Nvidia confirms the GTX 580

[tajoh111]

GTX 580 @ 772/1544/2004MHz ? 64TMU


GTX 480 @ 925/1850/1950 MHz with 260.99 WHQL


http://www.chiphell.com/forum.phpmod...%2F%3D1&page=4

This sounds too good to be true. This type of change would definitely be worthy of being called the gtx 580 in my honest opinion because that would be pulling a woman out of a hat. Much better performance and more importantly, much better performance per watt. If that speed is optained on that type of cooler(it looks smaller than the gtx 480), the overclocking potential would be ridiculous.

But how the F would they have done it? That has got me questioning that score. But since this card appears to be in a couple people's hands, that has got to be the only thing keeping that score plausible.

I thought it could initially be a SLI gf104 board, but that would score significantly higher at those clocks. But if those clocks are wrong, then that score is awfully close to SLI gtx 460s.

http://www.tomshardware.com/reviews/...80,2694-3.html

[Macadamia]

Sorry, but that's just linear scaling (512/448)*775 probs + 5% for drivers.

Cypress X9000(++) 480 X10000 (++)

[[XC] hipno650]

oh if that score is true that is some serious stuff going down for stock speeds, we all know that FERMI ocing like no other video card ever before and it just keeps getting better as new chips come out (GTX 460 for example). i hope people also realize that the GTX 480 with that type of OC has truly MASSIVE performance compared to anything on the market today, thats getting close to 5850 crossfire range and is around a stock 5970 in real game performance. so considering the rumored specs of Cayman.... this could put them dead even... meaning purely awesome price war. Maybe I will keep my eye out for cheap GTX 480's when this and cayman come out, to run a little bit of SLI goodness...

[-Sweeper_]

PHK: informant sent me a NV sent from the Email, the letter referred to knock down the Cayman is not a problem 580. Cayman Pro XT But the letter did not say the Cayman in the end is Pro or XT, but my guess is GTX 580 faster 480 20~30%. The letter also nominal GTX 580 faster than 480 in the game 20 to 30%. NV Cayman's performance should not reach level of GTX580. Finally NV said Cayman's performance should not reach level of GTX580.

http://translate.google.com.br/trans...3D1%26page%3D1

things are getting interesting

[kadozer]

PHK: informant sent me a NV sent from the Email, the letter referred to knock down the Cayman is not a problem 580. Cayman Pro XT But the letter did not say the Cayman in the end is Pro or XT, but my guess is GTX 580 faster 480 20~30%. The letter also nominal GTX 580 faster than 480 in the game 20 to 30%. NV Cayman's performance should not reach level of GTX580. Finally NV said Cayman's performance should not reach level of GTX580.

I'd say this is a legit leak.

[-Sweeper_]

I'd say this is a legit leak.

that would mean GTX580 is 35-50% faster than a HD5870, nearly the same I'm expecting for cayman XT...

[[XC] hipno650]

that would mean GTX580 is 35-50% faster than a HD5870, nearly the same I'm expecting for cayman XT...

im thinking it would be closer to 50% then 35%

[cegras]

And where is the magical 50% coming from from less than a 50% specs increase? Are we looking at an architecture rework here like Cayman?

[Pekalion]

And where is the magical 50% coming from from less than a 50% specs increase? Are we looking at an architecture rework here like Cayman?

Not rework but maybe making it unpredictable more efficient. Fermi has shown us that is really an scalable architecture, so is viable to think that if you can create a more efficient Fermi chip, you will end with more scalability, ergo, much more headroom for extra power.

Beside of that, will it pop up at $500 or it's possible to end more expensive? I was going to make an SLI of GTX 460 but thinking that it could be 20-30% faster than a GTX480, it will give me a really great performance jump.

[JohnJohn]

I think neliz hinted gf100 to have 128 TMUs, but only 64 were functional with gtx480. now this can be the reason of such an unpredicted performance increase...

[zerazax]

I think neliz hinted gf100 to have 128 TMUs, but only 64 were functional with gtx480. now this can be the reason of such an unpredicted performance increase...

Higher clocks + 512-full SP's + full 128 TMUs... sounds like they're getting the original Fermi right at last

[Hey Zeus]

Noob Question...

What's a TMU and what is it's function?

[-=DVS=-]

Noob Question...

What's a TMU and what is it's function?

A texture mapping unit (TMU) is a component in modern graphics processing units (GPUs). A TMU is able to rotate and re-size a bitmap to be placed onto an arbitrary plane of a given 3D object as a texture.

In short more TMU = more texture rendering prowess.

[tajoh111]

I wonder if this card is going to be a good overclocker? If they reduce power consumption enough that they can use a cooler like that, while increasing speeds, this must be able to reach higher clocks than the original.

If these two cards have similar performance from the onset, I would base my purchasing decision on overclocking. Morphological AA looks nice until you realize, this is the same type of AA consoles use. In addition, these cards have so much extra power on tap, that unless your using eyefinity or 3d, you wouldn't need to use an inferior AA even at 2560*1600.

