process/ efficency tweak / speed tweak equate to a whole new G7x numbering scheme ?
At present the only thoughts I have seen is that it may be due to DTR being melded into the desktop range, is it just this or is more?
Also, on average how soon do new card codes appear in the .inf file before the card makes an appearance ?
G70 can still stand for GeForce 7 series (G70 in short). It just looks fancy :)
Ailuros
20-Apr-2005, 09:09
Well, criminy, then where's that 110nm high-end part? Did it just go poof! again?
I've heard these pieces of information in reference to a forthcoming 110mn part:
Native PCIE high-end NV4x.
10% more transistors than NV40.
Target clock ~550MHz.
May be pitched (in SLI form?) against R520 for a while until next-gen is ready.
From separate sources, at different times - all very confusing. :?
Close to or equal to 550MHz sounds a lot; while not impossible I still wonder what the power consumption of such a beast may be; NV43's power consumption isn't relatively low either.
DegustatoR
20-Apr-2005, 10:29
NV43's power consumption isn't relatively low either.
Relatively to what? NV43's power consumption is lower than RV410's.
Well, criminy, then where's that 110nm high-end part? Did it just go poof! again?
I've heard these pieces of information in reference to a forthcoming 110mn part:
Native PCIE high-end NV4x.
10% more transistors than NV40.
Target clock ~550MHz.
May be pitched (in SLI form?) against R520 for a while until next-gen is ready.
From separate sources, at different times - all very confusing. :?
Close to or equal to 550MHz sounds a lot.
Yes. Especially when you consider that they've pushed the 12 pipe, 110nm NV42 up to only 450MHz (albeit with modest cooling and perhaps a relatively low Vcore). Incidentally, has anybody had a look in the latest Forceware betas to see if NV48 is still there? It'd be nice to get it out of the equation.
If we ignore G80 for now (assuming it's their WGF 2.0 offering), a 24 pipe G70 @550MHz might square up quite nicely against a 16 pipe R520 at 650-700MHz (especially if it doesn't have a 1:1, pipe:ROP mapping). You would think it'd be 90nm, that being the case. The turnaround would have to have been relatively quick, but they already have experience migrating NV4x across fab nodes.
Going round in circles here a bit, sorry. Mr Occam says it'll just be a battle of two very fast, 250 million transistor, 16 pipeline, 90nm parts. :cool:
If we ignore G80 for now (assuming it's their WGF 2.0 offering), a 24 pipe G70 @550MHz might square up quite nicely against a 16 pipe R520 at 650-700MHz (especially if it doesn't have a 1:1, pipe:ROP mapping). You would think it'd be 90nm, that being the case. The turnaround would have to have been relatively quick, but they already have experience migrating NV4x across fab nodes.
But NVidia's never done low-k before...
Jawed
But NVidia's never done low-k before...
They have, they just f**ked it up and got scared (allegedly). :lol:
Demirug
20-Apr-2005, 13:44
Incidentally, has anybody had a look in the latest Forceware betas to see if NV48 is still there? It'd be nice to get it out of the equation.
Yes, it is. And the chips are named as GF 6800 (LE, GT).
Incidentally, has anybody had a look in the latest Forceware betas to see if NV48 is still there? It'd be nice to get it out of the equation.
Yes, it is. And the chips are named as GF 6800 (LE, GT).
Ah yes, I remember. Well that's reassuringly boring, thanks.
dizietsma
20-Apr-2005, 14:00
What is the reasoning behind as yet to be launched gpu's to be mentioned in inf drivers ? Is it so that OEM's can decide on which driver to release on the CD when the card is released ?
What is the reasoning behind as yet to be launched gpu's to be mentioned in inf drivers ? Is it so that OEM's can decide on which driver to release on the CD when the card is released ?
I was always under the impression it is because they are actually prepping the drivers to support that chip in advance of release, particularly in a unified driver environment. Otherwise you end up with different development tracks and nasty versioning control issues are liable to crop up.
m'gawd, I think I just actually used some of my ancient developer background on b3d appropriately and in context. Will wonders never cease.
What is the reasoning behind as yet to be launched gpu's to be mentioned in inf drivers ? Is it so that OEM's can decide on which driver to release on the CD when the card is released ?
Well, it would be nice for driver support, right?
Why stupid. If you have the money and want the latest fastest h/w then you are entitled to buy it. If you don't have the money ... well then stay with the older cards and be happy.
US
I think we might be a bit stupid.
I'll elaborate, let's say you're playing game X-the latest greatest piece of software on the market.
You're playing at 1280*1024 4aa+8af at a solid 60+ most of the time and rare dips into the thirties.
