AMD CrossFire Vs. Nvidia SLI Scaling Analysis

[hixbot]

Is there a possibility we can revisit microstuttering in a future article? Perhaps an entire artcile devoted to the subject. We all really need an update on that important issue.

 

[rantoc]

Got 2x 480 GTX in SLI, they work wonders on a 2650x1600 display. And for you knuckleheads who like to add a zillion cards like quadcore ect... dont. Why you ask? Latency! I rather have one MONSTER CARD with one gpu than 4 weaker. Sadly there arent and better single/dual gpu solutions available. Hope ati/amd's new singlegpu flagship fixes it (and hope they dont spare too much on the framebuffer like they usually do, hurting the card at really high resolutions/aa).

 

[rantoc]

[citation][nom]rutoojinn[/nom]This gen has been great to both ATi and Nvid since the performance is quite similar. Prices are the only thing to consider when purchasing a card.[/citation]

I don't completly agree, always had both manfacturers cards in my rig and i have to give Nvidia lots of credits for their driver efforts. ATI have started to caught up but arent there just yet.

 

[scrumworks]

[citation][nom]Cleeve[/nom]Really? I'd be very interested in seeing some proof of that, scrum. I believe you might be making a huge assumption, there.[/citation]

Well if you for starters take a look at Catalyst 10.8 release notes you can spot "FarCry 2: Performance increases 2-6% on ATI Radeon™ HD 5800 Series single and CrossFire™ configurations". I don't expect huge gains but I believe the crossfire profiles have been improved since 10.7 betas.

 

[Khimera2000]

thats alot of power draw o.0 but really intresting as for the issue of drivers on test i dont mind reading reviews that have drivers one or two generations old, but if its more then 3 generations old, or beta plus one driver releas old i think there should be new drivers expected, otherwise where doing a comparison that might not be valid.

Just an opinion. i like the results but a chunk of it I take with a grain of salt... simply becaus the driver, and the 5850 issues in the article, but fun read none the less

 

[WarraWarra]

Amazing the constant ration of improvement ie: single 5870 vs 480 percentage diff is almost or the same as when in sli/xfire.

Would not have thought when the "ovens" NV480's came out that they would do this well against the 5870's .

Is the 480's 1st gen or 2nd gen with a few fixed heat / power issues resolved ??

PS> if you used the Win7 SP1Beta and likely 64k+ cluster on the hard drive you might have had better performance results during testing + games might actually have run better with the fixes in the Win7 SP1Beta.
SP1Beta is about to convert me from XP to win7 as much as I hate Win7 irritations.

 

[cleeve]

[citation][nom]scrumworks[/nom]Well if you for starters take a look at Catalyst 10.8 release notes you can spot "FarCry 2: Performance increases 2-6% on ATI Radeon™ HD 5800 Series single and CrossFire™ configurations". I don't expect huge gains but I believe the crossfire profiles have been improved since 10.7 betas.[/citation]

Yeah, we see those types of claims often but I'm yet to see one that holds up to actual testing. And like razberybandit claims, nVIDIA 260.89 driver notes lay claim to GTX 460 performance increases of 5-19% in four of the titles/resolutions specifically used within this test.

If it makes you all happy I can rebench Far Cry 2 just to show you how nothing will change at all... you will likely be disappointed by the lack of difference, though.

 

[Guest]

I see some of you saying that ATI is doing a good job but needs to pick it up a little bit lets think that through a little. If all of you Nvidia fan boys have forgotten ill remind you that ATI had their 5xxx series cards out 6 months before the 470 and 480 were on the scene. Sorry to say but ATI won this generation of cards war. You cant come into the ring 6 months later and be 16% faster or a little more in some situations and say you win sorry ATI kicked Nvidias butt.

 

[dEAne]

Oh yeah thank toms for this.

 

[scrumworks]

[citation][nom]Cleeve[/nom]Yeah, we see those types of claims often but I'm yet to see one that holds up to actual testing. And like razberybandit claims, nVIDIA 260.89 driver notes lay claim to GTX 460 performance increases of 5-19% in four of the titles/resolutions specifically used within this test.If it makes you all happy I can rebench Far Cry 2 just to show you how nothing will change at all... you will likely be disappointed by the lack of difference, though.[/citation]

So you are saying AMD people is a bunch of liars?

