Triple-GPU Scaling: AMD CrossFire Vs. Nvidia SLI

[dennisburke]

It does seem the games chosen for this test favor AMD, but even more important is it's well known in the Nvidia community the driver support for Tri SLI is lacking across many titles, and that many are waiting for better drivers.

 

[JPForums]

Today’s question isn’t which single-GPU card is best, but which cards operate best in pairs and trios.

The implication here is we are looking at how well the cards scale with numbers rather than the absolute performance.

The latest 3DMark version is supposed to remove any Nvidia bias as a result of artificially-inflated PhysX results, so we see big problems for AMD if these results are repeated in games.

The scaling percentages for multiGPU setups as given in your first chart are as follows:
3-way SLI (overall) %190
3-way SLI (GPU) %278
3-way Xfire (overall) %203
3-way Xfire (GPU) %286

3-way SLI (overall) %159
3-way SLI (GPU) %196
3-way Xfire (overall) %167
3-way Xfire (GPU) %200

The results are similar on every chart on this page. I'd say these result pose no problem at all for ATi given that they scale even better (if marginally) than nVidia. Of course, ATi might have something to worry about if you look at absolute performance. However, you would have to belie the stated focus of the article with the very first comment of the very first test result in your article. Why don't we check it out anyways.

We'll start with the premise you laid out earlier:

The closest competitor is Nvidia’s GeForce GTX 570

Asus ENGTX570 from article - $320 ($340 before MIR) at newegg
GTX570 - As low as ~$320 (~$335 before MIR) at newegg

HD6970 - As low as ~$310 (~$340 before MIR) at newegg

XFX HD-695A-CNFC from article - $260 ($290 before MIR) at newegg
HD6950 2Gb - As low as ~$250 (~$275 before MIR) at newegg

GTX560 Ti 2Gb - $280 at newegg

After evaluating prices, it would seem that the closest competitor is the GTX560 Ti 2Gb. Alternately, you can just compare the HD6970 to the GTX570.

It seems ATi still has little to worry about. Of course, you may find that the HD6970 still looses, but until you post such a result your comment is mute. In fact, rather than be worried, it would appear that ATi has a commanding advantage at this price point. They have the ability to support 3-way Xfire setups (with good scaling) with a lower cost card than the one nVidia limits to 2-way SLI.

It's not often I'm provoked into commenting on an article before I finish reading it, but I felt that there was no excuse for this comment. Next time please stick to your stated premise of the article.

 

[JPForums]

After reading the rest of the article, I retract my final statement of "Next time please stick to your stated premise of the article", but retain the sentiment that the above, as well as some other similar comments in the article were poorly conceived (as detailed in my previous comment). I would like to thank you for eventually returning to the focus of the article.

I understand a little better why you wanted to compare the absolute performance of these cards even though they were in different price classes. Put simply, at 2560x1600 the ATi cards were quite compelling, from an absolute performance standpoint, despite the price delta.

 

[KT_WASP]

[citation][nom]megamanx00[/nom]The 6950 is not the cheapest 3-way crossfire configuration. There are some 5770s with two connectors to allow up to 4-way crossfire, and the same goes for 6850s, and 6870s. I suppose if you're strictly talking about reference designs though then yes the 6950 would be the lowest option considering the 5800 series is being phased out.[/citation]

Yes, the HD 6950 is the cheapest card from THIS GENERATION that will do 3/4 card setups. The HD 6850 and HD 6870 only allow two cards max in multiple setups.

From guru3d:

As stated above, the Radeon HD 6850 and 6870 only have one CrossfireX connector, this literally means that you can place only two cards into a multi-GPU setup.

http://www.guru3d.com/article/radeon-hd-6850-6870-crossfirex-review/1

Same goes for the GTX 560ti... only allows 2 cards for SLI.. so for the current crop from Nvidia, the lowest priced option in the GTX500 series is the GTX 570.

Please.. show me a link that shows a 6850/70 or a 560ti that has more then on xfire/SLI link...I checked the cards on Newegg and it didnt matter if it was reference or not.. none of them have more then one link. And, since you need two links for more then a two card set-up.. your argument holds no water.

 

[youssef 2010]

[citation][nom]article[/nom]AMD's Radeon HD 6950 2 GB was its least expensive three-way CrossFire option when we set up this article.[/citation]

You're wrong as the 5750 is much cheeper and some of these support CrossfireX. Anything higher than that supports CrossfireX natively.IIRC, 5770 were always widely available.

[citation][nom]Crashman[/nom]Asus' 25" 1920x1200 display is still well priced, but I don't expect to see them around much longer. You should tell Chris to order three of these for every editor, before that bargain price vanishes![/citation]

lol.