Imagine I have in my hand fantastic card Y-all new and improved that supports the same technology your card has but you can play at 1600*1200 4aa+8af-would you spend 500$ on this card when not 6 months ago you bought another one for 400$?
I think many paople here might say yes-but I doubt the difference in IQ is one that justifies the price. So when a refresh is to be launched with a 10% gameplay improvement, I feel we're stupid consumers to get all hyped up about it.
There will most probably always be a better thing around the corner-but the rate of improvement has to be one that justifies our enthusiaism.
This is in a no way criticism towards ATi's or Nvidia's engineers-they need to sell a product.
I'll elaborate, let's say you're playing game X-the latest greatest piece of software on the market.
You're playing at 1280*1024 4aa+8af at a solid 60+ most of the time and rare dips into the thirties.
Imagine I have in my hand fantastic card Y-all new and improved that supports the same technology your card has but you can play at 1600*1200 4aa+8af-would you spend 500$ on this card when not 6 months ago you bought another one for 400$?
I think many paople here might say yes-but I doubt the difference in IQ is one that justifies the price. So when a refresh is to be launched with a 10% gameplay improvement, I feel we're stupid consumers to get all hyped up about it.
Who's that "we" you mentioned? ;)
I'd say someone who does this is just rich or doesn't care about money, not stupid.
trinibwoy
20-Apr-2005, 18:58
I think you make a good argument and it really isn't the IHVs to blame either. Hardware is simply improving at a much faster rate than games are so all we can use to justify our shiny new cards is a slight resolution or AA increase. And even with all the hype surrounding IQ, that resolution bump is not worth the $$$. Titles like Doom3 and HL2 make matters even worse since the top-end cards are not even required to experience all the games have to offer. (Of course, this generalization excludes all those people with superhuman perception that can spot a jaggy a mile away :))
I'd say someone who does this is just rich or doesn't care about money, not stupid.
I'm not claiming dumb people just "unsmart" consumers, we don't drive them to bring prices down by agreeing to pay rising premiums for hardly noticeable improvements.
Heck why not...maybe I'm just stupid for paying 400$ for cards.... :oops:
overclocked
20-Apr-2005, 20:21
i asked how many trnsistors app a ROP in the NV4x takes but got no answer.
My thought was that they shrink the chip on a new node plus dont add any ROP´s.
How many transistors would nVidia save by having a 16p/8ROP chip instead as a "new 6800NU" on a 110nm node(not that the node does anything for the count).
I Wonder because could they remove say 8 ROP´s and add a quad at equal transistorcount?
Ailuros
21-Apr-2005, 02:08
NV43's power consumption isn't relatively low either.
Relatively to what? NV43's power consumption is lower than RV410's.
All other >2 quad 110nm chips this far. It consumes also more than a 6800nonUltra.
If we ignore G80 for now (assuming it's their WGF 2.0 offering), a 24 pipe G70 @550MHz might square up quite nicely against a 16 pipe R520 at 650-700MHz (especially if it doesn't have a 1:1, pipe:ROP mapping).
Was the last parenthesis for R520?
In any case 550/700 (for the according chips) might be or have been the targeted clockspeeds, yet they still sound at least to me a bit too optimistic.
Besides that, anyone dare to speculate how many quads NV's 90nm fall part might have?
Ailuros
21-Apr-2005, 02:12
i asked how many trnsistors app a ROP in the NV4x takes but got no answer.
My thought was that they shrink the chip on a new node plus dont add any ROP´s.
How many transistors would nVidia save by having a 16p/8ROP chip instead as a "new 6800NU" on a 110nm node(not that the node does anything for the count).
I Wonder because could they remove say 8 ROP´s and add a quad at equal transistorcount?
No idea about transistor count estimates for ROPs. As far as ROP amount concerns I don't think we currently need more than 16 ROPs and it may very well be that NVIDIA's next high end part might have 6 quads with say 16 ROPs.
If we ignore G80 for now (assuming it's their WGF 2.0 offering), a 24 pipe G70 @550MHz might square up quite nicely against a 16 pipe R520 at 650-700MHz (especially if it doesn't have a 1:1, pipe:ROP mapping).
Was the last parenthesis for R520?
No, G70, although NV43 would suggest that for current games, not having a 1:1 mapping is barely a hindrance.
Ailuros
22-Apr-2005, 00:59
True; yet I doubt the ROP amount on NV43 could be indicative, since it's a mainstream part.
"G70 to be Power Hungry"
http://www.theinquirer.net/?article=22917
I couldn't get into the site they linked (requires userid/pwrd). Anyone else able to see that material?
