Leaked ATI S.I. 6870 benchmark

[tincart]

A larger GPU size certainly makes sense on the same process node. They have brought out RV770 plus the full R8xx series on it now, so the risk of bringing out a larger chip while keeping margins under control is probably as low as it can be.

 

[GaiaHunter]

wow, 380mm^2 and 35% boost would be great. that sounds more like what NI would have been in the first place at 28 nm however, almost too good to be true.

Some rumors have this as indeed being NI on 40nm. Guess it wont be long until we know.

 

[evolucion8]

Okay I see what you mean. The next page of the preview clears it up by giving the default 480SP score:

default 480SP = X9815

default 512SP = X11371

Therefore gain is 15.85% at default (non-identical) clock speeds

Overclocked 512SP = X12173

and is therefore a gain of 24.02% over default 480SP.

However, the 6870 might score around X12000, so they are still close.

That's far from true, enabling 32 shaders doesn't mean that the score will increase linearly, because there's more than shaders in gaming performance, TMU's, ROP's, cache performance etc, and GF100 is known to have weak TMU performance.

In Vantage, a single GTX 480 scores about X8986 while the HD 5970 scores about X11579, but in games, the difference between both is far more noticeable. So if an HD 6870 scores slightly higher than the HD 5970, means that the GTX 480 512SP is far from being able to even match an HD 5970.

 

[AtenRa]

That's far from true

You really have to read page 8 and 9 of the review, Kalniel is right

http://en.expreview.com/2010/08/09/world-exclusive-review-512sp-geforce-gtx-480/9070.html/8

 

[Davidh373]

Want the release date to get announced so I can put the 480s up on fleabay before these cards come out and devalue them

why do you have 2 480's... that's beyond pointless... unless you do a lot of work with After Effects or Premiere...

 

[Wreckage]

You really have to read page 8 and 9 of the review, Kalniel is right

http://en.expreview.com/2010/08/09/world-exclusive-review-512sp-geforce-gtx-480/9070.html/8

Yeah it looks like a full Fermi would still be ahead of the game.

 

[Idontcare]

Yeah it looks like a full Fermi would still be ahead of the game.

I would expect Fermi drivers, regardless whether applied to 512SP or 480SP enabled parts, to be vastly more mature than the drivers being used to bench pre-release hardware like this.

In fact that is the most plausible reasoning for the existence of this hardware at this stage prior to the release - driver optimization and debugging while production ramp details are getting underway.

To be sure fermi drivers are going to continue to improve between now and the release of SI, but I'd put more money on the SI's drivers improving even more still. Just my opinion, a guess at best.

 

[SHAQ]

Is everyone sure this is a legit benchmark? There are always FUD benchmarks before release.

 

[MrK6]

Yeah it looks like a full Fermi would still be ahead of the game.

A full Fermi probably isn't feasible until 28nm. NVIDIA would have to extensively redesign the core to get something marketable on 40nm (and who knows if it would still have the same performance, look at how much the GTX 460 lacks).

Is everyone sure this is a legit benchmark? There are always FUD benchmarks before release.

I'm not sure if it was in this thread or not, but I remember reading something where Wizzard (the gentleman who writes GPU-Z) said he did not see anything in the screen shot that indicated that this was not legit, at least as far as his own program goes. I'm not saying that verifies that the score is true, but I think it supports it.

 

[Wreckage]

To be sure fermi drivers are going to continue to improve between now and the release of SI, but I'd put more money on the SI's drivers improving even more still. Just my opinion, a guess at best.

That depends. If it's just a "minor improvement" as some rumors have suggested, it's still mostly R600 based and drivers are about as tweaked as they can get. Until they release a true new architecture I don't think they can do much more with the drivers.

40nm (and who knows if it would still have the same performance, look at how much the GTX 460 lacks).

Lacks? You mean like how in newer DX11 games it can keep up with a 5870? My sig can confirm this if you need proof.

 

[Grooveriding]

why do you have 2 480's... that's beyond pointless... unless you do a lot of work with After Effects or Premiere...

I game at 2560x1600.

So bring on these new SI cards, a good 30% or so faster than GTX 480, and if they come without the crazy temperatures and fan noise, I'll be happy to switch.

 

[OCGuy]

why do you have 2 480's... that's beyond pointless... unless you do a lot of work with After Effects or Premiere...

Huh? No it isnt.

