The GT300/Fermi Thread - Part 2!

[zalbard]

The GTX480 with 512 shaders running at full speed, 600 or 625MHz depending on which source, ran on average 5% faster than a Cypress/HD5870, plus or minus a little bit. The sources were not allowed to test a GTX470 which is likely an admission that it will be slower than Cypress/5870.

There is one bright spot, and it is a very bright spot indeed. No, not the thermal cap of the chip, but the tessellation performance in Heaven. On that synthetic benchmark, the numbers were more than twice as fast as the HD5870/Cypress, and will likely beat a dual chip HD5970/Hemlock. The sources said that this lead was most definitely not reflected in any game or test they ran, it was only in tessellation limited situations where the shaders don't need to be used for 'real work'.

Well, this better not be true. I mean, great performance in 'over-the-top-tessellation-using-games' is nice, but when are these coming out? In 2012? Waiting till Monday. But it would be a shame since these cards wouldn't compete with Cypress well just because of the price...

[Sly Fox]

I just read up to, "Small numbers of final cards have started to trickle in to Nvidia, and they are only showing it to people they consider very friendly for reasons that we will make clear in a bit."

And I'm already ROFL'ing here.

Oh man, Nvidia had better pray that these guys are pulling all of this out of there arse's.

Not that I expect much after the whole delaying for all of eternity and then holding up a blatantly fake card in order to lie to investors.

It's just sad if this is true, if only for the fact that I wanted a monster card. Crysis with 16X AA @ 60+ FPS on a single card would probably look so nice.

[Jowy Atreides]

ran on average 5% faster than a Cypress/HD5870....

hooo boy. The 5870's will overclock well beyond 480 then

[-Sweeper_]

Now you know the raw clocks, how does it perform? It is a mixed bag, but basically it is much below their original expectations publicly stated as 60% faster than Cypress. The numbers that SemiAccurate were told span a variety of current games, all running at very high resolutions. Here is where we can't list specifics or the Nvidia Keystone Kops might find their first SemiAccurate mole. We will bring you the full spreadsheets when the cards are more widespread.

The GTX480 with 512 shaders running at full speed, 600 or 625MHz depending on which source, ran on average 5% faster than a Cypress/HD5870, plus or minus a little bit. The sources were not allowed to test a GTX470 which is likely an admission that it will be slower than Cypress/5870.

Highly doubt things are that bad.

[Tim]

Oh snap if true......

[mapel110]

Even the clocks are wrong. There are 4 Clockdomains. Chip/half hot clock/hot clock/vram

[natty]

yea we believe you buddy...... lol

[Nedjo]

<CONSPIRACY_THEORY>there's always possibility that NVIDIA is feeding CD with false info to make him ridiculous and irrelevant news source after final launch?</CONSPIRACY_THEORY>

[Sly Fox]

<CONSPIRACY_THEORY>there's always possibility that NVIDIA is feeding CD with false info to make him ridiculous and irrelevant news source after final launch?</CONSPIRACY_THEORY>

Haha it would certainly be an amusing turn of events.

No matter what I do though I just can't get the feeling out of my gut that fermi is the GeForce FX series all over again.

[***Deimos***]

"SemiAccurate gets some GTX480 scores"
http://www.semiaccurate.com/2010/02/...gtx480-scores/

EDIT: Whoops, already posted, shoulda refreshed the page

Oh wow. SNAP.

I can kind of see how it makes sense though.
Its like nVidia fx vs 9700pro - tied in performance.. in DX8 games.

Vertex processing power in current GPUs has increased dramatically over last 10 years... but all games made nowadays are pretty much console ports. And XBOX360 and PS3 use GF7/X1900 type graphics...

Even "high end" GPUs selling today like GT240 and G92 score just "adequately" in Ozone3D 18million triangle, geometry instancing demo.