OUch..225 Watts! Dual something? ;)
Do they actually know the second connector will need to supply the full 75W, or are they just guessing? It seems more likely that the total power consumption simply tips over the PCI-E + Molex total (thus necessitating the second connector), rather than pushing the limits of PCI-E + 2x Molex :?
"G70 to be Power Hungry"
http://www.theinquirer.net/?article=22917
I couldn't get into the site they linked (requires userid/pwrd). Anyone else able to see that material?
..and with SLI you need 550watts just for the graphiccards?
Not surprising adding more trans. and having a higher clockspeed with 110nm which isn't lowk. Even r520 is expected to need more power than a x800 even though they using 90nm lowk
I guess this is some kind of safety measure. If they clock it low, they might keep it below 150W, but depending on what R520 brings to the table, they might want to cross that border.
225 Watts, what kind of Fan will they use to cool that? and keep it comfortable in SLI working environments..
Dave Baumann
29-Apr-2005, 17:37
That 150W power specification Fudo is linking to has been around for ages - we reported on this back in December (http://www.beyond3d.com/forum/viewtopic.php?t=19088&highlight=pci+express). From what I know of the specification it will not only call for a 150W power supply but will also effectively be a defacto dual slot solution.
That 150W power specification Fudo is linking to has been around for ages - we reported on this back in December (http://www.beyond3d.com/forum/viewtopic.php?t=19088&highlight=pci+express). From what I know of the specification it will not only call for a 150W power supply but will also effectively be a defacto dual slot solution.
So false alarm re G70, or just a poor choice of supporting links from Fudo?
So false alarm re G70, or just a poor choice of supporting links from Fudo?
I think as in a lot of cases (as with the D955 motherboard) Fuad has his facts right, just some bad wording.)
btw, D955 = AMR board and not SLI Compatible... nice move from intel...
Yep, the Dell MVP system can't be far away. Imagine how many R520s that will leave for the rest of us :!: :cry:
Jawed
Yep, the Dell MVP system can't be far away. Imagine how many R520s that will leave for the rest of us :!: :cry:
Jawed
8p
I'm just wondering why intel ripped of the extra peg16 slot for the "big" hardware sites (i.e. the ones with r520 demo models) and leaves it on for the smaller, non-english hardware sites...
Sorry, thread derailing..
digitalwanderer
29-Apr-2005, 18:43
Yep, the Dell MVP system can't be far away. Imagine how many R520s that will leave for the rest of us :!: :cry:
As long as there is one for me I'll be happy. :twisted:
martrox
29-Apr-2005, 23:39
I may need two........ :roll:
225 Watts, what kind of Fan will they use to cool that? and keep it comfortable in SLI working environments..
VR-Zone said this a couple of weeks ago:
The reference cooler design for NV50/G70 looks huge with 4 copper heat-pipes and a large fan.
G70 has to have a huge performance lead over R520 for me to even consider bringing that monster home. :)
Nah, who am i kidding? Of course i'll go with the fastest one, even if it means i'll burn my house down.
Why can't it be pre-launch times all year long? Gotta looove all the speculations etc. 8)
VR-Zone said this a couple of weeks ago:
The reference cooler design for NV50/G70 looks huge with 4 copper heat-pipes and a large fan.
G70 has to have a huge performance lead over R520 for me to even consider bringing that monster home. :)
Nah, who am i kidding? Of course i'll go with the fastest one, even if it means i'll burn my house down.
Why can't it be pre-launch times all year long? Gotta looove all the speculations etc. 8)
Yeah, I love the discussion, the feeling of "something BIG" waiting outside the door.. the endless speculation...
Whoever is providing L'inq with their information (it's probably someone we all know from posting) mentioned that the G70 would be faster, but not breathtaking, as in, if the r520 did score 14k in 3dm05, then the g70 would have 15k
For me it's okay though, since I spend enough bucks on my x800xt-pe, I can sing out this generation and make a big shift (still agp) to the card that plays wgf2.0 games best in 2006... :D
And for some reason, I do believe that that is ati.
Why? a lot of developers create their game for xboxes and it looks like a lot of games are thus mildly optimized for nV cards.
The future generations card (which should share some similarities with the next-gen consoles) thus benefit from the console foundations developers like to base their games on.
So, within a year or so, we should see a decline of the games that were made with the xbox in mind and a shift to the games that were developed on x800Pro's (todays xbox360 dev. kit.)