 

[pandemonium_ctp]

I've been waiting for an updated Nvidia SLI versus ATI Crossfire comparison for a while. I'm also glad to see that you're using Mushkin memory for your benchmarking. Mushkin is the good stuff.

[citation][nom]gkay09[/nom]The SLI scaling is one place where Nvidia wins almost all the time, but ATI has improved a lot though, it still need to work harder if they want to have a complete win over Nvidia...[/citation]

This has me confused, since previously CF's scaling was better than SLI's and recently Nvidia has improved, not ATI/AMD. I think you have it backwards.

>> http://www.tomshardware.com/reviews/crossfire-pci-express,2095-7.html

Scroll down to the high-res and high settings scaling graph. Nearly 200% for both the X48 and P45 boards, which should be theoretically as high as you can possibly scale (without CPU or PCI bus or whatever bottlenecking performance). If you're comparing scaling at low resolution or low settings, then you don't understand the point of multi-GPUs.

Now to take a general census of FPS scores in frames listed here on Tom's 2010 charts for ATI's and Nvidia's reference cards:
2010 Gaming Graphics Charts (High Quality) > Sum of FPS Benchmarks 1920x1200 TOP 4 reference cards for ATI CF and Nvidia SLI scalability difference
http://www.tomshardware.com/charts/2010-gaming-graphics-charts-high-quality/Sum-of-FPS-Benchmarks-1920x1200,Marque_fbrandx876,2491.html
Nvidia Geforce GTX 460 SLI (2x1024 MB not labeled as 2x on website?) 455.40
Nvidia Geforce GTX 460 (1024 MB) 271.00
168.04%
Nvidia Geforce GTX 460 SLI (2x768 MB not labeled as 2x on website?) 434.50
Nvidia Geforce GTX 460 (768 MB) 251.70
172.63%
Nvidia Geforce GTX 280 SLI (2x1024 MB) 387.90
Nvidia Geforce GTX 280 (1024 MB) 206.30
188.03%
Nvidia Geforce GTX 260 SLI (2x896 MB) 335.20
Nvidia Geforce GTX 260 (896 MB) 178.20
188.10%


http://www.tomshardware.com/charts/2010-gaming-graphics-charts-high-quality/Sum-of-FPS-Benchmarks-1920x1200,Marque_fbrandx4,2491.html
ATI Radeon HD 5870 CF (2x1024 MB) 509.00
ATI Radeon HD 5870 (1024 MB) 367.20
138.62%
ATI Radeon HD 5850 CF (2x1024 MB) 475.40
ATI Radeon HD 5850 (1024 MB) 312.10
152.32%
ATI Radeon HD 5830 CF (2x1024 MB) 419.40
ATI Radeon HD 5830 (1024 MB) 242.80
172.73%
ATI Radeon HD 4890 CF (2x1024 MB) 432.20
ATI Radeon HD 4890 (1024 MB) 241.80
178.74%

While this is pretty broad of a comparison, it does show from the collective data here that ATI's scaling has decreased. Of course, CF cards don't scale beyond 2 cards very well, and SLI does, so this could affect the data here, as well as many other variables including testing platforms used, benchmark system favors, instances tested, etc...etc... Regardless, it shows that Nvidia is scaling [against itself in SLI] well right now.

Back to the article, scaling versus a baseline card which is a competitor's of a different price point isn't exactly accurate and places a skew on the comparison. If one wants true accuracy with comparing SLI/CF, take points awarded or average FPS or whatever the benchmark uses and divide it by market cost. When it comes to it, cost is the bottom line and should be the common denominator used for dividing factors between brands.

And as I always like to mention, include monthly operational costs as well. (Don knows I'm a stickler for this one - from previous notes I've made. )

Thanks for the updated review. ^^

 

[hexsor]

Just to be fair, where is the crossfire 5970's????????.....if you going to bring out nVidias flag ship, why not bring out ATI flag ship too

 

[mapesdhs]

Cleeve,

If the GPU positions in the chart are based on pricing, why is GTX 460 SLI
listed above a single GTX 480? Here, the typical price of the Zotac GTX
480 AMP is about 400 UKP, whereas the Gigabyte GTX 460 GV-N460OC-1GI is
only about 169 UKP, ie. two of the Gigabyte 460s is nearly 60 UKP less
than the GTX 480 AMP.

Or are you factoring in something extra somehow for mbd/PSU requirements?
(that shouldn't be necessary these days given how cheap it is to obtain a
good mbd that supports SLI/CF, eg. the excellent Asrock P55 Deluxe is only
75 UKP here, while the X58 Extreme6 is a mere 177 - the same or faster than
the Gigabyte UD9 yet less than half the cost).

