GPU Showdown: Fermi Vs. Cypress - Both Overclocked on Air

[RussianSensation]

Main Benchmarks
http://www.overclock.net/overclock-...prehensive-gpu-showdown-fermi-vs-cypress.html

Update #1 (Page 12): GTX480 on water vs. 5970 on air
http://www.overclock.net/overclock-...hensive-gpu-showdown-fermi-vs-cypress-12.html

Update #2 (Page 17): MSI GTX 460 Cyclone 1GB @ 940 / 2300 memory
http://www.overclock.net/overclock-...hensive-gpu-showdown-fermi-vs-cypress-17.html

Videocards Compared:

5770CF - 1055/1350
5850 - 1115/1200
5870 - 1115/1200
5970 - 1000/1200
465 - 900/2100
470 - 875/2100
480 - 900/2300

* Note the above overclocks are suicidal for 24/7 on air operation!

System Setup:

Intel i7 980x (4.61GHz HT Off)
Corsair Dominator GTX2's
EVGA 4-Way Classified X58
Enermax EVO 1250W
Separated OS Installs to factor out driver conflicts
Samsung 2493HM - its a 1920x1200 Capable but all tests were run at 1080p

List of Benchmarks and Settings:

3dmark03 - default
3dmark05 - default
3dmark06 - default
3dmark Vantage - default no physx
AVP Benchmark - dx 11 4xAA all max
Heaven 2.0 - Default settings @ 1920x1080 with normal tessellation
Crysis - very high dx10 4xAA
Farcry2 - dx10 ultra high / max 4xAA
Dirt 2 - dx11 max and 4xAA
Stone Giant - default settings at 1920x1080 high tessellation
Resident Evil 5 - Max settings 8xAA dx10 fixed benchmark
Hawx - dx10.1 all max in game settings 8xAA
StreetFighter4 - All max settings 8xAA
Counter-strike: Source - All max settings including bloom 8xAA and max_fps 999
Just Cause 2 - All max settings minus the nvidia only functions 8xAA
Left4Dead2 - 1920x1200 all max in game and 8xAA
Metro2033 - dx11 all max in game settings minus DoF and Advanced PhysX 4xAA
Battlefield Bad Company 2 - dx11 all max 8xAA

Enjoy

 

[tviceman]

Wow the gtx470 beat the 5870 in nearly every benchmark.

 

[KIAman]

I am truly amazed how close the HD5850 and HD5870 overclocked. Something looks wrong software wise considering the architectures are identical except GPU horsepower.

 

[zagood]

I am truly amazed how close the HD5850 and HD5870 overclocked. Something looks wrong software wise considering the architectures are identical except GPU horsepower.

"The reference 5850 I had did not touch the same clocks as the 5870 lightning therefore I decided to experiment a little. I took a 5850 bios and flashed it to my 5870 lightning and everything worked to plan. I got the same clocks as the 5870 but with the 5850 shader count."

 

[OILFIELDTRASH]

Fermi really starting to shine. 470 beating 5870 consistently and the Gtx 480 looking like an absolute beast.

 

[Hauk]

Man that guy worked his arse off, fresh OS installs even. I love seeing this kind of stuff.. good find OP..

 

[lavaheadache]

i don't like that all the benchies are overclocked gpu's

i think a stock run is more important

 

[busydude]

i don't like that all the benchies are overclocked gpu's i think a stock run is more important

There are countless websites which show performance at stock run. I guess the article is first of its kind for this generation.

 

[aggressor]

I agree. There are plenty of other sources for stock clocks. With these things overclocking so well, it's great to have this as a reference point.

I do wish he had a GTX 460, though. That's the real gem.

 

[thilanliyan]

Great find. I was itching to see how the cards compare when OCed.

 

[KIAman]

i don't like that all the benchies are overclocked gpu's

i think a stock run is more important

I like the OC'd results. It shows maximum potential rather than targeted performance.

 

[TheDrake]

I agree. There are plenty of other sources for stock clocks. With these things overclocking so well, it's great to have this as a reference point.