So, the G70 is 110nm an produced at IBM?
nutball
30-Apr-2005, 20:43
Whoever is providing L'inq with their information (it's probably someone we all know from posting) mentioned that the G70 would be faster, but not breathtaking, as in, if the r520 did score 14k in 3dm05, then the g70 would have 15k
That's not what Charlie reckons :)
http://www.aceshardware.com/forums/read_post.jsp?id=331&forumid=3
But when an Inq guy admits he has no solid facts... ;)
I think its going to be just like the r420vs the nv40 . THey will trade blows but be close
trinibwoy
30-Apr-2005, 21:14
Why? a lot of developers create their game for xboxes and it looks like a lot of games are thus mildly optimized for nV cards.
The future generations card (which should share some similarities with the next-gen consoles) thus benefit from the console foundations developers like to base their games on.
So, within a year or so, we should see a decline of the games that were made with the xbox in mind and a shift to the games that were developed on x800Pro's (todays xbox360 dev. kit.)
The only games I know of that gave Nvidia an edge due to their console roots are Halo and Riddick - not exactly 'a lot'. IMO the main differentiators this generation are Nvidia's OpenGL/stencil advantage and ATI's shader/clockspeed advantage. Console titles had very little impact in the wide scheme of things.
Also the Xbox was pretty much a PC in a box so ports were relatively straightforward. What makes you think that more games will be ported from the Xbox360 than the PS3? Both architectures are pretty different from the PC. Anyone know if games on the Xbox360 will be any more PC friendly than those on the PS3? What about Microsoft's XNA initiative?
karlotta
30-Apr-2005, 23:09
why..1.x PS, to 3.0 PS. the Xbox doesnt use 2.x PS so games are being made 1.x and or 3.0 if a port...... splintercell is one. Its a IHV(nvda) working over DEV/PUB. like UBIsoft. And all the big guys (EA UBI, ect) are making console games first, PC as a after thought/port
The only games I know of that gave Nvidia an edge due to their console roots are Halo and Riddick - not exactly 'a lot'. IMO the main differentiators this generation are Nvidia's OpenGL/stencil advantage and ATI's shader/clockspeed advantage. Console titles had very little impact in the wide scheme of things.
Also the Xbox was pretty much a PC in a box so ports were relatively straightforward. What makes you think that more games will be ported from the Xbox360 than the PS3? Both architectures are pretty different from the PC. Anyone know if games on the Xbox360 will be any more PC friendly than those on the PS3? What about Microsoft's XNA initiative?
As a PowerPC (the xbox was a pc, the xbox360 is a mac) there shouldn't be too much differences in hardware.
As "cell" is something new, it also makes the hurdle of programming a bit higher.
There are more games that came from the xbox to the pc, that aren't used for benchmarking, but which do have impact.
Ati got a "bad" reputation in the "kotor" scene because of a limiting bug in kotor2, turned out, there was a vertex buffering routine installed which ran fine on nv hardware, but brought x800's to a crashing standstill in certain places.
an "undocumented" line in the .ini file fixed the problem.
I'm just wondering why that never showed up in bug testing.
We'll probably see something like that as well when Fable jumps to PC.
trinibwoy
01-May-2005, 02:06
why..1.x PS, to 3.0 PS. the Xbox doesnt use 2.x PS so games are being made 1.x and or 3.0 if a port...... splintercell is one. Its a IHV(nvda) working over DEV/PUB. like UBIsoft. And all the big guys (EA UBI, ect) are making console games first, PC as a after thought/port
The Xbox doesn't support SM3.0 either so I'm not sure how SC:CT's console roots has anything to do with it (in the context of neliz's earlier statement to which I was replying). Nvidia's influence in that decision is a completely different matter.
trinibwoy
01-May-2005, 02:09
As a PowerPC (the xbox was a pc, the xbox360 is a mac) there shouldn't be too much differences in hardware.
As "cell" is something new, it also makes the hurdle of programming a bit higher.
There are more games that came from the xbox to the pc, that aren't used for benchmarking, but which do have impact.
Isn't the xbox360 going to be multi-core as well? I don't really know much about the inner working of the PowerPC architecture or how much more powerful these console CPU's will end up being but you may be right. What other Xbox-->PC ports are there besides Riddick, SC and Halo?
Demirug
01-May-2005, 08:28
why..1.x PS, to 3.0 PS. the Xbox doesnt use 2.x PS so games are being made 1.x and or 3.0 if a port...... splintercell is one. Its a IHV(nvda) working over DEV/PUB. like UBIsoft. And all the big guys (EA UBI, ect) are making console games first, PC as a after thought/port
The Xbox doesn't support SM3.0 either so I'm not sure how SC:CT's console roots has anything to do with it (in the context of neliz's earlier statement to which I was replying). Nvidia's influence in that decision is a completely different matter.