18 million triangles
8800GTX - 13fps
HD3870 - 25fps

8 million triangles
8800GTX - 28 fps
HD3870 - 33 fps

Perhaps this is why games keep the poly count down, and how/why Fermi is so much faster with vertex processing... GF8 was just slow at it to begin with.

[Chumbucket843]

Both of our sources said that their cards were smoking hot, one said they measured it at 70C at idle on the 2D clock.

i think they are measuring in kelvin. i dont see how idle power could be worse by this much when it consumes in worst case scenario 50% more power at peak.

<CONSPIRACY_THEORY>there's always possibility that NVIDIA is feeding CD with false info to make him ridiculous and irrelevant news source after final launch?</CONSPIRACY_THEORY>

maybe this is why charlie writes libelous articles about nvidia.

[spursindonesia]

With this article, CD puts A LOT of his credibility, well, basically all of it, on jeopardy. Not just to Green camp fanbois, or the general mass, but even to people who has been skeptic toward nVidia's Fermi, like me. In my skepticism, i still see 512 SP's Fermi to be around 15-20 % faster overall than plain jane HD 5870, so this claim is just rather unbelievable to me at this moment.

[weston]

he has been right about a lot of things so far, EOL of GTX 200 series before fermi, no high end offerings, not able to drop prices to compete with the 48x0 series, late in production, problems with yields, problems with heat, etc...

Not an ATI fanboy here, I have a GTX 275, but fermi looks to be full of fail for gamers. the GPGPU will subsidize any losses in the gaming segment though.

[Chumbucket843]

Oh wow. SNAP.

I can kind of see how it makes sense though.
Its like nVidia fx vs 9700pro - tied in performance.. in DX8 games.

Vertex processing power in current GPUs has increased dramatically over last 10 years... but all games made nowadays are pretty much console ports. And XBOX360 and PS3 use GF7/X1900 type graphics...

Even "high end" GPUs selling today like GT240 and G92 score just "adequately" in Ozone3D 18million triangle, geometry instancing demo.

18 million triangles
8800GTX - 13fps
HD3870 - 25fps

8 million triangles
8800GTX - 28 fps
HD3870 - 33 fps

Perhaps this is why games keep the poly count down, and how/why Fermi is so much faster with vertex processing... GF8 was just slow at it to begin with.

vertex processing is cheap and is bound by vram and tri-setup. the 3870 is faster because the core clock is higher. most games have ~1million polys/frame so its not that important. triangle setup/rasterization/vertex culling is much faster on fermi than any gpu. if clocks on fermi are 600MHz then setup rate is 2.4 billion tris/sec. 5870 is 850 million tris/sec. this boost was added for tessellation. most triangles today cover 30 pixels most of the time. with tessellation micropolys can cover half a pixel.

Well, this better not be true. I mean, great performance in 'over-the-top-tessellation-using-games' is nice, but when are these coming out? In 2012? Waiting till Monday. But it would be a shame since these cards wouldn't compete with Cypress well just because of the price...

it doesnt take time to get good at adding a lot of tessellation which is one of its advantages. all game devs have to do is remodel characters and there are already software tools for that. if you have a displacement map that has very small intricate detail it will require higher tessellation levels to be visible.

[damha]

So now nvidia realizes tessellation is a good idea when ATI has been saying this for the past, what, 10 years?!

[cegras]

I don't know, Charlie hates a lot, but his GTS 250 hoopla back at the inquirer from a while ago really was an 'aha' moment.

[cegras]

i think they are measuring in kelvin. i dont see how idle power could be worse by this much when it consumes in worst case scenario 50% more power at peak.


maybe this is why charlie writes libelous articles about nvidia.

People around here don't seem to be able to differentiate between power and temperature. An idle temperature of 70 may not mean much, since the 4850 idled really hot too. If this is all true, likely this is done to keep acoustics in check, with the screaming coming during load.

[grimREEFER]

there's a good chance that "most games" means cpu limited games
it's sorta a convenient way to ignore crysis and the like.