I do wish he had a GTX 460, though. That's the real gem.

He has a graph with the 460 on page 17 IIRC.

Love the OC results too, motivates me more to grab the 470 from TG for $230.

 

[tviceman]

Fermi really starting to shine. 470 beating 5870 consistently and the Gtx 480 looking like an absolute beast.

By the looks of this graph, I think this shows how ATI and Nvidia strategies are going to converge with their refresh lineups (6x00 series and Fermi II). Both have inherently different problems, while the strength of one lineup is the weakness of the other's.

ATI absolutely needs to improve performance, while somehow maintaining manageable die sizes, power consumption, and good thermals.

Nvidia absolutely needs to improve yields, power consumption, and thermals without sacrificing performance.

I think the price wars everyone has been wanting and waiting for will finally happen this winter when the refresh parts go head to head.

 

[Daedalus685]

The 5800 clock scaling is terrible compared to gf100.

Nice info!

 

[Soleron]

@tviceman

Have you heard about a Fermi refresh this year? Link? We've heard a little about Southern Islands but a Fermi II was just speculation as far as I know - and Nvidia is known to leak more than AMD.

I mean, GF106 and GF108 launch around August; rollouts like that don't usually happen close to a new gen.

 

[Daedalus685]

@tviceman

Have you heard about a Fermi refresh this year? Link? We've heard a little about Southern Islands but a Fermi II was just speculation as far as I know - and Nvidia is known to leak more than AMD.

I mean, GF106 and GF108 launch around August; rollouts like that don't usually happen close to a new gen.

There are murmurs of a gf104 with full shader count, even a new card with gf100 counts based on gf104. No one knows for sure but at least a GTX'475' based on a full gf104 is likely. With 28nm not in the works until a year from now or so they will likely have to come up with something else on 40nm.

I wonder what technical aspects of cypress limit its clock scaling so much. The % gain on the 470 (note that it is actually an unlocked 465 which could make a difference in OC) is certainly larger than the 5870 but it blazes past it in some tests it clearly loses at stock. Hmm... I'll ahve to look loser to see if the poor showing from cypress is simply lower OC values or sub linear scaling.

 

[Genx87]

I suspect the refresh of Fermi will be something based around the GF104. I was really surprised at how they managed to tackle the performance\watt with a significant degree with the GF104.

 

[nyker96]

i don't like that all the benchies are overclocked gpu's

i think a stock run is more important

I have to agree here. Since OC sometimes depend on the card you get or even the particular card's ability to change core-voltage, cooling system its used, this might not be reproducible across setups. On another setup maybe you get a nice 5970 with great cooling system and it shows up much better than here etc. you might get some trend off these data but have to take it with a grain of salt.

But even knowing that, it's great work and hard work OP for going through so many benchmarks, this must be commended. great job!!

 

[MrK6]

Probably one of the best articles that have been put out recently. That really shows what the capability of the hardware is, not the same generic stuff that every review site produces. It would have been nice if he had tested with higher resolutions, but good stuff so far.

 

[lifeblood]

I have to agree here. Since OC sometimes depend on the card you get or even the particular card's ability to change core-voltage, cooling system its used, this might not be reproducible across setups. On another setup maybe you get a nice 5970 with great cooling system and it shows up much better than here etc. you might get some trend off these data but have to take it with a grain of salt.

I agree as well. I also would point out that it doesn't even give an idea of potential. To quote the OP:

* Note the above overclocks are suicidal for 24/7 on air operation!

So really unless your planning on replacing your vid card after a marithon 24 hour gaming session, this doesn't have much relevance to the real world.

Still, it is interesting and gives an idea of the benefits of the different architectures. Pity TSMC couldn't get 32nm going in time for this generation. I bet things would have been dramatically different with nVidia dominating this generation.

 

[tviceman]

@tviceman

Have you heard about a Fermi refresh this year? Link? We've heard a little about Southern Islands but a Fermi II was just speculation as far as I know - and Nvidia is known to leak more than AMD.