Don't think about the roots. Think about the future. I am sure UBI will use the SCX Engine for the next SC, too.
Isn't the xbox360 going to be multi-core as well? I don't really know much about the inner working of the PowerPC architecture or how much more powerful these console CPU's will end up being but you may be right. What other Xbox-->PC ports are there besides Riddick, SC and Halo?
Kotor 1&2, Fable (release this year) are the other "big hits" that were xbox exclusive and went to pc.
And a lot of studios would be mad if they didn't share resources between compatible platforms like the pc/xbox.
Tim Murray
01-May-2005, 20:19
Psychonauts
Area 51
Brothers in Arms
Cold Fear
GTA: San Andreas (okay, that's a PS2 to Xbox and PC port, but whatever)
that's just from current games, too. there are plenty Xbox->PC ports, or at least games that were released on Xbox and PC at around the same time.
trinibwoy
01-May-2005, 21:10
Oh, thanks. Didn't know the list was that long.
Dave Baumann
01-May-2005, 22:06
The original Splinter Cell is a very obvious example of this - it was initially built for the XBox and the shadowing mechnaism uses the DST implementation specific to NVIDIA; although the shadowing method doesn't use a defined method of filtering as per PC DX it obviously does work on NVIDIA's chips. This is why with NVIDIA boards the defauly mode is "buffer" shadows, but a second, lower detail "projector" method was implemented to make the title work on the PC on non-NVIDIA boards.
trinibwoy
01-May-2005, 23:18
The original Splinter Cell is a very obvious example of this - it was initially built for the XBox and the shadowing mechnaism uses the DST implementation specific to NVIDIA; although the shadowing method doesn't use a defined method of filtering as per PC DX it obviously does work on NVIDIA's chips. This is why with NVIDIA boards the defauly mode is "buffer" shadows, but a second, lower detail "projector" method was implemented to make the title work on the PC on non-NVIDIA boards.
Yeah I remember that. Does ATI have any similar 'features' that may be shared between R500 and R520 but may not be compatible with Nvidia cards? ATI doesn't seem to be as willing to provide 'extras' above and beyond the official spec.
Dave Baumann
01-May-2005, 23:33
Yeah I remember that. Does ATI have any similar 'features' that may be shared between R500 and R520 but may not be compatible with Nvidia cards?
Neither are announced or have been openly discussed - even then, you rarely find found out everything about each chips capabilities in finite detail.
ATI doesn't seem to be as willing to provide 'extras' above and beyond the official spec.
Chips have capabilities built into them that are often a superset of the "official spec" as far as fitting it into PC DX graphics, this is the same with all IHV's - 9700 had more temps than called for by the PS2.0 specification, it also had centriod sampling support (initially fixed with PS3.0) and capabilities for geometry instancing support, none of which were part of the DX9 SM2.0 spec. On the otherhand, when you're tlaking about a closed box console environment whats an "official spec"?
Kanyamagufa
02-May-2005, 03:25
As a PowerPC (the xbox was a pc, the xbox360 is a mac) there shouldn't be too much differences in hardware.
Being an avid Mac user since my beginning with computers, this is quite a remarkable statement. It's very exciting.
I also heard rumors that XBox360 Developer Kits were sent out to Mac users who had 6800 Ultras installed because of their PowerPC cores - It was the closest machine to the new XBox and therefore the best to start on.
Exciting news.
As a PowerPC (the xbox was a pc, the xbox360 is a mac) there shouldn't be too much differences in hardware.
Being an avid Mac user since my beginning with computers, this is quite a remarkable statement. It's very exciting.
I also heard rumors that XBox360 Developer Kits were sent out to Mac users who had 6800 Ultras installed because of their PowerPC cores - It was the closest machine to the new XBox and therefore the best to start on.
Exciting news.
:lol: I'm not sure where you heard these "things". The xbox360 is as close to a mac as the xbox was to the mac. Its running a windows kernal which automatically makes it not a mac. XDKs were sent to developers, not "mac users". And I have NO idea what you are trying to say about the graphic card? 6800 Ultras have IBM power processors on board? what? The 6800 is from nVidia, the xbox360's graphics are done by ATI, the XDK didn't have a 6800 in it.
AlStrong
02-May-2005, 05:34
Ati got a "bad" reputation in the "kotor" scene because of a limiting bug in kotor2, turned out, there was a vertex buffering routine installed which ran fine on nv hardware, but brought x800's to a crashing standstill in certain places.
an "undocumented" line in the .ini file fixed the problem.