[Iconyu]

there's a good chance that "most games" means cpu limited games
it's sorta a convenient way to ignore crysis and the like.

That would be a creative use of the word most, especially considering most games are not CPU limited.

[Marios]

This is what I would like to see ...

GTX480 Ultra 245W

GTX285	HD5870	GΤΧ480	%Faster	Game
65,7	83,8	109,72	30,93	Far Cry 2
115	111,2	192,05	72,71	Batman Arkham Asylum
85,5	88,4	142,79	61,52	Tom Clancy's H.A.W.X
88,5	89,3	147,8	65,5	Resident Evil 5
60,3	65,4	100,7	53,98	World in Conflict 
6376	7953	10649	33,89	3DMark Vantage Extreme GPU

[kemo]

People around here don't seem to be able to differentiate between power and temperature. An idle temperature of 70 may not mean much, since the 4850 idled really hot too. If this is all true, likely this is done to keep acoustics in check, with the screaming coming during load.

But when you read ideling @ 70C with Fan at 70% makes you think that his source is smoking something or sitting in a 50C ambient

[shoopdawoopa]

This is what I would like to see ...

GTX480 Ultra 245W

GTX285	HD5870	GΤΧ480	%Faster	Game
65,7	83,8	109,72	30,93	Far Cry 2
115	111,2	192,05	72,71	Batman Arkham Asylum
85,5	88,4	142,79	61,52	Tom Clancy's H.A.W.X
88,5	89,3	147,8	65,5	Resident Evil 5
60,3	65,4	100,7	53,98	World in Conflict 
6376	7953	10649	33,89	3DMark Vantage Extreme GPU

I pray this is the real result, I want Fermi to be good so bad so we can get rid of this onesided nonsense.

[saaya]

I'm not so sure about that.

There are multiple constraints, and bandwidth is just one. #GPUs has to be >= #frames being rendered ahead. Hypothetically, if you had 100 GPUs running at 50fps, those 100 frames would be 2 sec of lag. And D3D and OpenGL have limitations on #chain/buffers.

There is also diminishing returns due to cost of partitioning work, and how uniform the work units are. Most games I play dont have exact same pixel/texture/vertex workload on all parts of the screen all the time.

Bottom line:
If you cant get anywhere near 50% let alone 100% scaling in many (not all) games with just 2 GPU, how do expect it to be possible for quad?

we were talking about gpgpu... the limitations for graphics are different... especially for sfr the system bus definately plays a big role, but for gpgpu each gpu works on independent tasks (i thought?)

I don't know, Charlie hates a lot, but his GTS 250 hoopla back at the inquirer from a while ago really was an 'aha' moment.

hmmm i must have missed that... what was it about?

This is what I would like to see ...

GTX480 Ultra 245W

GTX285	HD5870	GΤΧ480	%Faster	Game
65,7	83,8	109,72	30,93	Far Cry 2
115	111,2	192,05	72,71	Batman Arkham Asylum
85,5	88,4	142,79	61,52	Tom Clancy's H.A.W.X
88,5	89,3	147,8	65,5	Resident Evil 5
60,3	65,4	100,7	53,98	World in Conflict 
6376	7953	10649	33,89	3DMark Vantage Extreme GPU

why is a 285 as fast as a 5870 in this table? 0_o

[Iconyu]

hmmm i must have missed that... what was it about?


why is a 285 as fast as a 5870 in this table? 0_o

When nvidia renamed the 9800GTX to the 250 GTS, but didn't send parts for preview/review to anandtech or hardocp as they knew they'd say something bad about it.

And the 285 is that speed because the table has been pulled directly out of the posters arse.

[96redformula]

When nvidia renamed the 9800GTX to the 250 GTS, but didn't send parts for preview/review to anandtech or hardocp as they knew they'd say something bad about it.

And the 285 is that speed because the table has been pulled directly out of the posters arse.

Ah, I didn't realize that is what all the fuss was about...Thanks for filling me in .