Beyond3D Wolves Needed to Tear This One Apart
- Thread starter JoshMST
- Start date
- Status
It's a well written and thoughtful article. So yeah, maybe not entirely accurate, but whatever. I've yet to see an article written by anyone where I agree with every point made.
What I find particularily funny is how some people spit words like "bias" at Josh when they themselves need to take a close look in the mirror.
What I find particularily funny is how some people spit words like "bias" at Josh when they themselves need to take a close look in the mirror.
Josh, you gullible, narcissistic avocado--I thought it was a good piece. 
Three notes, if this wolf in sheep's clothing (literally, sheeps' hair) may:
1) It was pointed out to me that the 6600 doesn't have/need a crossbar between its pipes and ROPs (p. 3), but rather a simple FIFO buffer.
2) I thought the PS and (especially the) VS units offered essentially the same functionality as those on R300, with just the PS units gaining registers to execute longer shaders?
3) Re: your conclusion, I think it's more correct to say nV has surpassed ATi in features in this gen, and caught up in terms of overall performance. They done good this round.
Three notes, if this wolf in sheep's clothing (literally, sheeps' hair) may:
1) It was pointed out to me that the 6600 doesn't have/need a crossbar between its pipes and ROPs (p. 3), but rather a simple FIFO buffer.
2) I thought the PS and (especially the) VS units offered essentially the same functionality as those on R300, with just the PS units gaining registers to execute longer shaders?
3) Re: your conclusion, I think it's more correct to say nV has surpassed ATi in features in this gen, and caught up in terms of overall performance. They done good this round.
MasterBaiter
Regular
OMG!!!! Your parents must be brothers!!! How did you come to such crazy conclusions!?!
I'd agree as well. If these two companies were compared to surgeons, ATI would be using a scalpel, and Nvidia a broadsword though.
I thought the article was an interesting read, and since it is speculation... I don't think anyone should get in much of a huff if they are taking it with the proper amount of salt.
Pete said:
I'd agree as well. If these two companies were compared to surgeons, ATI would be using a scalpel, and Nvidia a broadsword though.
I thought the article was an interesting read, and since it is speculation... I don't think anyone should get in much of a huff if they are taking it with the proper amount of salt.
Diggy, I'm rather shocked and awed that you're replacing Hellbinder as the Rx20 cheerleader when almost no information is available on it.
Also, ATI's products are behind on:
AF (still. RAA)
drivers (Windows, still, and Linux? it's about as much of a contest in a Ray Lewis versus Mary Lou Retton boxing match)
they're ahead, significantly, in AA. I'm too lazy to get into SM3.0 versus SM2.0x.
Diggy, for you to be accusing someone of green-tinted bias is kinda funny.
Also, ATI's products are behind on:
AF (still. RAA)
drivers (Windows, still, and Linux? it's about as much of a contest in a Ray Lewis versus Mary Lou Retton boxing match)
they're ahead, significantly, in AA. I'm too lazy to get into SM3.0 versus SM2.0x.
Diggy, for you to be accusing someone of green-tinted bias is kinda funny.
Hellbinder
Banned
Uh.. I have not said anything about R520 of any significance like ever. I have no interest whatsoever in Anyones hardware nor have i for a long time. So i have been replaced by no one.The Baron said:
Also.. Its funny that you have become more and more *biased* as time goes on as your post indicates. Where as I can honestly say I am Completey, unbiased at the present time. Yet you still make little digs at me like this and others.
Its completely baseless.
The Baron said:
What does RAA mean? Regardless, from my expereances the AF quailty is a toss up; each have their own issuesbut both are good for the most part. Although it would be nice if we could have cards with all the power these new ones have and the AF quality of a Geforce3.
MasterBaiter
Regular
Most likely they will be only modestly faster than the current versions is my guess.
CMAN said:
Yup correct. Though there are rumours floating around that ATI's "winter collection", that is the high-end part has been made in 110 nm and runs at higher clocks.
Normally a process shrink is just all about reducing die sizes -> more dies per wafer, lower costs, higher yields.
I'm still assuming that both Nvidia and ATI will release 90 nm parts next year.
RAA is me complaining about a lack of angle-independent AF.kyleb said:
Yeah, I think the days of straight trilinear filtering on every stage are gone... *sniff* I'll miss it.
Hi guys, sorry it took so long for me to get back here. Had to sit and think about things for a while, and really consider my stance on a few things. So I will go over each individual's questions and comments to me (where applicable).