I'm just wondering why that never showed up in bug testing.
We'll probably see something like that as well when Fable jumps to PC.
Er... sorry, but I don't know the details of this... what is the fix? What does it do? And do you know what the problem is specifically? Just curious. :)
Kanyamagufa
02-May-2005, 13:30
As a PowerPC (the xbox was a pc, the xbox360 is a mac) there shouldn't be too much differences in hardware.
Being an avid Mac user since my beginning with computers, this is quite a remarkable statement. It's very exciting.
I also heard rumors that XBox360 Developer Kits were sent out to Mac users who had 6800 Ultras installed because of their PowerPC cores - It was the closest machine to the new XBox and therefore the best to start on.
Exciting news.
:lol: I'm not sure where you heard these "things". The xbox360 is as close to a mac as the xbox was to the mac. Its running a windows kernal which automatically makes it not a mac. XDKs were sent to developers, not "mac users". And I have NO idea what you are trying to say about the graphic card? 6800 Ultras have IBM power processors on board? what? The 6800 is from nVidia, the xbox360's graphics are done by ATI, the XDK didn't have a 6800 in it.
The 6800 was not significant, it is merely the only card that was available (that has now changed) for Apple that could possibly provide a level of image quality near enough to develop for next generation consoles.
Also, I was not trying to say that developer kits were sent out only to Mac owners. I was simply saying that if the rumors were true, this marked the first time kits had ever been sent out to them, and that this was probably due to the presence of a PowerPC in both.
CyFactor
02-May-2005, 14:06
The 6800 was not significant, it is merely the only card that was available (that has now changed) for Apple that could possibly provide a level of image quality near enough to develop for next generation consoles.
Also, I was not trying to say that developer kits were sent out only to Mac owners. I was simply saying that if the rumors were true, this marked the first time kits had ever been sent out to them, and that this was probably due to the presence of a PowerPC in both.
The first XBox360 developer kits had ATI X800 class graphics boards. Not sure where you're getting your information from. They did not have nVidia graphics.
Er... sorry, but I don't know the details of this... what is the fix? What does it do? And do you know what the problem is specifically? Just curious. :)
It had something to do with Vertex buffering, disabling it in the ini file solves it on ati cards..
dizietsma
04-May-2005, 07:46
For nvidia would not the cheapest option to have G70/80 be two nv40 chips on the same die using the smaller but cheaper TSMC 110nm process, as hinted about above ? This would explain the large heatsink and also the requirement for two molex connectors as well for it.
This solution presumably would require little developement cost compared to a new chip, the money, time and physical resources saved thus going to the PS3 chip and the offshoot of that for the nv50 just in time for Longhorn.
This seems a very low risk solution, the SLi drivers are maturing, Gigabyte have shown that two chips on one card can be done ( though only 6600's the Asus 6800 card seemingly just an ES ) and it does give the possibility of 2 x 2 of these G70/80 cards being possible in a system.
This might also explain why nv47/nv48 have not appeared ( or have they ?? ) and a new naming convention is being done. Of course 40+40 is also 80, however G70 cannot be 2 nv35's of course. Maybe some light version though ? 2 6600GT's on a core ?
I have no evidence whatsoever for any of this, almost certainly will be wrong, but I was just letting my mind wander a bit.
Chalnoth
04-May-2005, 08:56
For nvidia would not the cheapest option to have G70/80 be two nv40 chips on the same die using the smaller but cheaper TSMC 110nm process, as hinted about above ? This would explain the large heatsink and also the requirement for two molex connectors as well for it.
It depends upon yields and the cost of linking two chips on the same packaging (both in dollars and in performance).
This solution presumably would require little developement cost compared to a new chip, the money, time and physical resources saved thus going to the PS3 chip and the offshoot of that for the nv50 just in time for Longhorn.
Except then you need to do more R&D on how to get the chips to communicate in order to keep performance efficient, and I don't know how involved nVidia would be in designing the packaging.
This seems a very low risk solution, the SLi drivers are maturing, Gigabyte have shown that two chips on one card can be done ( though only 6600's the Asus 6800 card seemingly just an ES ) and it does give the possibility of 2 x 2 of these G70/80 cards being possible in a system.
SLI would be far from optimal for a two-chip, one package solution, particularly since you couldn't very well double the memory bus without dramatically increasing cost. So, you actually need it to be efficient with memory bandwidth.
Ailuros
04-May-2005, 11:03
For nvidia would not the cheapest option to have G70/80 be two nv40 chips on the same die using the smaller but cheaper TSMC 110nm process, as hinted about above ? This would explain the large heatsink and also the requirement for two molex connectors as well for it.
