[dizietsma]

Komplett have 150+ 512Mb GTX's coming in for early Feb it seems, so it looks like another batch will be ready for then, whether it be due to core or memory supply being finally resolved, and so this cannot be described as a one off product. Whether it is good value or not then is another matter.

As most reviewers on the original 7800 GTX got 490-520Mhz range I would say that the absolute yield for a volt bumped 7800 at 550 is also pretty good, but then nvidia ( and Ati for that matter ) will always factor in a safety margin for less than ideal conditions out there in the big wide world ( SLI in a non air condition flats near the equator using $10 700w PSU's being a worst case scenario ) .

Some of the AIB manufacturers are less circumspect of course and they are putting out 7800's nearer to 500, with default voltage, to get that marketing advantage.

Note also that Samsung is over volting it's really high speed GDDR3 as well, the lower speed chips use 1.8v whilst the fastest uses 2v+ . This seems to indicate that yields there might be an issue, but you would still assume many thousands can be chunked out at this level and not just hundreds

In summary I just do not think there is enough information at present to be able to accurately guess the clockspeed of the G71, even if it had the same 110nm process and number of pixel pipelines / ROPs etc, when you throw in the move to 90nm and the changing spec it makes a speed guess very tricky.

Trust me though, it's 650Mhz

[neliz]

Originally Posted by dizietsma

Komplett have 150+ 512Mb GTX's coming in for early Feb it seems, so it looks like another batch will be ready for then,

They have been saying that for two months now.. so 3 months after the product first went to market, a second batch would appear

The problem is, February seems a little bit to late for me, when the GTX512 will finally be on the shelves AND priced at $649, how hard would it be to resist the 1900XT lying next to it for, say $599?

the 512 was one squirt in a peeing contest.

Besides, if the memory of the GTX512 was the problem of it's shortage, how long would we have to wait for mass distribution of the G71 then?

[satein]

Originally Posted by neliz

Besides, if the memory of the GTX512 was the problem of it's shortage, how long would we have to wait for mass distribution of the G71 then?

There would be another option that NV may stockpile the memory for G71 too . As if they want the part to be available at lunch date, NV need any measure to make sure of it then. GTX512 would be something like a thing that make people stop and wait not really go immedietly to X1800XT (and it works its spell really well), and then by that Fab time frame they may offer G71 instead of GTX512.
Edit: In addition point, if the G71 is also pin2pin compatible with G70... that means it can be fixed on GTX512 board and any AIB would produce it easily with previous process setup for G70 too.

[Megadrive1988]

not wanting to read through the entire thread, how many ROPs are rumored to be in G71 ?

[Razor1]

Originally Posted by satein

There would be another option that NV may stockpile the memory for G71 too . As if they want the part to be available at lunch date, NV need any measure to make sure of it then. GTX512 would be something like a thing that make people stop and wait not really go immedietly to X1800XT (and it works its spell really well), and then by that Fab time frame they may offer G71 instead of GTX512.

This is exactly what I'm thinkin

[Chalnoth]

Originally Posted by Megadrive1988

not wanting to read through the entire thread, how many ROPs are rumored to be in G71 ?

I think we should all be expecting 16 again. Not really much reason to have more with a 256-bit memory interface.

[walterman]

Originally Posted by Ailuros

Yes it truly is 500MHz, but that still doesn't mean that your theory about the ROPs has any legs. Not by far.

Well, help me then to understand the big picture.

[Geo]

Originally Posted by ninelven

And why would the burden of proof fall upon the more likely, more logical answer? Or why would the less likely scenario not require any proof and be accepted as fact? Maybe it is the same logic which holds that G71 being a smaller core on the same proccess as R580 must be clocked significantly lower.

We know that, do we?

With GTX512 being a spectacular exception (and most likely planned as a short-term stop-gap), Jen-Hsun has been very vocal (that's not even strong enuf --more like "written on my soul in letters of fire" ) that the lesson NV learned from NV30 is that relying on clocks rather than transistors sucks. He's even pointed at that multiple times as a major factor in the NV renaissance over the last 18 months and philosophical difference with ATI.

That needs to be taken into consideration here, tho I don't have a clue how it will manifest itself in G71 --tho I tend to think they won't push the clocks as hard as ATI, unless G71 is another "short-termer". And maybe it is, if it comes out in March and NV50 comes out in June, which would be one way to read the timeframes that NV has pointed at (but not the only way).