The best price I could find for the Zotac GTX 480 AMP was 386.51 UKP
(envizage.com) while the best price for the Gigabyte card was 159 UKP
(ebuyer.com) so the saving with two of the 460s is even higher than
with average pricing if one makes the effort to shop around.

Hence, GTX 460 SLI is in the wrong place on the graphs. It should be below
the GTX 480 because two 460s always costs significantly less - a point
which was strongly made in all the launch reviews of the 460 and is still
the case today.

At the low cost end of the GTX 480, the pricing is about 340 UKP, eg.
the ASUS ENGTX480/2DI/1536MD5. Likewise, the cheapest GTX 460 1GB is
about 153 UKP (eg. Gainward Golden Sample). Thus, even going for the
minimum possible cost, the 460 SLI config still costs much less than a 480.

Or is US pricing the reverse with respect to 460 vs. 480? Does a typical
480 in the US cost less than two 460s? That would certainly be a change
in the way things have been for a while, but it's not filtered elsewhere
yet if so.

Ian.

PS. If I was asked to recommend a 460 1GB card, it would have to be the Palit
Sonic Platinum (800MHz GPU); at typically 180 UKP, it matches or often
outperforms a 5850 and two of them are amazingly faster than a 480 in SLI,
eg. see the reviews on tweaktown.com, bjorn3d.com, techpowerup.com, etc.

 

[ITsonic]

Don't it really matter which card we buy? Both companies make good video cards so I won't take sides on this battle.

 

[pandemonium_ctp]

[citation][nom]ITsonic[/nom]Don't it really matter which card we buy? Both companies make good video cards so I won't take sides on this battle.[/citation]

Indeed it does. Brands tend to have their quirks and benefits though, thus the loyalty or dismissal.

 

[jaquith]

I run a 3-WAY GTX 470/Koolance on (3) three monitors I only "wish" the resolution was that LOW -> RE "Frankly, there aren’t a lot of 2560x1600 monitors out there in the wild, so this probably isn’t a major concern for most folks. But for gamers planning on a high-resolution 30” monitor or a triple-monitor gaming rig, GeForce GTX 470/480 cards or Eyefinity Edition Radeon cards with 2 GB of RAM will be something to look into."

Typically, folks {Gamers} who ARE running SL/CF have multiple monitors, or 30" monitors so this makes no sense.

 

[pandemonium_ctp]

Oh, if you're curious on active market share, check out http://www.videocardbenchmark.net/30dayshare.html.

That site is very useful, albeit very broad and specific to their testing platform. Enjoy!

 

[pandemonium_ctp]

[citation][nom]jaquith[/nom]I run a 3-WAY GTX 470/Koolance on (3) three monitors I only "wish" the resolution was that LOW -> RE "Frankly, there aren’t a lot of 2560x1600 monitors out there in the wild, so this probably isn’t a major concern for most folks. But for gamers planning on a high-resolution 30” monitor or a triple-monitor gaming rig, GeForce GTX 470/480 cards or Eyefinity Edition Radeon cards with 2 GB of RAM will be something to look into."Typically, folks {Gamers} who ARE running SL/CF have multiple monitors, or 30" monitors so this makes no sense.[/citation]

I wouldn't necessarily say that. I, along with other [gamers] still prefer their CRTs to large display LCD or LED displays (not in Eyefinity fashion - that'd be annoying). I'm using CF simply because it's more cost effective to use than a more powerful and more costly single GPU setup. (I know I'm not typical though. )

But yes, typically the best argument to use for CF/SLI is for very high resolutions on larger/multiple display areas.

 

[nottheking]

[citation][nom]Twoboxer[/nom]12 million people with WoW subscriptions,[/citation]
Actually, WoW isn't at 12 million now; they were past-tense referenced as 12 million at the beginning of 2009, though the statement made by Vivdendi was in February, so there's a good chance it was rounded up, and it never reached that number. This contrasts to all other figures they provide, which they regularly use, and include one certified by Guiness World Records.

At the very least, since then they've dropped to 11.6 million, after which came the temporary shut-off of WoW China, which briefly dropped it to 6.6 million. (yes, China accounts for almost half of WoW players, so its impact on the English-speaking gaming world isn't as great as you might expect) And since then, Blizzard themselves have made statements around a year ago that they had 11.5 million.