In a highly relative sense one could consider a single quad as a "chip" already; if there really will be a 6 quad GPU from NVIDIA, you could eventually view it as a NV40+NV43 in that rather weird sense.
I have no doubt that such a board at 110nm and a =/>430MHz frequency would draw a lot of power no idea if it would exceed the 150W barrier though. How impossible is it for NVIDIA to apply for PCI-Sig compliance for that second molex just in order to cover better some vendors own incentives (multi-chip boards)?
Megadrive1988
04-May-2005, 13:15
225 Watts, what kind of Fan will they use to cool that? and keep it comfortable in SLI working environments..
VR-Zone said this a couple of weeks ago:
The reference cooler design for NV50/G70 looks huge with 4 copper heat-pipes and a large fan.
G70 has to have a huge performance lead over R520 for me to even consider bringing that monster home. :)
Nah, who am i kidding? Of course i'll go with the fastest one, even if it means i'll burn my house down.
Why can't it be pre-launch times all year long? Gotta looove all the speculations etc. 8)
well VR-Zone is most likely wrong, because G70 is not NV50 or NV5X. the G70 is probably an NV4X architecture
and G80? *maybe* NV50 / NV5X, or maybe not.
In a highly relative sense one could consider a single quad as a "chip" already; if there really will be a 6 quad GPU from NVIDIA, you could eventually view it as a NV40+NV43 in that rather weird sense.
NV41+NV41 or NV42+NV42 makes more sense I think since those are native 3 quads.
Ailuros
04-May-2005, 23:53
In a highly relative sense one could consider a single quad as a "chip" already; if there really will be a 6 quad GPU from NVIDIA, you could eventually view it as a NV40+NV43 in that rather weird sense.
NV41+NV41 or NV42+NV42 makes more sense I think since those are native 3 quads.
I said on purpose in a highly relative sense; my weird example gives 20 hypothetical ROPs and your two examples 24. I can't know how many ROPs G70 will have, but I can't figure either why more than 16 would be absolutely necessary at this stage either.
Won't the G70 have a big speed disadvantage compared to the r520 if it's on 110nm? 110nm doesn't add much in performance it's more of a die shrink. Is there a chance that the G70 could end up being 90nm instead?
Ailuros
05-May-2005, 02:46
Judging from sterile numbers alone (which are merely indications) not really.
Purely hypothetical wouldn't 6 quads @ 400MHz give the same fill-rate as 4 quads @ 600MHz? Presupposition that the estimated amount of quads is correct and quad=quad between the two different architectures.
DudeMiester
05-May-2005, 03:35
Just my 2 cents:
I think they will have 8 quads, however each quad will be simpler in design then in the NV40. The reason being is that with dynamic branching, the more pipes you have the less costly a stall will be, as the less of the overall chip is stalled. The extra slowness of the quads will be compensated by increased clockspeed thanks to the simpler design and process tech.
Chalnoth
05-May-2005, 03:44
I doubt they'll make that much architectural difference until WGF.
Megadrive1988
05-May-2005, 06:17
I doubt they'll make that much architectural difference until WGF.
agreed. therefore G70, and probably G80, are refreshes of the current Nv4X architecture.
G70 is to counter R520 Fudo
G80 is to counter R580 ?? probably.
(and others have said that R580 is more than just a speed bumped R520)
Ailuros
05-May-2005, 06:28
G80 will be most likely their first 90nm GPU. I wonder if they'll keep the hypothetical 6 quads of G70 or not.
overclocked
06-May-2005, 18:58
When are they launching the cards btw. 8)
trinibwoy
06-May-2005, 19:06
I can't believe we know so little about NV's next gen part. Soon they will have the R520+AMR upon them. For their sake I hope they haven't been twiddling their thumbs and basking in "But ATI don't have SM3.0!!" cause that ship has sailed baby :)
I can't believe we know so little about NV's next gen part. Soon they will have the R520+AMR upon them. For their sake I hope they haven't been twiddling their thumbs and basking in "But ATI don't have SM3.0!!" cause that ship has sailed baby :)
It's a bit freaky. As I've said before, I can understand why some people are convinced the existance of such a card is just a head fake and not the real thing. I think it's too elaborate for that, but I would guess we will get a little more visibility next week at the conference call.
wireframe
06-May-2005, 20:12
I can't believe we know so little about NV's next gen part. Soon they will have the R520+AMR upon them. For their sake I hope they haven't been twiddling their thumbs and basking in "But ATI don't have SM3.0!!" cause that ship has sailed baby :)
That's certainly one way of looking at it and exploring the possible outcomes. Another would be that it is ATI that was caught basking, not for one moment thinking Nvidia would launch the 6000 series in the form that it took. ATI may have been thinking that NV40 (or whatever) was going to put enough stress on them to just catch up to the SM 2.0 pace they had set and then Nvidia releases something that absolutely trounces their X800 offering. Maybe ATI thought they would be first with the SM 3.0 specs in R520 and maybe they have now had to go back and rethink the R520 so as not to be caught with their pants around their ankles once more.