 

[MrK6]

That depends. If it's just a "minor improvement" as some rumors have suggested, it's still mostly R600 based and drivers are about as tweaked as they can get. Until they release a true new architecture I don't think they can do much more with the drivers.

People said that about the 5xxx cards and drivers have improved performance 15-20% in many applications: http://forums.anandtech.com/showthread.php?t=2100754 (thanks dug777 for the research!). It remains to be seen what can be done with this new generation, but ruling it out before the hardware has even been released is ridiculous at best.

Lacks? You mean like how in newer DX11 games it can keep up with a 5870? My sig can confirm this if you need proof.

One result from one trial hardly counts as conclusive evidence. You'd be laughed out of any scientific discussion. Let's look at some other trials, shall we?

http://www.hardocp.com/article/2010/07/12/nvidia_geforce_gtx_460_review/5 - Here, in actual gameplay, the GTX 460 can't even keep up with the HD5830 while playing Metro 2033

http://www.hardwarecanucks.com/foru...perclocked-ee-external-exhaust-review-13.html - Once again, the GTX 460 can't keep up with the HD 5830, nevermind it gets completely stomped by the 5850, all at multiple resolutions. What do you think that says about the performance compared to the 5870? And do think that makes the conclusion you're parading in your signature a misconception, purposefully misleading, or completely dishonest? Please answer this as well; don't conveniently ignore the post or just stop posting in this thread, everyone's seen it and will hold you to it.

Furthermore, one could characterize that said one trial as biased since NVIDIA heavily participated in the game's development.

However, I'd like to clarify that my statement more reflected what was lost in the redesign of GF100 to GF104. In order to get power consumption more acceptable and at least closer to the competition, NVIDIA forfeited roughly 15-20% performance going from the GTX 470 to the GTX 460 (reference: http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_460_SLI/25.html ). Now, a heavily overclocked GTX 460 has difficulty keeping up with even a stock GTX 470, and hence why I say it's severely lacking.

 

[Wreckage]

http://www.hardocp.com/article/2010/07/12/nvidia_geforce_gtx_460_review/5 - Here, in actual gameplay, the GTX 460 can't even keep up with the HD5830 while playing Metro 2033

No AA

Here are some more examples.
http://www.xbitlabs.com/articles/video/display/hardware-tesselation.html

Lost planet 2 is another good example here you can see it around 20% faster than a 5850.
http://www.legitreviews.com/article/1391/4/

Also the 460 is not a "full" GF104 so we are not seeing everything that chip can do.

 

[MrK6]

No AA

Here are some more examples.
http://www.xbitlabs.com/articles/video/display/hardware-tesselation.html

According to this link, the 5870 is still much faster than the GTX 460, specifically, the 5870 is 22% faster at 1680x1050, 41% faster at 1920x1200, and 44% faster at 2560x1600. Furthermore, once tesselation is enabled, the 5870 extends its lead at higher resolutions, and is 19% faster at 1680x1050, 49% faster at 1920x1200, and 46% faster at 2560x1600. The 5870 actually trades blows with the GTX 480 when tesselation is enabled. This is ironic, considering tesselation performance was supposedly one of the Fermi architectures major strengths. I'm excited to see what AMD can do with several of its tesselation engines in the upcoming SI cards.

Lost planet 2 is another good example here you can see it around 20% faster than a 5850.
http://www.legitreviews.com/article/1391/4/

Again, this is a benchmark developed by NVIDIA (it's even available for download on NVIDIA's site). If AMD hasn't had a chance to work with it too, is it no surprise that its cards don't perform as well? Is the performance of GF104 that poor that you have to link to biased results to show any advantage? Again, if you want to use biased or immature results as conclusive, it only detracts from your argument as it makes it look like you are either being purposefully misleading or that you lack a general understanding of the scientific process.

Also the 460 is not a "full" GF104 so we are not seeing everything that chip can do.

If the "full" GF104 comes to market, it will be ~10-12% faster clock for clock, in the absolute best case scenarios. The review of the "full" 512SP Fermi showed us that Fermi gets about 80% performance back with increasing shaders. If we get the full 14.2% increase in shaders, net performance will still be below that of the GTX 470. Even if the "GTX 475" gets a clock speed adjustment, it might perform as well as the GTX 470, and its power consumption will probably be close to the same as well. It doesn't seem like a worthwhile endeavor for NVIDIA, and doesn't bring anything new to the table for us or this discussion.

 

[Outrage]

Lost planet 2 is another good example here you can see it around 20% faster than a 5850.

A benchmarking tool posted by nvidia, thats great... what a good example.

