Some news on the 7900!

[chilled]

Over here at Hexus.

I think the main point was that the 7900 would only need a core clock of 430MHz to put it on a level playing field with the 7800 GTX 512 @ 550MHz. That figures due to the 32 pipeline design.

NV are moving to the 90nm process and so the 7900 is likely to come in at 700MHz corespeed. Is it just me, or is that quite a big refresh to the 7800? I know its a year on in March 2006, but still that almost seems next gen!

ATI's X1900 is rumoured to be faster than the 7800 512, but I very much doubt it'll be as fast as this. I hope they have a 20 or 24 pipe card as a refresh refresh.

Personally I couldn't care less who is actually a little faster, as long as its more or less a level playing field. I just hope we're not returning to the 6 month like product cycles as we do still need ATI in the mix otherwise its bad news for us consumers...

 

[ScrewFace]

Don't worry. ATI is as big and rich as nVidia. ATI'll most likely come out with X2800 XT before nVidia's next generation as they did with the 9700 Pro. Now if only if Intel and AMD would get off their asses and start ramping up CPU speeds!

 

[Wentelteefje]

Although those figures really seem/are impressing, ATI is trying something else... Instead of focusing on a great number of pixel pipelines, their X1900 will have 16 TMU's, but will be able to process 3 pixel shaders per pipeline, equaling 48 pixel shader processors...

nVidia decides to do otherwise, but it must be said that the previous X1800 managed to beat the 7800 as well, despite having 8 true pixel pipelines less... BTW ATI has something very nice with their programmable memory controller (seeing that one driver update could improve frame rates with 30% in OpenGL)...

Their trying both to take another path (nVidia relying on textures, while ATI sees the future in more pixel processing), but IMO both know very well what they're doing... The PC market has grown too big for one of them to release a bad product...

 

[akugami]

Not any different from the wild ATI claims on how much better it's X1800's would be compared to the 7800's. In some very specific cases with heavily optimized game code it may even be true. However, unless nVidia has done some serious tinkering with the GPU core beyond the process shrink and adding some pipelines...I doubt they'll get such a heavy boost in GPU power, at least on a per mhz comparison.

That the G71's core is an improvement on the G70 is pretty much without doubt. I'm just doubting the wild claims that are being put out on just how much it has improved. That a completely new generation of video cards can double the performance should be without question. That a refresh product (for the most part) can double the performance of the previous card seems wildly optimistic.

Either way, I'd like to see real world benchmarks before I make my next video card purchase.

 

[Munky]

Does anyone honestly think Nv can make a 32 pipe gpu clocked at 700+ mhz? What would be the transistor count and what kind of yields can they expect from it. If I had to chose between higher clocks and more pipes, I'd pick the more pipes approach because it's a "safer" bet when designing a gpu, but clocking all those pipes at 700 mhz is a very ambitious expectation. I'd put the wide availability model clocks somewhere between 550-600 mhz, and then NV can release a limited edition ultra using cherry picked cores at 700 mhz. I dont think 700mhz off the bat is a realistic expectation, but it would be impressive if they pulled it off (and the cards were available in retail for more than 2 weeks )

 

[allies]

Originally posted by: munky
Does anyone honestly think Nv can make a 32 pipe gpu clocked at 700+ mhz? What would be the transistor count and what kind of yields can they expect from it. If I had to chose between higher clocks and more pipes, I'd pick the more pipes approach because it's a "safer" bet when designing a gpu, but clocking all those pipes at 700 mhz is a very ambitious expectation. I'd put the wide availability model clocks somewhere between 550-600 mhz, and then NV can release a limited edition ultra using cherry picked cores at 700 mhz. I dont think 700mhz off the bat is a realistic expectation, but it would be impressive if they pulled it off (and the cards were available in retail for more than 2 weeks )

Agreed :thumbsup:

 

[Killrose]

All just rumors.

R

 

[Elfear]

Originally posted by: allies

Originally posted by: munky
Does anyone honestly think Nv can make a 32 pipe gpu clocked at 700+ mhz? What would be the transistor count and what kind of yields can they expect from it. If I had to chose between higher clocks and more pipes, I'd pick the more pipes approach because it's a "safer" bet when designing a gpu, but clocking all those pipes at 700 mhz is a very ambitious expectation. I'd put the wide availability model clocks somewhere between 550-600 mhz, and then NV can release a limited edition ultra using cherry picked cores at 700 mhz. I dont think 700mhz off the bat is a realistic expectation, but it would be impressive if they pulled it off (and the cards were available in retail for more than 2 weeks )

Agreed :thumbsup:

I'm thinking the same thing. Two extra quads and a 150-270MHz core increase (depending on how you look at it) sounds a little far-fetched to me but I'd love to be proven wrong as that would be a monster card.

 

[nitromullet]

Originally posted by: Killrose
All just rumors.

R

yeah, c'mon...