Banksie- it seems that with each major GeForce product family, there has been fairly significant changes. While these may not have always been sweeping architectural changes, the feature and performance characteristics have been significant.
TnT -> TnT2 had more optimized pixel pipelines, more efficient in comparison even when clocked at the same speed. There was a process change (350 nm to 250 nm), which required a redesign of the entire chip in regards to the new design libraries the process brought. No such thing as a "simple shrink".
GF 256 -> GF2 increased to two texture units per pipeline (though there is some interesting debate about if GF256 merely had a defective 2nd unit that could only do some work), as well as another new process shift (220 nm to 180 nm). Again, more design work, not a simple shrink.
GF2 -> GF3 essentially a new architecture, but still based on previous NV work. Programmability introduced, cross bar memory controller, other types of compression. I would consider this a major architectural change.
GF3 -> GF4 was perhaps the smallest leap, but still a lot of work was done. Extra vertex unit, pixel pipelines received a lot of work, a more efficient architecture (as evidenced by the pretty significant speed increase between the two).
Of course the least change was with the GF3 to GF3 Ti series. Speed bump only in terms of memory and core clocks. For the others, there is no lack of work being done by NV and its engineers. Same can be said for ATI and its products, while some may appear at the surface to be simple changes, the amount of work that goes into each new product is mind boggling. Seriously, hats off to both companies for their dedication to the industry, and pushing the feature and performance limits.
Dave Baumann- I am slowly getting the hints (trail of crumbs left on the ground), and I have a pretty good idea what you are getting at. For better or for worse, I don't have the contacts you have that will tell the concrete information. This is probably a good thing, since if I knew for sure, and didn't want to piss of my contacts, then I wouldn't be able to speculate anymore and write such articles!
DW- oh, its the man! Me and my green bias indeed!
As for transistor count, I really have no clue how either of them count, but if you look at the physical size of the chips, the R420 and NV40 are very close together in size. So, while overall transistor count may not be that big a deal in comparison, the fact that ATI has a die that is almost the same size as the NV40 means that they are both hitting some of the same problems and limits that current process technology impose. NV may have a slight edge because they use IBM with 300 mm wafers. From my understanding the TSMC 130 nm Low-K line is based on 200 mm wafers. ATI needs to order many more wafer starts than NV does to supply the market with product. I have recently heard from a source (ooooh, scary!) that ATI is having a hell of a time with their top of the line parts, and that we won't really be seeing anymore X800 XT's for some time, until ATI is able to release a 110 nm version- take that puppy with a grain of salt, but looking at current availability of such products, I find that I agree with the overall feeling of that rumor. Again, NV is hitting the same wall with their 6800 Ultra products, but it seems that they are faring slightly better than ATI in terms of supply a top of the line card to OEM's.
The "2nd best" portion you talked of with regards to a loyal consumer base... I am afraid I worded that poorly. It should have said something like this: "ATI has built a very loyal consumer base that would continue to support ATI even if they had a future part that was 2nd best in the industry". Sorry, my bad on that one!
The "matching NV step for step" comment. My point was that NV released the 6800, and half a month later ATI came out with the X800 products. In the Fall NV came out with the 6600 products, and half a month later ATI had the X700. I would say that was proper usage of "matching NV step for step" with regards to product releases. NV may have been first, but ATI has always stepped in with a competing product some time afterwards. Now, if NV had released after ATI, then I would have said, "NVIDIA has been able to match ATI step for step".
What I considered the Golden Age of 3D Graphics was around the time when the Voodoo 2 was launched. 3dfx, Rendition, NVIDIA, Real 3D, ATI, and a couple of others all had 3d chips out that were very competitive. Not only that, but companies like STB, Diamond, Orchid, Creative, Canopus, and others all had products that had many different characterstics in overall design, as well as unique drivers for each product (reference drivers were not as available, plus these companies liked to put their own kind of flare into them). Just my opinion of course, but I have fond memories of that time.
PeterACE- you are kind of a nasty pants
I had actually based my "shallow and wide" comments on Dave's work, and from a high level perspective I still stand by the comment. I read through that post again, and I think I can argue that when comparing the R300 to the NV3x, ATI took a more shallow and wide approach than NVIDIA did. I don't think I am wrong in saying that.
I had forgotten about the geometry instancing support on the earlier R3x0 series. My bad on that one!
