Articles - So just how fast is NVIDIA's Fermi graphics card? We find out.

Quote:

---

Benchmarked against a GeForce GTX 285 and Radeon HD 5870.

---

Read more.

Do us a favour and bench Farcry 2 on a 5970, same settings of course. :)

Thing is the 5970 is really a 5870x2 (AMD just changed the branding scheme), it's two cores (which gives wonky performance on games not built to deal with multiple GPUs like a lot of games that aren't brandspanking new - which is why I went for 5870 myself [being delivered tomorrow]), so you could only really fairly compare it to two Fermi cards in SLI.

However there is absolutely no reason why Dark Void should have not been tested on a 5870 too otherwise the second review page is largely pointless (comparing it to old hardware when people that wanted high performance already switched to 5870), though again looking at reviews it seems to be a game that is also biased in design for nVidia cards like far cry 2, so probably not a good idea to trust nvidias tests at all til independent ones are carried out on games that they haven't anti-competitively bribed developers to exclude AMD from ("The Way It's Meant To Be Played" scheme).

I don't care about brands, just what's best within reason (I was going to buy gtx285 before 5870 appeared). Far Cry 2 and Dark Void are on all accounts crap games (PC Gamer magazine did great reviews as well as zero punctuation, dark void is this month in PCG and apparently the flying is a gimmick and you only get to use it at the very start and very end of the game, for the rest it's a substandard corridor shooter), I don't care how well they run as an nvidia-sponsored benchmark program. ;)

I wonder how much "tweaking" was done for that presentation... It's rather sad, but one can never believe either companies claims on performance.

Quote:

---

Originally Posted by Perfectionist

Thing is the 5970 is really a 5870x2 (AMD just changed the branding scheme), it's two cores (which gives wonky performance on games not built to deal with multiple GPUs like a lot of games that aren't brandspanking new - which is why I went for 5870 myself [being delivered tomorrow]), so you could only really fairly compare it to two Fermi cards in SLI.

However there is absolutely no reason why Dark Void should have not been tested on a 5870 too otherwise the second review page is largely pointless (comparing it to old hardware when people that wanted high performance already switched to 5870), though again looking at reviews it seems to be a game that is also biased in design for nVidia cards like far cry 2, so probably not a good idea to trust nvidias tests at all til independent ones are carried out on games that they haven't anti-competitively bribed developers to exclude AMD from ("The Way It's Meant To Be Played" scheme).

I don't care about brands, just what's best within reason. Far Cry 2 and Dark Void are on all accounts crap games (PC Gamer magazine did great reviews), I don't care how well they run as an nvidia-sponsored benchmark ;)

---

Yep but the talk is that the g100 should go up against the 5970. They both use 8-pin and 6-pin connectors putting g100 above 225w - some claims of 280w are being made.

http://vr-zone.com/articles/-rumour-...ble-/8314.html

This g100 is a lot closer to a 5970 than it is to a 5870, and that is also what it has to beat.

Also the 5970 is not 2x 5870 is it? Its a mixture of 5850 and 5870 joined IIRC (please correct me if im wrong). The card nvidia used is probably going to be their range toping card, it has a double die size of the ati equivilant along with a much more expensive design process and other hardware that costs a bomb in comparison! That means in price wise, this is probably the equivilant to a 5970.

I could be wrong and hope i am as competition would be great, however this is how i see it!.

It is funny that Nvidia did not try DiRT 2 or BattleForge as these both support DX11!

Quote:

---

Originally Posted by Jimbo75

Do us a favour and bench Farcry 2 on a 5970, same settings of course. :)

---

Yup, will do, just for you :)

Perhaps the article didn't make it clear. The NVIDIA numbers were produced by NVIDIA during an event in Las Vegas. We did not run them.

The AMD/ATI numbers for Far Cry 2 were run in our labs in the UK, to provide some basis for comparison, however tenuous given that we don't know the exact spec. of the NVIDIA machines.

We will add the HD 5870 numbers for Dark Void as soon as we have the game in our hands and can confirm the exact settings used by NVIDIA.

This is not intended as a review. Rather, it's a very brief look at what kind of performance to expect from Fermi silicon, based on numbers specifically generated by NVIDIA.

We do not have the GF100 card in our labs and don't expect to do so for a month.

Sorry for any confusion caused.

You may have used a 975 extreme edition, a decent CPU no doubt, but are you sure Nvidia weren't using one clocked to 4.5-5Ghz, thereby largely invalidating your results?

If you ran the same test with a 2Ghz dual core, the results would have also been different, making it problematic.