 

[cusideabelincoln]

Lacks? You mean like how in newer DX11 games it can keep up with a 5870? My sig can confirm this if you need proof.

HD5870, and even HD5850, is superior, sometime significantly, than a 460.

Summary across all DX 11 titles:
GTX 460: 100%
HD 5850: 109%
HD 5870: 124%

The cheaper HD5830 is 17% faster in minimum framerate in Bad Company 2 than a GTX 460 1GB. INCREDIBLE VALUE!

 

[Wreckage]

. The 5870 actually trades blows with the GTX 480 when tesselation is enabled.

Trades blows, it lost every single benchmark in that review except 1 at 1 resolution. It still lost 2 of the 3 in that game and AA was not enabled. In the other benchmarks it was not even close. The 5870 actually lost to the 460 in Stone Giant, Uniheaven and 2 of the 3 resolutions in Stalker.

Hilarious.






Now what does this have to do with the topic at hand??? Quit derailing threads. And stay out of this one.



esquared
Anandtech Administrator

 

[MrK6]

Trades blows, it lost every single benchmark in that review except 1 at 1 resolution. It still lost 2 of the 3 in that game and AA was not enabled. In the other benchmarks it was not even close. The 5870 actually lost to the 460 in Stone Giant, Uniheaven and 2 of the 3 resolutions in Stalker.

Hilarious.

But I thought one result was good enough to be conclusive for you? I mean, look at your signature. So are you being hypocritical or lying?

 

[Outrage]

In the other benchmarks it was not even close. The 5870 actually lost to the 460 in Stone Giant, Uniheaven and 2 of the 3 resolutions in Stalker.

Hilarious.

Buy a 460 and look att unplayable benchmarks, or buy something else that... you know can actually play games.

 

[Wreckage]

But I thought one result was good enough to be conclusive for you? I mean, look at your signature. So are you being hypocritical or lying?

Yeah I posted from 3 sites and still it was not good enough. So what are you being? :thumbsdown:

I suppose I could add them to my sig if it makes you happy though.

 

[bryanW1995]

But I thought one result was good enough to be conclusive for you? I mean, look at your signature. So are you being hypocritical or lying?

dude, quit arguing with him. he's paid to do it.

 

[MrK6]

Yeah I posted from 3 sites and still it was not good enough. So what are you being? :thumbsdown:

I suppose I could add them to my sig if it makes you happy though.

Well yes it wasn't good enough, because in none of those links did you show any conclusive evidence that what you're holding as a general truth is actually true or universal. My references and links prove that. In fact, I pointed out in every single one of your links the actual or logical flaws in your argument or "evidence." You have yet to show anything to back up your statement. If you cannot, again, are you being hypocritical, misrepresenting, or lying?

dude, quit arguing with him. he's paid to do it.

That's exactly what he wants, so he doesn't have to answer to any of my challenges. About this time he'll either just stop posting, ignore my posts, or still attempt to dig himself out of this one. Do I get points for calling it?

 

[evolucion8]

No AA

Here are some more examples.
http://www.xbitlabs.com/articles/video/display/hardware-tesselation.html

Lost planet 2 is another good example here you can see it around 20% faster than a 5850.
http://www.legitreviews.com/article/1391/4/

Also the 460 is not a "full" GF104 so we are not seeing everything that chip can do.

LOL with the Xbit labs link again, you are the only one here in the world who thinks that because a single GTX 460 can match the HD 5870 in tessellation, the same will happen in every scenario, lol, what a way to lie to yourself. The GTX 470 barely matches the HD 5870, let alone a slower GTX 460 1GB.

http://www.anandtech.com/show/3868/quick-look-powercolors-radeon-hd-5770-pcs-vortex-edition/2 <<Latest review with the latest drivers from both vendors and the review was published on August 26, 2010.

Summary: HD 5870 vs GTX 470

From 21 tests, the HD 5870 won 11 tests, tied 3 and lost 7 compared to the GTX 470

HD 5850 vs GTX 460 1GB

From 21 tests, the HD 5850 won 15 tests, tie 2 and lost 4 compared to the GTX 460 1GB.

Sorry to snap you from your green bubble, back to reality.

 

[zebrax2]

That depends. If it's just a "minor improvement" as some rumors have suggested, it's still mostly R600 based and drivers are about as tweaked as they can get. Until they release a true new architecture I don't think they can do much more with the drivers.

If it does change their current shaders me thinks that is not a "minor improvement".