My intention for this article was not only to share my information and perhaps inform a reader or two, but also to drive discussion about this topic and the industry in general. So, would you suggest that I just shut up and not write anything? I know I have learned alot from such discussions, so I apologize if this has been a waste of your time.
Pete- good comments, I enjoy the humor!
I am not sure of the 6600 and its crossbar, or that it just needs a FIFO to get all 8 pipes to work with 4 ROPS. You would have to ask a NV engineer for specifics.
As for the changes in the R420 pipeline as compared to the R3x0 series, you would have to ask one of the ATI guys. I would imagine that such a change is not trivial from any standpoint. Even throwing in a bunch of new registers would significantly impact the overall layout of any design.
As for who has the performance lead in the industry? I think that both NV and ATI excel in different areas, but overall it is a wash. This is definitely a personal preference decision when buying a product from either company. Both have unique features over the other, but when everything is added together they both have very competent products that perform well in current applications.
Hellbinder- nice to see you again!
CMAN- looking at the clock speeds that both NV and ATI are able to achieve with their 120 million transistor products, I think it is becoming pretty obvious that TSMC has in fact tweaked the 110 nm process to have better transistor performance than their stock 130 nm process. I think that it is quickly approaching the transistor performance of TSMC's 130 nm Low-K process. I would say that by Q2 2005, it will probably be faster than 130 nm Low-K.
Thats all for now, I hope to hear from you all again soon!
Banksie- it seems that with each major GeForce product family, there has been fairly significant changes. While these may not have always been sweeping architectural changes, the feature and performance characteristics have been significant.
TnT -> TnT2 had more optimized pixel pipelines, more efficient in comparison even when clocked at the same speed. There was a process change (350 nm to 250 nm), which required a redesign of the entire chip in regards to the new design libraries the process brought. No such thing as a "simple shrink".
GF 256 -> GF2 increased to two texture units per pipeline (though there is some interesting debate about if GF256 merely had a defective 2nd unit that could only do some work), as well as another new process shift (220 nm to 180 nm). Again, more design work, not a simple shrink.
GF2 -> GF3 essentially a new architecture, but still based on previous NV work. Programmability introduced, cross bar memory controller, other types of compression. I would consider this a major architectural change.
GF3 -> GF4 was perhaps the smallest leap, but still a lot of work was done. Extra vertex unit, pixel pipelines received a lot of work, a more efficient architecture (as evidenced by the pretty significant speed increase between the two).
Of course the least change was with the GF3 to GF3 Ti series. Speed bump only in terms of memory and core clocks. For the others, there is no lack of work being done by NV and its engineers. Same can be said for ATI and its products, while some may appear at the surface to be simple changes, the amount of work that goes into each new product is mind boggling. Seriously, hats off to both companies for their dedication to the industry, and pushing the feature and performance limits.
Dave Baumann- I am slowly getting the hints (trail of crumbs left on the ground), and I have a pretty good idea what you are getting at. For better or for worse, I don't have the contacts you have that will tell the concrete information. This is probably a good thing, since if I knew for sure, and didn't want to piss of my contacts, then I wouldn't be able to speculate anymore and write such articles!
DW- oh, its the man! Me and my green bias indeed!
As for transistor count, I really have no clue how either of them count, but if you look at the physical size of the chips, the R420 and NV40 are very close together in size. So, while overall transistor count may not be that big a deal in comparison, the fact that ATI has a die that is almost the same size as the NV40 means that they are both hitting some of the same problems and limits that current process technology impose. NV may have a slight edge because they use IBM with 300 mm wafers. From my understanding the TSMC 130 nm Low-K line is based on 200 mm wafers. ATI needs to order many more wafer starts than NV does to supply the market with product. I have recently heard from a source (ooooh, scary!) that ATI is having a hell of a time with their top of the line parts, and that we won't really be seeing anymore X800 XT's for some time, until ATI is able to release a 110 nm version- take that puppy with a grain of salt, but looking at current availability of such products, I find that I agree with the overall feeling of that rumor. Again, NV is hitting the same wall with their 6800 Ultra products, but it seems that they are faring slightly better than ATI in terms of supply a top of the line card to OEM's.
The "2nd best" portion you talked of with regards to a loyal consumer base... I am afraid I worded that poorly. It should have said something like this: "ATI has built a very loyal consumer base that would continue to support ATI even if they had a future part that was 2nd best in the industry". Sorry, my bad on that one!