 

[youssef 2010]

Keep in mind that Nvidia's cards are worth $100 higher than AMD's cards

 

[youssef 2010]

So, 3-way SLI turns into a 3-way loss for Nvidia in power,value, and performance.lol

[citation][nom]Article[/nom]If CrossFire’s value-oriented victory in two-way and performance win in three-way configurations weren’t enough to convince us, the Radeon HD 6950's lower power certainly did. AMD started out with the highest efficiency at one card, and superb multi-GPU scaling allowed its efficiency to increase with every card we added. When’s the last time adding parts made your machine more efficient?[/citation]

Beat that, Nvidia.

To clarify, I am NOT an AMD fanboy. But I respect AMD's vlaue oriented approach and the fact that it is competing on CPU and GPU fronts.I just feel that Nvidia is very stuck-up about its cards.

 

[Computerrock1]

I think the results would have been different had it been the 580

 

[Marcus52]

Looks to me like the choice is AMD if the single-card performance of the game is within 10% or so of Nvidia's cards, but if Nvidia's GTX performs significantly better to begin with, then 3-way SLI will also.

I do have to wonder if the test suite isn't representative of the gaming world anymore - are the 3DMark scores that far off of real-world results? Somehow, I don't think so. What I do know is my main app is World of Warcraft, not represented here, but thanks to Chris Angelini's testing I know that it runs much better on Nvidia GTX cards, so it likely will on SLI 3-way. It also runs better on SLI than Crossfire in dual-card setup, so that's 2 good reasons for me to choose Nvidia over AMD.

 

[17seconds]

The only problem, and it is a major flaw, is comparing a higher performance card against a lower performing card. The higher performing card will always be more bottlenecked by the CPU than the lower performing card. The lower performing card will always show higher scaling as a result of less bottlenecking. A comparison of the 6970 in 3 way Crossfire vs. the GTX 570 cards would be the exact comparison to make and would show a much more accurate picture. It is premature to declare a winner until those results are posted.

 

[wildeast]

@matto17secs
An Intel Core i7-2600K overclocked to 4.0GHz cannot be bottlenecked by today's setups, even if it'll (and it won't), you are testing according to the strongest cpu on the market and you won't get better performance than this.

 

[stingstang]

"Local banshees abandoned our neighborhood at the commencement of power testing." ...What the hell?

 

[Marcus52]

[citation][nom]obarthelemy[/nom]and nobody cares... the %age of multiple-GPU users (and gamers) is what ... 0.01% and 0.1%, respectively ?[/citation]

Forum troll. Has to be, no one in their right mind would say such a thing.

 

[Marcus52]

[citation][nom]xciter[/nom]Why are you comparing two gpu`s that are not even in the same price range. Accroding to newegg the 6950 is ~240$ and the 570 is ~350$. Multiply that by three or two and you get a lot of price difference.Favoring nvidia are we?[/citation]

Another forum troll. The rest of you probably think these guys are idiots, but I know better; they are actually posting stupid things to get a reaction out of you. Well, that doesn't mean they aren't idiots, but I don't think that's their main purpose. It couldn't be, no one who can actually post could be that dumb. I mean, if you can write a post, you can read the article, right?

 

[davewolfgang]

I mean, if you can write a post, you can read the [strike]article[/strike] title, right?

fyp



:-D

 

[Crashman]

[citation][nom]xciter[/nom]Why are you comparing two gpu`s that are not even in the same price range. Accroding to newegg the 6950 is ~240$ and the 570 is ~350$. Multiply that by three or two and you get a lot of price difference.Favoring nvidia are we?[/citation]Wait...the cheaper cards perform better...favoring AMD were we? How many ways would you like that question answered that weren't already covered in the article, or in previous responses?[citation][nom]matto17secs[/nom]The only problem, and it is a major flaw, is comparing a higher performance card against a lower performing card. The higher performing card will always be more bottlenecked by the CPU than the lower performing card. The lower performing card will always show higher scaling as a result of less bottlenecking. A comparison of the 6970 in 3 way Crossfire vs. the GTX 570 cards would be the exact comparison to make and would show a much more accurate picture. It is premature to declare a winner until those results are posted.[/citation]This was "cheapest to cheapest" for the current generation of cards. Perhaps you should ask Nvidia why it doesn't have 3-way capability on cheaper cards

 

[Guest]

I use 3-way of 5770s, which only works great at 5760x1080, and it only cost me 350, and I average 40fps min in all my games at almost max settings. So Ya, go crossfire for the fact you can do triple configurations with lower-end cards.

 

[gamerk316]

Benchmark FAIL. CPU bottlenecks makes the testing invalid.

Also, no listing MINIMUM FPS numbers, where NVIDIA's numbers tend to be better.

