The GT300/Fermi Thread

[saaya]

SKYML:

great article. Best article out there, starting to like your site over AT. keep up the great work.

yeah his tech reviews are really good, better than AT... not as detailled as xbit, but he cuts to the chase faster and points out what matters faster

so looking at those FC2 numbers...
fermi is 70% faster than a GTX285?
is that a GTX380? 448sps?

All in all, we saw NVIDIA’s 3D Vision Surround in action and while it was extremely impressive to say the least, we can give any more thoughts about it, more testing on our part must be done.

but you saw it with the overhead projects, not the 3 monitors with bezel, right?

hmmm seeing the supersled closeup.... doesnt look that impressive :/
blowing up buildings in FC2 and Crysis looks better imo...
http://www.youtube.com/watch?v=EIB41...layer_embedded

It should also be mentioned that we aren’t providing you with any pictures of the card simply because the final heatsink and PCB layout aren’t finalized yet.

wha wha whaaaa? didnt jensen just tell the BBC in an interview that they started mass production of GF100 cards? How can they mass produce something if the spec isnt finalized?

either way, id have loved to see the prototype card and heatsink, even if its way different from the final product... :/
i mean, you know, if something is a prototype and ES, thats actually MORE interesting to enthusiasts... its a shame nvidia doesnt seem to understand that

thanks for the writeup though, better than the PCP thing imo!

[NCspecV81]

God, why on Earth do you continue posting overclocked results while it has absolutely no relevance?

Here is something from AT, with a 3.33GHz i7:

I posted one.

And when were those benchmarks done exactly? Obviously far from what is current.

[randomizer]

You'd need to upload to photobucket or imageshack etc to use clickable thumbnails. That last pic is ok but the first is stretchin' ma forumz

EDIT: In reference to the last pic before the AT benchies. You people post too fast!

[annihilat0r]

I posted one.

And when were those benchmarks done exactly? Obviously far from what is current.

I can assure you 60 fps is not going up to 70 in two driver updates.

[saaya]

Sorry. A lot of stuff to type.

The benchmark is built into MS's DX11 toolkit for testing of geometry shader code. It is an industry standard with developers as far as I know. I have asked for a bit more information about it but my developer contacts don't think it is publicly available.

Stay tuned for more on that.

hey, no prob, didnt mean to poke at ya, it was meant as a headsup

mmhhh sounds to me like its not really a propper benchmark standard tho... they used MS DX11 toolkit, well yeah, just like EVERY DX11 benchmark and game uses the MS DX!! toolkit... that doesnt mean its an objective unoptimized benchmark... in that case they should have compared fermi to gtx285 like they did in the other tests...

[zerazax]

yeah his tech reviews are really good, better than AT... not as detailled as xbit, but he cuts to the chase faster and points out what matters faster

so looking at those FC2 numbers...
fermi is 70% faster than a GTX285?
is that a GTX380? 448sps?

According to Anandtech, specs and names haven't been finanlized. OTOH, according to HardOCP, GTX 380 will have the full 512SP's...

Then again I also see elsewhere saying that GTX 380 is ~20% faster than a 5870 so who knows

[NCspecV81]

I can assure you 60 fps is not going up to 70 in two driver updates.

Really? So my 5870 defies all logic then?

[saaya]

pcperspective = too much reading :P
TPU = short and too the point
HWC = more infos and details without too many words
AT= too much reading, too theoretical, no benchmark numbers

i dont get why AT didnt post the FC2 and dark void perf numbers
they just say they dont trust those numbers... sure, many people dont trust them, but its still very interesting to know, isnt it?
anandtech didnt even mention how fast nvidia claimed fermi was going to be... i can see that they dont trust them, neither do i, but still, not showing anything is lame..

Power usage (we only know that it’s more than GT200)

remember how nvidia originally claimed GT300 would be about the same as GT200 at GTC?

That leaves us on a final note: clocks. The core clock has been virtually done away with on GF100, as almost every unit now operates at or on a fraction of the shader clock. Only the ROPs and L2 cache operate on a different clock, which is best described as what’s left of the core clock. The shader clock now drives the majority of the chip, including the shaders, the texture units, and the new PolyMorph and Raster Engines. Don’t be surprised if GF100 overclocking is different from GT200 overclocking as a result.

interesting...