Chances are, by now, they're down to 11 million; I expect some more to 'come back' for Cataclysm, but overall, I think they've hit their peak, and will gradually decline from here on out.

[citation][nom]cangelini[/nom]I'd like to hear some discussion on the best way to reliably test WoW, actually.[/citation]
Well, unfortunately, at least from my understanding of the game's design, the way WoW works means that the networking interface lags everything else badly; the reason a GPU has any effect is that reducing frame render times can compensate for lagging packets.

If one wanted to study the effect of computer hardware on the client side for performance of the game client, I'd recommend the use of a third-party server "emulator," preferably located on a local network to best minimize the impact of connection. I know that these emulators exist, though from what I hear they're rather inaccurate in gameplay terms. However, in mimicking the environment the client has to handle, it'd likely be perfect for benchmarking.

For further testing, it might be interesting to run these benchmarks, then also repeat them with different connection types to mimick more real-world situations, such as making the server remote, (potentially in a differnet Tom's lab) and connecting with different types of Internet connection. This could help place things into perspective. (i.e, "would you be better with a GPU upgrade or switching to a higher bandwidth/lower latency Internet connection?")

Actually, I'd suggest that an article dedicated to benchmarking WoW in this manner might be a rather good idea.

I would somewhat caution, though, that while the server emulators THEMSELVES are fully legal*, the terms of service for playing on WoW's own servers do state that you're not allowed to connect to third-party servers. While I doubt they'd act on anyone for it (I've yet to see evidence they have) technically they'd be in their right to ban you from playing on their Pay-2-Play servers. (*server emulators are made to contain no actual Blizzard-made code or content, as obviously, no one's allowed to look at Blizzard's own server source code.)

[citation][nom]Cleeve[/nom]If it makes you all happy I can rebench Far Cry 2 just to show you how nothing will change at all... you will likely be disappointed by the lack of difference, though.[/citation]
Cleeve, you never cease to be amazing. First this article, and now, of course, responses like this. Always there to inject a bit of reality to chill some of these heated arguments.

 

[Anthelvar]

ATI/AMD 6870 crossfire scaling from 90% to over 100%. Hilbert just did a review. VERY GOOD RESULTS. Huge performance improvement.

 

[kelemvor4]

3.61ghz cpu bottlenecks a single gtx 480. This needs to be done with a much faster processor and at higher resolutions/video settings to be an effective review.

 

[cleeve]

[citation][nom]scrumworks[/nom]So you are saying AMD people is a bunch of liars?[/citation]

Liar is a strong word. More like stretching the truth to a theoretical best-case scenario, the specifics of which they do not publish along with their outrageous performance claims.

Interesting that you would specifically refer to AMD when the example you quoted from me referenced Nvidia drivers. Certainly the problem is not from not AMD employees specifically, but the folks who write driver notes in general, from any camp. Also from Nvidia for sure. And likely this will be also be perpetrated by Intel if they ever offer serious graphics hardware and driver updates that people bother to notice.

Not company specific, despite your desperate, fanboyish attempts to spin as such.

 

[cleeve]

If the GPU positions in the chart are based on pricing, why is GTX 460 SLI
listed above a single GTX 480?

Pricing changes, often on a daily basis. Pricing data is only valid the day of writing.

 

[mapesdhs]

What I meant was, GTX 460 SLI has _never_ been more expensive than a GTX 480.
This was true when the 460 launched and it's still true now. It really is in the wrong
place, no matter what happens on a daily basis. What 460s did you use for this test?
What 480? What were their prices, and which crazy site was selling the 460s at such
a high price?

Either way, it definitely does not reflect current pricing. Please move the 460 SLI placing
down below the 480. Remember people will assume the former is more expensive because
of its position in the charts, but the implication is false. Afterall, all of this work is intended
to help people make better purchasing decisions, yes? 8)

Ian.

 

[cleeve]

What I meant was, GTX 460 SLI has _never_ been more expensive than a GTX 480.
This was true when the 460 launched and it's still true now.

That's quite untrue.

You can get 480's off of newegg for as low as $440 right now. When the 460 1GB was $240, it cost $480 SLI'd. I believe you could get a GTX 480 for about $460 when the article was written.

As far as changing the article, I'm not going to retrofit every article I've ever written every time there's a pricing change. That is not a reasonable request. People will see the current prices when they look to buy.