This was already a repost 2 days ago...

http://forums.anandtech.com/messageview...atid=31&threadid=1777294&enterthread=y

Does the OP really think this news would go unnoticed by this forum for 2 days?

 

[knifemyglitter]

actually, nvidia is a much bigger corp, just check the stock/company specs out. Lets just hope both cards coming out will be comparable, so that they'll be priced competitively.

 

[The Sly Syl]

Originally posted by: knifemyglitter
actually, nvidia is a much bigger corp, just check the stock/company specs out. Lets just hope both cards coming out will be comparable, so that they'll be priced competitively.

Considering how much pure console research and devolopment ATI does, its surprising that nvidia is a bigger corp.

Then again, I suppose Nvidia has a more "mature" chipset line that takes some time to devolop and produce as well. Nforce has been much more succesful than any of ATi's chips, i'd imagine.

 

[Pete]

ATI's still got more employees, but NV's currently got almost twice the market cap. If ATI executes their X1900 launch well and PS3 doesn't blow X360 out of the water and Revolution turns out nice, I'm guessing market caps will somewhat equalise. But NV's been firing on most/all cylinders for quite a few quarters, so ATI's got to get some consistency going in the good news dept. to boost their stock price. Click on "Key Statistics" to see that ATI's got more shares outstanding (so stock prices aren't exactly comparable, but pretty close) and that NV's whipping them in profit margins (tho revenue is about even).

 

[Keysplayr]

Originally posted by: munky
Does anyone honestly think Nv can make a 32 pipe gpu clocked at 700+ mhz? What would be the transistor count and what kind of yields can they expect from it. If I had to chose between higher clocks and more pipes, I'd pick the more pipes approach because it's a "safer" bet when designing a gpu, but clocking all those pipes at 700 mhz is a very ambitious expectation. I'd put the wide availability model clocks somewhere between 550-600 mhz, and then NV can release a limited edition ultra using cherry picked cores at 700 mhz. I dont think 700mhz off the bat is a realistic expectation, but it would be impressive if they pulled it off (and the cards were available in retail for more than 2 weeks )

Isn't NV going low-k with the G71? ATI's clock speeds jumped 170MHz from a 350MHz 9800pro with 8 pipes to a 520MHz X800XTPE with 16 pipes (100% more pipes) utilizing low-k.

If the G70 can hit 550 without low-k on 110nm's, I do not see any issue whatsoever with a 90nm low-k G71 with 32 pipes (only a 25% increase in pipes) gaining 150MHz.

 

[A554SS1N]

Originally posted by: munky
Does anyone honestly think Nv can make a 32 pipe gpu clocked at 700+ mhz? What would be the transistor count and what kind of yields can they expect from it. If I had to chose between higher clocks and more pipes, I'd pick the more pipes approach because it's a "safer" bet when designing a gpu, but clocking all those pipes at 700 mhz is a very ambitious expectation. I'd put the wide availability model clocks somewhere between 550-600 mhz, and then NV can release a limited edition ultra using cherry picked cores at 700 mhz. I dont think 700mhz off the bat is a realistic expectation, but it would be impressive if they pulled it off (and the cards were available in retail for more than 2 weeks )

Well, ATi is looking at 48 - so if they can do that, NVidia can manage 32 - I very much doubt you'd be thinking the safe approach is 32 pipes compared to 48 pipes would you?? If anything is being ambitious, it's ATi's approach which is more innovation than evolution. I think both companies will do just fine in there usual low-yield-high-end-chip way....

 

[fierydemise]

Originally posted by: A554SS1N
Well, ATi is looking at 48 - so if they can do that, NVidia can manage 32 - I very much doubt you'd be thinking the safe approach is 32 pipes compared to 48 pipes would you?? If anything is being ambitious, it's ATi's approach which is more innovation than evolution. I think both companies will do just fine in there usual low-yield-high-end-chip way....

Its not 48 pipes, I don't exactly understand it but as far as I know its 16 pipes with each pipe capable of doing 3 shader operations. Someone please clarify/correct me.

 

[A554SS1N]

Originally posted by: fierydemise

Originally posted by: A554SS1N
Well, ATi is looking at 48 - so if they can do that, NVidia can manage 32 - I very much doubt you'd be thinking the safe approach is 32 pipes compared to 48 pipes would you?? If anything is being ambitious, it's ATi's approach which is more innovation than evolution. I think both companies will do just fine in there usual low-yield-high-end-chip way....

Its not 48 pipes, I don't exactly understand it but as far as I know its 16 pipes with each pipe capable of doing 3 shader operations. Someone please clarify/correct me.

Yeah, I know, not in the traditional sense, but there are other units as far as i know which make an effective 48 pipes (I think) - although I'm only going on wild website stories

 

[MBrown]

I can't we are talking about 7900's now. It seems like the 7800's and the x1800's just came out! What a race this is.