[ajaidev]

"esult they did not give a GF100 the ability to drive more than 2 displays at once. The shipping GF100 cards will have the traditional 2 monitor limit, meaning that gamers will need 2 GF100 cards in SLI to drive 3+ monitors"

dam

It was me that simulated the HD 5870 after simulating the GTX 285 results they gave out. Trust me, unless they used a totally different test system for the GTX 285 and GF100, the results are pretty accurate.

SKYMTL pls add 5970 in gaming results!

[jaredpace]

While disappointing that here are no easy to read benchmarks, it's possible to discern general performance from a few key points: The HAWX benchmark in the deep dive slides and the FARCRY2 GF100 vs. GTX285 video. (Check out guru3d's video of the Farcry2 benchmark for extra info). GF100 is ~ 68% faster in FC2, and 61% faster in HAWX, both using 4xAA, compared to a GTX285.

HD5870 is ~35% faster than a GTX285 1920 4xAA, according to Computerbase.de. Also, an HD5970 is ~90% faster than a GTX285. The tested GF100 part sits between a HD5870 & HD5970, closer towards the speed of the 5970. And it appears GF100 handles 8xAA better than HD5000. I rank them as so:

GTX285: 100%
HD5870: 135%
GTX3xx: 168%
HD5970: 190%

[annihilat0r]

Really? So my 5870 defies all logic then?

On a 920@3.6GHz my result is 66fps. Either it's the difference arising from completely different setups (phenom vs i7) or yeah. Yours defies all logic.

[Carfax]

I don't know which article, but it was there.

Phenom is a nice gaming CPU, but the Core i7 is undisputably faster, so I don't see how AT could make such a claim.

And there are also graphs available on their website which contradict the statement.

[saaya]

"esult they did not give a GF100 the ability to drive more than 2 displays at once. The shipping GF100 cards will have the traditional 2 monitor limit, meaning that gamers will need 2 GF100 cards in SLI to drive 3+ monitors"

dam

oh come on... whos really gonna game on 3 monitors... :P

SKYMTL pls add 5970 in gaming results!

oh yeah, that would be cool!
oh and how about HAWX, did they bench HAWX or did nvidia only offer slides for the HAWX perf?

While disappointing that here are no easy to read benchmarks, it's possible to discern general performance from a few key points: The HAWX benchmark in the deep dive slides and the FARCRY2 GF100 vs. GTX285 video. (Check out guru3d's video of the Farcry2 benchmark for extra info). GF100 is ~ 68% faster in FC2, and 61% faster in HAWX, both using 4xAA, compared to a GTX285.

HD5870 is ~35% faster than a GTX285 1920 4xAA, according to Computerbase.de. Also, an HD5970 is ~90% faster than a GTX285. The tested GF100 part sits between a HD5870 & HD5970, closer towards the speed of the 5970. And it appears GF100 handles 8xAA better than HD5000. I rank them as so:

GTX285: 100%
HD5870: 135%
GTX3xx: 168%
HD5970: 190%

mhhhh yeah, looks interesting...

[annihilat0r]

Phenom is a nice gaming CPU, but the Core i7 is undisputably faster, so I don't see how AT could make such a claim.

And there are also graphs available on their website which contradict the statement.

I found what I was talking about:

"The situation is this: in some cases, Nehalem can go from being much faster than Phenom II, to being measurably slower within the same benchmark depending on resolution. Gary was the first to tie the issue to the GPU used. Gary found that NVIDIA GPUs appeared to behave this way on Nehalem/Phenom II while AMD GPUs didn't. In other words, NVIDIA GPUs were running faster on AMD hardware while AMD GPUs were running faster on Intel hardware. It's all very strange."

http://www.anandtech.com/cpuchipsets...spx?i=3640&p=3

[STEvil]

oh come on... whos really gonna game on 3 monitors... :P

Me.

[ajaidev]

Ok read through several sites, i am sad that GF100 =GTX380 and not the GTX360 that means the farcry 2 bench was done by Nvidias best and most likely the GF104 may be the lesser card.

The other thing i am confused about is the tessellation implementation on this thing, logic would dictate that the huge scaling is due to the use of CUDA cores as geometry processing units which are in parallel config. Now i read that use of tessellation does not give a big hit to the performance and i am confused as to how that would be achieved?