The "matching NV step for step" comment. My point was that NV released the 6800, and half a month later ATI came out with the X800 products. In the Fall NV came out with the 6600 products, and half a month later ATI had the X700. I would say that was proper usage of "matching NV step for step" with regards to product releases. NV may have been first, but ATI has always stepped in with a competing product some time afterwards. Now, if NV had released after ATI, then I would have said, "NVIDIA has been able to match ATI step for step".
What I considered the Golden Age of 3D Graphics was around the time when the Voodoo 2 was launched. 3dfx, Rendition, NVIDIA, Real 3D, ATI, and a couple of others all had 3d chips out that were very competitive. Not only that, but companies like STB, Diamond, Orchid, Creative, Canopus, and others all had products that had many different characterstics in overall design, as well as unique drivers for each product (reference drivers were not as available, plus these companies liked to put their own kind of flare into them). Just my opinion of course, but I have fond memories of that time.
PeterACE- you are kind of a nasty pants
I had actually based my "shallow and wide" comments on Dave's work, and from a high level perspective I still stand by the comment. I read through that post again, and I think I can argue that when comparing the R300 to the NV3x, ATI took a more shallow and wide approach than NVIDIA did. I don't think I am wrong in saying that.
I had forgotten about the geometry instancing support on the earlier R3x0 series. My bad on that one!
My intention for this article was not only to share my information and perhaps inform a reader or two, but also to drive discussion about this topic and the industry in general. So, would you suggest that I just shut up and not write anything? I know I have learned alot from such discussions, so I apologize if this has been a waste of your time.
Pete- good comments, I enjoy the humor!
I am not sure of the 6600 and its crossbar, or that it just needs a FIFO to get all 8 pipes to work with 4 ROPS. You would have to ask a NV engineer for specifics.
As for the changes in the R420 pipeline as compared to the R3x0 series, you would have to ask one of the ATI guys. I would imagine that such a change is not trivial from any standpoint. Even throwing in a bunch of new registers would significantly impact the overall layout of any design.
As for who has the performance lead in the industry? I think that both NV and ATI excel in different areas, but overall it is a wash. This is definitely a personal preference decision when buying a product from either company. Both have unique features over the other, but when everything is added together they both have very competent products that perform well in current applications.
Hellbinder- nice to see you again!
CMAN- looking at the clock speeds that both NV and ATI are able to achieve with their 120 million transistor products, I think it is becoming pretty obvious that TSMC has in fact tweaked the 110 nm process to have better transistor performance than their stock 130 nm process. I think that it is quickly approaching the transistor performance of TSMC's 130 nm Low-K process. I would say that by Q2 2005, it will probably be faster than 130 nm Low-K.
Thats all for now, I hope to hear from you all again soon!
but... GF3 and GF3 Tis used the same core, so it's not even a revision...
okay, I misunderstood then. I thought you were implying that they used a respun core with helpful clock boosts.JoshMST said:
JoshMST,
Thanks for the reply.
I'm sorry if you thought I was being nasty to you (please don't take my critsims of your article(s) as a personal attack).
It's just I don't like articles that try to resynthersize lots of complex issues and 'water it down' for the masses. Newspapers do this all the time. Lots of the essential nuances are usally missed out completely.
You second article was much much better than the first (I couldn't beleve the amount of errors and misassumptions). But It could still be much improved - You might want to consider and describe some of the design decisions behind each IHV's respective architectures (examples could be decisions about: about Power draw, OEM requirements ETC).
The reason I used the word bias was I picked up (like Dig did) on some of the 'poorly worded' parts which implied to me a 'nv orientated' way of describing some of the issues.
Thanks for the reply.
I'm sorry if you thought I was being nasty to you (please don't take my critsims of your article(s) as a personal attack).
It's just I don't like articles that try to resynthersize lots of complex issues and 'water it down' for the masses. Newspapers do this all the time. Lots of the essential nuances are usally missed out completely.
You second article was much much better than the first (I couldn't beleve the amount of errors and misassumptions). But It could still be much improved - You might want to consider and describe some of the design decisions behind each IHV's respective architectures (examples could be decisions about: about Power draw, OEM requirements ETC).
The reason I used the word bias was I picked up (like Dig did) on some of the 'poorly worded' parts which implied to me a 'nv orientated' way of describing some of the issues.
radar1200gs
Regular
What are some specific examples of these "nV oriented" explanations you pick up on?
- Status
Similar threads