[fbomber]

I´ve seen some people here saying that the r580 will be "too much for the g71 to handle".
What if NVIDIA adds one texture unit per pipeline, leaving the two full alus free for pixel shader processing, instead of one having to deal with texture operations?

Just my thoughts, and I want to know from you if this would be any good, and if so, what kind of performance gains would be expected, compared to the g70.

[Razor1]

Originally Posted by fbomber666

I´ve seen some people here saying that the r580 will be "too much for the g71 to handle".
What if NVIDIA adds one texture unit per pipeline, leaving the two full alus free for pixel shader processing, instead of one having to deal with texture operations?

Just my thoughts, and I want to know from you if this would be any good, and if so, what kind of performance gains would be expected, compared to the g70.

Well the problem with that is nV will have to redo thier drivers a bit, ATi got around that by introducing a mid range x1600, so ATi tuned thier drivers already

[snk]

Originally Posted by Chalnoth

I think we should all be expecting 16 again. Not really much reason to have more with a 256-bit memory interface.

What about this?

24-48-24-48 CATZ (Color, Alu, Texture, Z?) would indicate 24 ROPs.

[Apple740]

Sorry if i missed it but any news of FP HDR + AA support on G71?

[ninelven]

Originally Posted by AlphaWolf

Nothing hypocritical about it, its called sarcasm. You don't need proof or evidence to support your claims neither does anyone else weeeee lets all make stuff up and pretend its the truth.

Lame. This kind of stuff goes on all the time. People take Dave's comments on good faith with no problem (which is because he usually knows what he is talking about but the point should be clear), and DW, ANova, Chalnoth, Uttar, or prettymuch anyone else for that matter never catches flak for unsubstantiated viewpoints. Prime example: the state of yields for G70. I fail to see why I should be held to a higher standard. Hell, glancing over your posts it took me just a few seconds to find a nice example.

Originally Posted by Geo

unless G71 is another "short-termer"

Well, more short-term than R580 (cores). I have no idea what G71 is, so I was only talking with regard to one of the speculated configurations (16/32) in making the size comment. I would think though that, barring substantial changes, nvidia is rather anxious to depart from G7X in the high end.

[Razor1]

Originally Posted by Apple740

Sorry if i missed it but any news of FP HDR + AA support on G71?

Nope too little info out there about the g71 to really presume anything at the moment, other then the process.

[AlphaWolf]

Originally Posted by ninelven

Lame. This kind of stuff goes on all the time. People take Dave's comments on good faith with no problem (which is because he usually knows what he is talking about but the point should be clear), and DW, ANova, Chalnoth, Uttar, or prettymuch anyone else for that matter never catches flak for unsubstantiated viewpoints. Prime example: the state of yields for G70. I fail to see why I should be held to a higher standard. Hell, glancing over your posts it took me just a few seconds to find a nice example.

Mostly people at least provide an argument to back their statements. You made a statement like it was fact and feel no need to support it.

I was one of the first to suspect ram might be an issue because the ram being used for the 7800gtx512 wasn't even announced yet. However samsung has announced mass production of it since the announcement of the gtx512 and still very few gtx512s have made it to retail. That is why I begin to doubt my own initial assertion.

[karlotta]

Originally Posted by ninelven

Lame. ...I fail to see why I should be held to a higher standard. ....

Well i see them being questioned all the time on there comments. BUT they do have a track record... you dont yet. thats the real dif. AND the phyiscal truth of the lack of 550cores in volume. Which goes to the topic of the thread ,"...750hz..." that is just insane to me for a jump to 90nm as a core. And no lowk wont give that boost, maybe in the CPU/intel therory , which has no gates or transistor in the measurment. A 400mil tran part at 750hz stock is a big leap. IMO

[Razor1]

Originally Posted by karlotta

Well i see them being questioned all the time on there comments. BUT they do have a track record... you dont yet. thats the real dif. AND the phyiscal truth of the lack of 550cores in volume. Which goes to the topic of the thread ,"...750hz..." that is just insane to me for a jump to 90nm as a core. And no lowk wont give that boost, maybe in the CPU/intel therory , which has no gates or transistor in the measurment. A 400mil tran part at 750hz stock is a big leap. IMO

Why not, ATi wasn't able to get thier .11 x700 over 500 mhz, so they increased (more then double if you take the x700 as a base) thier tranisitor counts and increased thier clocks 45%, why can't nV do the same? Are they not as competent as ATi's enginneers? If anything nV's cores are structured better for higher clocks then previous generateion ATi cores.