Use of physx and tessellation together should alter the score quite a bit because the no of CUDA cores will be decreased...



Also lol at the sites since they are comparing 5870 to GF100 while 5970 should be compared given the 8+6pin config and the segment each card is suppose to represent "GF100 and 5970 are ultra high end" if price was disclosed it would have been much simpler tough. Where is the cheapo GTX 360 BTW anyone know?

[NCspecV81]

I found what I was talking about:

"The situation is this: in some cases, Nehalem can go from being much faster than Phenom II, to being measurably slower within the same benchmark depending on resolution. Gary was the first to tie the issue to the GPU used. Gary found that NVIDIA GPUs appeared to behave this way on Nehalem/Phenom II while AMD GPUs didn't. In other words, NVIDIA GPUs were running faster on AMD hardware while AMD GPUs were running faster on Intel hardware. It's all very strange."

http://www.anandtech.com/cpuchipsets...spx?i=3640&p=3

And that proves? Absolutely nothing. Why? Even if you consider that proof there is nothing said there that proves anything with what we know about fermi. I mean, even they say "in some cases", and just perhaps they just fubard their test results?

[god_43]

http://www.pcper.com/article.php?aid=858

[annihilat0r]

That proves nothing but those oddities might explain why your lower clocked Phenom 2 is doing better than our higher clocked i7's with the same graphics card.

[Solus Corvus]

AT gets 14fps more on a 5870 @ 2560x1600 in FC2. With SKYMTL's numbers at 2560x1600 fermi has a 39% lead over 5870. Sub AT's numbers for 5870 and it's a 6% lead.

What a mess. I can't wait to see this in an independent reviewer's hands so we can get real apples-to-apples tests in a variety of games. I'm curious to see some 8xAA numbers as well.

[~CS~]

I found what I was talking about:

"The situation is this: in some cases, Nehalem can go from being much faster than Phenom II, to being measurably slower within the same benchmark depending on resolution. Gary was the first to tie the issue to the GPU used. Gary found that NVIDIA GPUs appeared to behave this way on Nehalem/Phenom II while AMD GPUs didn't. In other words, NVIDIA GPUs were running faster on AMD hardware while AMD GPUs were running faster on Intel hardware. It's all very strange."

http://www.anandtech.com/cpuchipsets...spx?i=3640&p=3

AMD GPUs were running faster on Intel hardware.

You know you are not proving your point with this right , quite opposite actually .Instead of bashing each other results , post more information Driver release and so on ... Am sorry whole simulated , doesn't really interest me , it doesnt mean fermi isn't a good performer ,but am sure it is to benefit of some people ,especially at release, to make numbers look more groundbreaking rather then marginal ...

Also am sure fermi has a way to go with drivers as well.

[NCspecV81]

You know you are not proving your point with this right , quite opposite actually .Instead of bashing each other results , post more information Driver release and so on ... Am sorry whole simulated , doesn't really interest me , it doesnt mean fermi isn't a good performer ,but am sure it is to benefit of some people ,especially at release, to make numbers look more groundbreaking rather then marginal ...

Also am sure fermi has a way to go with drivers as well.

9.12 officials. But yeah you are right. Simulated numbers mean squat. I'm actually surprised those were included. In fact, they should probably be removed.

[BrowncoatGR]

TPU = short and too the point

Cant find it on their webpage

[kadozer]

So all signs so far point to being faster than a 5870 by 20-30% and tad slower than a 5970. Looks like won't be that big of a price war for us consumers then.

[ajaidev]

Now lets see how much guts does each have "5970 vs GF100":


GF100 GPU features 512 CUDA cores, 16 geometry units, 4 raster units, 64 texture units, 48 ROPs, and a 384-bit GDDR5 memory interface.

The 5970 has * 3200 Stream Processing Units * 160 Texture Units * 256 Z/Stencil ROP Units * 64 Color ROP Units and a 256-bit*2 GDDR5 memory interface.

GF100 does all this with 3 billion 40nm transistors and 5970 uses 4.3 billion, i highly doubt that a GF100 can take on a 5970 in most games.


In performance a picture like this may emerge - 5850<5870<GTX380<5970...