THen the last problem, Fermi, you don't know which "Fermi" was in the machine. Its all well and good being 67 % faster than a single 285GTX, which lets be honest, isn't fantastically brilliant in and of itself, in its best game scaling. But what if its true that this is one of a very few cores that is 512sp's and 1.6Ghz shaders, but if the only cards you can ever buy in stores are 448sp's and 1.2Ghz, it could be anything from 30-40% slower, if thats the case a 380gtx with lower clocks and only 448SP's might only be 25-35% faster than a 285GTX in one of its best games, and lets be clear, if its one of only 2 real games on show and the only non physx game on show, you can be 100% certain this is the game it scales best in.

Its all well and good that they might release a 390GTX with 512sp's and 1.6Ghz clocks, but if no one can physically buy one, its a joke really.

I'm happy to be proved wrong, I'd be exstatic to buy a £300 512sp 1.6Ghx 380GTX that spanks a 5870 by 40%, but lets get a show of hands of who believes that will happen.

AS for physx, well, Dark Void(having played the demo and finding it largely like other "physx" titles, all show, no substance) is likely faster due to a significant boost in double precision power on top of the better functionality of switching a cluster out between different threads(gpu to gpgpu switching). Though in a fairly basic looking game thats likely not actually pushed very hard in terms of GPU processing, it makes more sense to dedicate a cluster to the overused phsyx effects than to constantly switch them back and forth, so I'd think its increased DP power available and better physx efficiency, than being down to faster switching between threads.

The single problem being, its yet another fairly poor game in terms of gameplay and feel, loaded up with physx effects. Seems constantly unsurprising that companies that spend all their time on adding physx effects seem to spend less time on everything else.

But again the main issue is simply assuming anything you can buy in the real world will match Nvidia's results in any way at all. It would be like if the entire 5870 release was based on a 5890 card, with 25% better performance, that you simply can't buy.

Basically the launch gave us the MAXIMUM ARCHITECTURE specs, nothing more, or less, no commitment to actually make a card with those full specs available, which everyone all over the net is hearing, simply won't happen. Outside of a handful being sent out just so someone can officially claim it was available so Nvidia can say "ask Newegg, they sold..... 8, in the past 6 months, of course its a real card" its a very shakey ground you're standing on by comparing performance when you quite literally have no idea what was in that computer and if it will ever be available.

Good stuff Tarinder. We know it's going to be a while before we get the real fermi but any little ideas we can get on comparitive performance will keep us going. :)

@ Drunken - we know that it's all up in the air regarding the cpu and clocks, but any little bits of info we can get is better than nothing at all.

Actually the test rig was a 3.2GHZ Core i7 960:

http://www.hardwarecanucks.com/forum...oscope-13.html

Thing is, nVidia have got no impetus to beat ATI at the price/performance level. So the idea that they'll release a GF100 at the same performance level as the 5970 and undercut it is a complete nonstarter.

They'll harp on about physx and CUDA, and almost certainly charge you through the nose for the experience. On top of that, the fact that their 2xx range is still so hideously expensive and yet has people clambering to buy them proves that there is still a massive market of nVidia fanboys who wouldn't touch an ATI with bargepole. Hence my complete lack of interest in Fermi thus far.

Quote:

---

Originally Posted by CAT-THE-FIFTH

Actually the test rig was a 3.2GHZ Core i7 960:

http://www.hardwarecanucks.com/forum...oscope-13.html

---

Yep good catch. The i7 975 should be the same with turbo off I think, but 133mhz isnt going to make much difference to the result anyway.

Just run the Far cry 2 Benchmark on my PC (specs Below) and get the following results at the same setting as Fermi. I'm happy with what I have!

Average 112.03
Max 152.49
Min 76.37

Spec's

Core i7 920 @ 3.2Ghz
6Gb Ram
2 x HD5870 @ Stock Speeds - CrossfireX

Ross

Thats pretty interesting. I thought the 5970 was over 100fps average at those settings and looking at your results it should be true.

If this is Fermi at its best, it's not looking good for Nvidia. On the other hand, look at those minimums - 64fps is quite impressive for a single gpu. Interesting times ahead. ;)

Quote:

---

Originally Posted by Jimbo75

Yep good catch. The i7 975 should be the same with turbo off I think, but 133mhz isnt going to make much difference to the result anyway.

---

Considering that a Core i7 at 2.6GHZ does not bottleneck an HD5970 here should be no issues with a CPU bottleneck:

http://www.legionhardware.com/document.php?id=869

The fact that Nvidia are using an i7 960 in their official test rig also indicates this!