I mean, GF106 and GF108 launch around August; rollouts like that don't usually happen close to a new gen.

No I don't have any evidence. I am using past launches, market conditions, and general timetables for my reasoning.

We know that Fermi yields aren't great and on top of that the chip is really big so it costs a whole lot more to manufacture than Cypress. At the same time, Fermi prices have dropped considerably in a short amount of time since release. So Nvidia is likely making very little to no money at all on their desktop GF100 based cards. We also know that thermals are very high which require (what many people consider) obtrusive cooling solutions to manage. We also know that draw power is very high which somewhat limits their consumer market and also significantly cuts back into professional/ HPC sales.

Fermi was 6 months late, which obviously had a trickle down affect on Fermi derivatives (GF104, 106, and 108). But, at the same time, 6 months late now does not mean future refreshes and/or generations will be behind schedule. Nvidia refreshed the Geforcefx 5800 ultra in 5 months with a vastly improved Geforcefx 5900 ultra. GT200 came out in June of '08, and was refreshed in January of '09. That took 7 months, and the architecture didn't scream to be refreshed and improved as much as Fermi does. And also, sense the general consensus seems to be that 28nm is the next GPU node to be used and won't be available until the second or third half of 2011 - which makes sense for a refresh to come sometime this fall or winter of this year.

We could just rely on Charlie 100% of the time for all our Nvidia related info, but lets be honest - Charlie doesn't report good news for Nvidia. And even when good things happen, if he has to talk about it he does his best to spin the info at hand. He said only 10,000 GF100's would be made - but does anyone really believe that now? And he said weeks before GF104 came out that it would be just as bad as GF100 in nearly every aspect, but we all know that couldn't have been further from the truth. GF104, GF106, and GF108 weren't set in stone last year during the muck and mire of Fermi's development.

It's obvious Nvidia has applied some of the lessons they learned to GF104, and it's safe to assume they were able to make adjustments and/or improvements to GF106 and GF108. Nvidia, just like ATI, has multiple teams working on several projects and I think it's reasonable to assume that the work on refreshing Fermi began before Fermi was even out with the engineering teams already applying and working on this iteration of Fermi's shortcomings.

So, in essence, I once again think that the GF100 refresh and ATI's 6x00 series are going to be very competitive vs. one another and will finally incite a price war we've all wanted.

 

[Daedalus685]

I agree as well. I also would point out that it doesn't even give an idea of potential. To quote the OP:So really unless your planning on replacing your vid card after a marithon 24 hour gaming session, this doesn't have much relevance to the real world.

Still, it is interesting and gives an idea of the benefits of the different architectures. Pity TSMC couldn't get 32nm going in time for this generation. I bet things would have been dramatically different with nVidia dominating this generation.

If TSMC had 32nm ATI would have released NI by now... Ifs and buts being candy and nuts as it were.

 

[RussianSensation]

i don't like that all the benchies are overclocked gpu's

i think a stock run is more important

To a certain extent I agree with you as most people will not overclock. However, it has been very difficult to find benchmarks where both camps are overclocked and compared to one another. This just provides that something little extra for anyone who was curious.

But even knowing that, it's great work and hard work OP for going through so many benchmarks, this must be commended. great job!!

Full credit to NCspecV81 from overclock.net forum, not me. I only linked his work as I thought it was well done.

 

[MagickMan]

I am wondering how much juice those Fermis require to OC that high, not to mention the heat coming off of them. Impressive, though.

Also, it really shows how much the ATI cards are in need of a price cut, and what I've said for months and months now, if you're will to OC there's no reason at all to buy a 5870 over a 5850.

 

[RussianSensation]

Magick, completely agree on your comment for 5850 > 5870 considering their pricing.

Anyone who is interested for CF/SLI benches -- Xbitlabs just did a round up including:

480 vs. 5870
480 2-way SLI vs. 5870 2-way CF
480 3-way SLI vs. 5870 3-way CF
and 5870 4-way CF

Can be found here: http://www.xbitlabs.com/articles/mainboards/display/ga-x58a-ud9_6.html#sect1