These are some pretty big ifs, though. I seriously wonder how they are getting more power out of the shaders of these things, the 772mhz vs 925mhz suggests a significant increase increase in shader performance. E.g 512/480 *772 = 823mhz and this thing is scoring probably what a 950mhz gtx 480 would score. This means a 15% power per core increase, which is similar to how much they lost from gtx 280 to gtx 480. I guess it should be quite the miracle then. But if they did this while decreasing power consumption, that would make fermi a scalable architecture again which is probably the most positive thing that could happen. But these are alot of ifs and we won't know till launch. I really didn't expect NV to provide real competition to cayman, so maybe more aggressive pricing on both sides has to occur.

[zerazax]

I wonder if this card is going to be a good overclocker? If they reduce power consumption enough that they can use a cooler like that, while increasing speeds, this must be able to reach higher clocks than the original.

If these two cards have similar performance from the onset, I would base my purchasing decision on overclocking. Morphological AA looks nice until you realize, this is the same type of AA consoles use. In addition, these cards have so much extra power on tap, that unless your using eyefinity or 3d, you wouldn't need to use an inferior AA even at 2560*1600.

These are some pretty big ifs, though. I seriously wonder how they are getting more power out of the shaders of these things, the 772mhz vs 925mhz suggests a significant increase increase in shader performance. E.g 512/480 *772 = 823mhz and this thing is scoring probably what a 950mhz gtx 480 would score. This means a 15% power per core increase, which is similar to how much they lost from gtx 280 to gtx 480. I guess it should be quite the miracle then. But if they did this while decreasing power consumption, that would make fermi a scalable architecture again which is probably the most positive thing that could happen. But these are alot of ifs and we won't know till launch. I really didn't expect NV to provide real competition to cayman, so maybe more aggressive pricing on both sides has to occur.

Probably similar to what GT200b did for the GTX 280... process improvements and not having a broken core will help

[tajoh111]

Probably similar to what GT200b did for the GTX 280... process improvements and not having a broken core will help

This is far more of a change though it appears. The difference between this 3dmark is the same between a gtx 285 and a gtx 480. I know this review is a bit older but the gtx 285 scores about 7000 and maybe a hair over, the gtx 480 around 9800 with new drivers. And this is scoring 3000 points higher than even that.

http://www.legitreviews.com/article/1258/8/

30% higher score is quite significant because I thought they were at a limit for power consumption and couldn't really roid anything up or increase the clocks.

[cronicash]

Forgive me if i posted something here that has already been posted...

Not sure what speculation to belive, this almost seems more credible. (was still hoping for 512 bit, prolly wouldn't make use of the extra bandwidth anyway). Hope this is wrong cause 1.5gb isint enough ram imo to run three decent 120mhz monitors (4x gtx480 always saturated mem in tri screen for me)...


http://vr-zone.com/articles/nvidia-g...ked/10184.html

[kadozer]

Forgive me if i posted something here that has already been posted...

Not sure what speculation to belive, this almost seems more credible. (was still hoping for 512 bit, prolly wouldn't make use of the extra bandwidth anyway). Hope this is wrong cause 1.5gb isint enough ram imo to run three decent 120mhz monitors (4x gtx480 always saturated mem in tri screen for me)...


http://vr-zone.com/articles/nvidia-g...ked/10184.html

Well if you believe this one its more like 10-15% over the 480. If you want to believe the hype which includes the 3dmark score above its more like 20%+ over the GTX 480.

I can't wait to see this magical wand.

[kuroikenshi]

I can't wait to see this magical wand.

magic wand crushed by cayman and antilles

[Sn0wm@n]

nvidia's magic driver .. nvidia's magic silicon .. what magic doesnt nvidia possess allready :S ...

[Mechanical Man]

Well if you believe this one its more like 10-15% over the 480. If you want to believe the hype which includes the 3dmark score above its more like 20%+ over the GTX 480.

I can't wait to see this magical wand.

Game performance != Vantage performance.

So it is possible for it to be 10-15% faster in games and ~20% faster in vantage. Just saying its possible. Not saying it is. I can wait and see.

[Sam_oslo]

And where is the magical 50% coming from from less than a 50% specs increase? Are we looking at an architecture rework here like Cayman?

It's naive to think nVidia is going to release the same thing again. They had good time to work on a "architecture rework", and they have learned a lot from GF104 too. We can expect to see a better performance/power ratio than 480.

GF104 has proven that Fermi-architecture has a good flexibility. If AMD can tweak more out of the same old 40nm, then nVidia should be able too. We have to wait and see how much better they can get, but both GTX 580 and Cayman are going to be dependent on the flexibility of the architecture, because both are based on the same "old" 40nm node.

[JohnZS]

I must admit I am keeping a close eye out on this GTX 580 :P
Finally perhaps a card to replace my Single PCB GTX 295 with.
John

[Carfax]

, if the GTX 580 is as good as rumors indicate, I may have to ditch my 480s

[Flying duck]

It's naive to think nVidia is going to release the same thing again. They had good time to work on a "architecture rework", and they have learned a lot from GF104 too. We can expect to see a better performance/power ratio than 480.

GF104 has proven that Fermi-architecture has a good flexibility. If AMD can tweak more out of the same old 40nm, then nVidia should be able too. We have to wait and see how much better they can get, but both GTX 580 and Cayman are going to be dependent on the flexibility of the architecture, because both are based on the same "old" 40nm node.

^^^
You do realise that GF104 isn't that much better than GF100 in terms of performance per watt and performance per mm^2?