 

[cburke82]

Based on the lowest newegg prices the cheapest 6950 is $274 and the cheapest 570 is $334 so Unless you just cant stand the performance of one 6950 and cant by 2 cards to start there is no reason to go Nvidia except pysix. the $120 you save going with the 6950 could almost buy you a new mobo that supports xfire if your current one does not and your still getting better performance in most games. And the games that you dont get better performance in you still get VERY playable FPS. Add to that more heat and power usage from the Nvidia cards and that gap is amazing. Im glad I got a 6950 now. Mine also unlocks and is stable with the 6970 bios, so if I can get a second card that unlocks and is stable those 2 will basically take a dump on 2 570's for much less wile requiring less in the way of a PSU I do wish they had tested this on Metro 2033 it seems to be the current standard for seeing what kind of power your system has. Im sure these cards would do good there though as my one card is doing just fine playing the game by itself

 

[cburke82]

someone here posted something about WoW performing better with Nvidia......That is true i have seen the articles but if you are looking to have 2-3 card setup for WoW my question is why? I have one 6950 and except for rare occasions i never go below 60FPS in stormwind i hit 35-40FPS and if I take v-sync off it reg gets over 100FPS....so unless you have a screen with a 250hz refresh rate are you really going to see a performance increase worth $300 in WoW with AMD or Nvidia? The only reason Im posting this is because this is a muti card test if this where a single card comp you would be 100% right in pointing that out.

 

[JackNaylorPE]

This is a good article in concept, but the problem is I could write the same article ..... that is someone pick the desired, (ATI or nVidia as winners) and one can easily "prove" the desired conclusion. Then I can do it again and get the opposite conclusion, simply by handpicking the games in the test.

My issues with this article are twofold:

...... 4 games ???????

....... for consistency, would have liked to see consistency in the test suite so as to compare with past THG articles. Choosing a new set makes results less relevant and perhaps even suspect as to why those 4 games.

Here's Guru3D's results on dual SLI.

Card 570 570 SLI 6950 6950 X-fire

COD-MW - 148 230 127 150
Bad Company 2 - 56 103 50 96
Dirt 2 - 80 138 72 131
Far Cry 2 - 85 153 76 147
Metro 2033 - 27 54 25 50
Dawn of Discovery - 74 109 78 90
Crysis Warhead - 54 86 51 87

Total 524 873 479 751
Scaling % 166.60% 156.78%

Here nVidia scales better 166.60 % to 156.78 % but more telling is the fps battle ...... 873 in SLI to 751 in CF .... and 825 fps for the 6970 in CF

This is probably the only resolution in today’s test that really matters, since a single card will pass most tests at anything less.

Matters to who ? Most of the builds I'm asked to do (averaging two a month) are getting a 570 or 6970 with the pretty solid intent of adding a 2nd in SLI or CF ...... mainly for Metro 2033 and the other games which will arrive in XMas 2011.....and te focus is 1920 x 1200 / 1080. I have done one 2560 x 1600 build in my life and that was for a photography buff. I have just under 6,500 posts here and maybe 2 or 3 times I put together a recommended build with a 2560 x 1600 resolution. Almost every build has a CF or SLI recommendation either from the getgo or as a future consideration. At 2560 x 1600, though, no argument ....ATI's clearly has the edge at this resolution.

The 570 and 6970 are about as closely matched a pair of cards from the two camps as I can remember. The relative performance between the two as single cards makes either one a sound choice. The tri SLI take here is interesting but the conclusions make some very broad comments not only about 3 cards but about twin card SLI / CF as well. Considering the contrary conclusions in other articles on the topic, including just 4 games in the test suite, and THG's decision to vary from their own standard game lineup is one I would have rethought. The sweeping conclusions drawn from such a small test sample seem more than over reaching.

 

[dark_lord69]

I really don't like SLI or XFire. I think if a single card can't do the job then you have the settings too high or the programmers didn't optimise the code enough to make the game playable on today's current hardware.

 

[megamanx00]

Next review, three cards, three 1080p monitors. Come on guys, do it.

 

[17seconds]

http://www.guru3d.com/article/radeon-hd-6950-crossfirex-review/
http://www.guru3d.com/article/geforce-gtx-560-ti-sli-review/
http://www.guru3d.com/article/geforce-gtx-570-sli-review/

Average dual card scaling across all 8 benchmarks at 1920x1200:
6950: 1.67x
560: 1.69x
570: 1.7x

Noise levels (vs. single card)
6950: 49 Dba (+8 Dba)
560: 40 Dba (+3 Dba)
570: 42 Dba (+2 Dba)

GPU Temps (vs. single card)
6950: 89c (+12c)
560: 74c (+6c)
570: 81c (+4c)

 

[spentshells]

No thanks bad review you should be reviewing like cards 3x6970 is what we should have seen, they still would have be walked by the 570's but unfair none the less