 

[Munky]

Originally posted by: A554SS1N

Originally posted by: fierydemise

Originally posted by: A554SS1N
Well, ATi is looking at 48 - so if they can do that, NVidia can manage 32 - I very much doubt you'd be thinking the safe approach is 32 pipes compared to 48 pipes would you?? If anything is being ambitious, it's ATi's approach which is more innovation than evolution. I think both companies will do just fine in there usual low-yield-high-end-chip way....

Its not 48 pipes, I don't exactly understand it but as far as I know its 16 pipes with each pipe capable of doing 3 shader operations. Someone please clarify/correct me.

Yeah, I know, not in the traditional sense, but there are other units as far as i know which make an effective 48 pipes (I think) - although I'm only going on wild website stories

It's 16 pipes but with 3 pixel shaders in each pipe, so it's 48 PS total. Just like the x1600 is 4 pipes with 12 PS. By going this route, they're not putting 48 full-fledged pipes, so it will have roughly the same transistor count as the 32 pipe gf7900, IMO.

 

[rbV5]

Originally posted by: MBrown
I can't we are talking about 7900's now. It seems like the 7800's and the x1800's just came out! What a race this is.

Well, you got that half right

 

[BenSkywalker]

The 'pipe' numbering is getting bothersome.....

The x1900 is expected to have 48ALUs- the same number as the 7800GTX- while it has 16ROPs, compared to the GTXs 24. Now, the level of functionality between the ALUs is different so comparing them simply by using the raw number of them doesn't come close to working. The X360's GPU as 8 ROPs and 48 ALUs, so you can call it a 8 pipe part, that performs like a 24 ALU part although it has 48 ALUs.......

I don't like the current naming convention. We should establish a baseline for a simple string command and figure out how many ALUs will be tied up to execute it and then divide that number to come up with a useable number for the sake of the masses. Come up with something easy to digest for those who want a simple answer like was done with the pixel/texel numbers.

 

[Keysplayr]

Originally posted by: munky

Originally posted by: A554SS1N

Originally posted by: fierydemise

Originally posted by: A554SS1N
Well, ATi is looking at 48 - so if they can do that, NVidia can manage 32 - I very much doubt you'd be thinking the safe approach is 32 pipes compared to 48 pipes would you?? If anything is being ambitious, it's ATi's approach which is more innovation than evolution. I think both companies will do just fine in there usual low-yield-high-end-chip way....

Its not 48 pipes, I don't exactly understand it but as far as I know its 16 pipes with each pipe capable of doing 3 shader operations. Someone please clarify/correct me.

Yeah, I know, not in the traditional sense, but there are other units as far as i know which make an effective 48 pipes (I think) - although I'm only going on wild website stories

It's 16 pipes but with 3 pixel shaders in each pipe, so it's 48 PS total. Just like the x1600 is 4 pipes with 12 PS. By going this route, they're not putting 48 full-fledged pipes, so it will have roughly the same transistor count as the 32 pipe gf7900, IMO.

Any thoughts on the low-k thing I mentioned above?

 

[Keysplayr]

Originally posted by: BenSkywalker
The 'pipe' numbering is getting bothersome.....

The x1900 is expected to have 48ALUs- the same number as the 7800GTX- while it has 16ROPs, compared to the GTXs 24. Now, the level of functionality between the ALUs is different so comparing them simply by using the raw number of them doesn't come close to working. The X360's GPU as 8 ROPs and 48 ALUs, so you can call it a 8 pipe part, that performs like a 24 ALU part although it has 48 ALUs.......

I don't like the current naming convention. We should establish a baseline for a simple string command and figure out how many ALUs will be tied up to execute it and then divide that number to come up with a useable number for the sake of the masses. Come up with something easy to digest for those who want a simple answer like was done with the pixel/texel numbers.

Unless I read that wrong, you just said the GTX has 24 ROP's. It has 16. Did you mean pipes?

 

[RussianSensation]

G71
"the money-shot is that a 700MHz NVIDIA GeForce 7900 is currently coming in at circa 13,000 3DM2K5's on an AMD Athlon 64 FX 60 system." - hexus

R580
"A single card can score 11000+ in 3Dmark05 which is faster than both single Geforce 7800 GTX 512 card and Radeon X1800 XT including the overclocked ones. The Hexens were right that with the fast CPU you can get around 13500 in 3Dmark05. I know that in games such as FEAR Nvidia is loosing significantly to R580 as much as fifty per cent faster in some scenarios." - the inquirer.

So both cards are getting close to 13000 marks in 3dmark05. This is going to be close

Now if all refreshes brought this much increases. ...I can't even imagine the jump with G80 and R600.

 

[Elfear]

Hmmm, I'm not too familiar with VR-Zone but it looks like they have some real results posted. Looks to be inline with the rumors we've been hearing.

 

[the Chase]

Seems stuff is starting to leak out all over the place on these cards. Thread over on XS has a photo of an Asus 1900 with the retail box it came out of in the same photo. These must be or about to be shipping to retailors!