[AlphaWolf]

Originally Posted by Razor1

Why not, ATi wasn't able to get thier .11 x700 over 500 mhz, so they increased (more then double if you take the x700 as a base) thier tranisitor counts and increased thier clocks 45%, why can't nV do the same? Are they not as competent as ATi's enginneers? If anything nV's cores are structured better for higher clocks then previous generateion ATi cores.

Are you comparing x700 to r520 and using that for a basis of g70 to g71?

[Jawed]

Who knows, ATI's overall feeling about 90nm at TSMC was that it was the easiest transition ever done.

Though I have absolutely no qualifications for saying this (and am in two minds about being so bold), I think the fact that the transition was from 130nm low-K to 90nm low-K was prolly a key factor in the ease of transition. I expect TSMC designed it to work like that.

NVidia didn't, seemingly, have problems moving NV4x between IBM's 130nm and TSMC's 110nm - so I guess sometimes we shouldn't be too pessimistic...

Jawed

[karlotta]

Originally Posted by Razor1

Why not, ATi wasn't able to thier .11 x700 over 500 mhz, so they increased (more then double if you take the x700 as a base) thier tranisitor counts and increased thier clocks 45%, why can't nV do the same? Are they not as competent as ATi's enginneers? If anything nV's cores are structured better for higher clocks then previous generateion ATi cores.

i would not take the x700 @110nm as the base, i would use the r420 @130nm, but then again they are diff in the design than the r520 and not a real shrink. The 130nm, and 90nm are lowk and considerd real nodes, 110 i is just a die shink, for cost/ yeilds. NVDA did a bang up job at 110nm. Clocked the hell out of it for the uber 550hz, but realy just stay at 430hz for best case in cooling and yields. So if i go by the "shrink" of ATI r420 to r520 the clocks are...520hz to 600hz. NOT a huge jump. then 430hz to 750hz is a huge jump.

[Razor1]

Originally Posted by karlotta

i would not take the x700 @110nm as the base, i would use the r420 @130nm, but then again they are diff in the design than the r520 and not a real shrink. The 130nm, and 90nm are lowk and considerd real nodes, 110 i is just a die shink, for cost/ yeilds. NVDA did a bang up job at 110nm. Clocked the hell out of it for the uber 550hz, but realy just stay at 430hz for best case in cooling and yields. So if i go by the "shrink" of ATI r420 to r520 the clocks are...520hz to 600hz. NOT a huge jump. then 430hz to 750hz is a huge jump.

Why would it be a bang up job if the .11 process is not made for increasing clocks from .13, its a cost cutting process from what I understand.

So we might see a huge increase in clocks going to .09 with low k, because the .09 process is made for increasing clocks which even TSMC stated 100% performance increase over .13. (pretty sure they were talking about thier low k process but they didn't specifically mention that, but that was the only process they listed for .13)

[Razor1]

Originally Posted by AlphaWolf

Are you comparing x700 to r520 and using that for a basis of g70 to g71?

No just saying that it can be a possiblity by going to .09 the clock increases were made in mind with the .09 process.

[karlotta]

Originally Posted by Razor1

Why would it be a bang up job if ...

Bangup job means a realy realy good job. They maxed the 110nm. BUT 110 to 90nm is not considerd a shrink its a whole new node. And i believe its a new design for NVDA not just a 90nm g7x , NVDA has the whole rename , keep name thang all messed up.. But i dont know, just my understanding of the history of sillycone is a 750hz leap is too big. And the cooling would have to be loud. IMO

[Razor1]

Originally Posted by karlotta

Bangup job means a realy realy good job. They maxed the 110nm. BUT 110 to 90nm is not considerd a shrink its a whole new node. And i believe its a new design for NVDA not just a 90nm g7x , NVDA has the whole rename , keep name thang all messed up.. But i dont know, just my understanding of the history of sillycone is a 750hz leap is too big. And the cooling would have to be loud. IMO

Oh ok ok sorry miss understood .

I think it is possible to up the clocks quite a bit from a process change like this one, unfotunately other then CPU's there really isn't much data on it.

Could be a new design, nV could have just put up a smoke screen.

[ninelven]

Originally Posted by karlotta

AND the phyiscal truth of the lack of 550cores in volume.

What physical truth? Replace cores with cards and you would have something, unless you are referencing something else?