I don't think I need to comment on the "AMR" aspect of your suggestion as I believe we are fully aware that Nvidia already has that one covered with SLI. It is ATI that is playing catch up.
trinibwoy
06-May-2005, 20:21
I don't think I need to comment on the "AMR" aspect of your suggestion as I believe we are fully aware that Nvidia already has that one covered with SLI. It is ATI that is playing catch up.
Exactly. Now that ATI has caught up to Nvidia with SM3.0 and a multi-GPU solution I was wondering what Nvidia is doing to make their next generation parts more attractive than the competition. I guess you could look at it as - while ATI was playing catch up - Nvidia was doing ?????.
I guess you could look at it as - while ATI was playing catch up - Nvidia was doing ?????.
What ATI did with R360? :lol: Actually, not even that much (yet).
Actually, I guess at this point you'd have to consider SLI as their "refresh" strategy, given we are one year on from NV40.
DegustatoR
06-May-2005, 23:43
Actually, I guess at this point you'd have to consider SLI as their "refresh" strategy, given we are one year on from NV40.
SLI is not an NVIDIA refresh.
caboosemoose
07-May-2005, 00:39
Whoever is providing L'inq with their information (it's probably someone we all know from posting) mentioned that the G70 would be faster, but not breathtaking, as in, if the r520 did score 14k in 3dm05, then the g70 would have 15k
That's not what Charlie reckons :)
http://www.aceshardware.com/forums/read_post.jsp?id=331&forumid=3
But when an Inq guy admits he has no solid facts... ;)
Well, for starters, many are not expecting R520 to be twice as fast as R420, as Charlies suggests it will be in that post, except in very specific instances (SM3 stuff only). So, that makes the rest of his post pretty hard to swallow.
Having said that, I very much go along with the theory that G70 will be:
NV4X-based
6 Fragment Quads
Less than 24 ROPs
Dave Baumann
11-May-2005, 22:13
6 Fragment Quads
IMO, yup.
6 Fragment Quads
IMO, yup.
They really pushing the limits don't they. Upping the VS to 8?
on 110nm
What's the target for the clockspeed?
6 Fragment Quads
IMO, yup.
So we've settled that G70 really is the next high-end part from NV, at 110nm? Glad to hear it. I've been getting tired of the circumlocutions I've been using to avoid that label. Labels are handy.
Any visibility on wether there is a significant difference between the old NV50 and G70 other than the naming convention change? Or is this actually the NV48 replacement?
When the dust settles a bit, it'd be nice to get an interview with someone at NV to walk thru all that "from there to here". Some loose ends flapping about.
Dave Baumann
11-May-2005, 22:44
I've heard a suggestion of a little north of 450MHz, but don't know if thats the GT or Ultra version (I suspect GT will be pushed out first and the Ultra waiting to see what ATI has).
trinibwoy
11-May-2005, 23:26
Have you heard anything regarding efficiency improvements? Or will a G70 quad be comparable to an NV40 quad?
overclocked
12-May-2005, 04:14
Also if they can make HDR work with AA is of interest for many i think.
When the dust settles a bit, it'd be nice to get an interview with someone at NV to walk thru all that "from there to here". Some loose ends flapping about.
Nah, by then we'll have some new speculations to argue about :wink:
DegustatoR
12-May-2005, 13:30
Have you heard anything regarding efficiency improvements? Or will a G70 quad be comparable to an NV40 quad?
There will be some changes in pixel shaders afaik...
Presumably the major thing NVidia will do is disconnect texturing from the first shader unite so that texturing doesn't reduce pipeline capacity.
http://www.beyond3d.com/previews/nvidia/nv40/index.php?p=10
Jawed
Nobody putting much faith in more than six vertex units? Along with that, I'd expect higher performance while filtering and blending FP surfaces and possibly some new AA ability.
trinibwoy
13-May-2005, 03:57
If ATI is going 90nm while Nvidia is going 110nm, doesn't that give ATI a lot more headroom in terms of transistor count and clockspeed?
vBulletin® v3.8.6, Copyright ©2000-2013, Jelsoft Enterprises Ltd.