Quad-GPU Performance Review NVIDIA vs. AMD @ [H]

I really enjoy quality and thoroughness of the reviews here but what I don't understand is why you constantly test at resolution that a majority of people are unable to access(2560+ etc.). Could someone explain this?
 
nice analysis. I wasn't expecting much as far as quad scaling goes. I always thought you were best off getting the second best thing out on the market. You save a lot of money. Of course this has nothing to do with saving money, but just a thought.
 
JMccovery said:
Wow... Just wow! Not to nitpick, but whenever Nvidia decides to launch the 'new' GTX590 (hopefully this one will have 3GB per GPU), will there be a retest at 5760 or no?
Click to expand...

The problem with a 2x2GB 590 is that the 590 is already power limited, and using 2x as many ram chips or 2x capacity ram chips would increase the amount of power needed for the memory subsystem and leave less behind for the GPUs themselves. As severely throttled as they already are; having to cut their power even more would cause it's own set of problems, and potentially leave performance unchanged but now shader bound.

Beyond that, look at the last page of the review. 3GB 580's are on the maybe list; hopefully we'll be able to see reviews of them soon.
 
I would personally be interested, if there is a 3/4 way discrete review, to see the 6950 thrown in the mix. Unlike stepping down to the smaller nvidia card (the 570), there are no memory limitations by going to the 6950.

Of course some of them can also be unlocked, but even left alone I bet the 4-way performance would be pretty damn impressive for only $1000 if you catch them on sale.
 
I'm not sure how much power RAM actually uses, but I doubt it can be all that much. Maybe 20, 30 watts? And in the case of the 580, doubling the RAM means doubling the capacity of each RAM chip, so it can't add all that much to the power consumption. And I seriously doubt that nVidia would release a 3gb version without seriously beefing up the power system.

@bologna: The reason why they test at insanely high resolutions is because the only reason to get these high-end cards is to run them at high resolutions. If you look at reviews of single cards, they test them at 1080P in addition to higher resolutions.
 
Haha, awesome review as always. This was just way too awesome to see as I think I got turned on by the kind of performance numbers that were being put up by these video cards. Being able to see this kind of performance in Metro 2033 and Crysis Warhead was especially impressive from the ATI setup. I guess we finally have a combination that can completely max Crysis out.

While I was a bit disappointed to see problems that interfered with testing from both sides of the camp, I have to admit that I pretty much expected this considering the complexity of it.
 
Bologna Tits said:
I really enjoy quality and thoroughness of the reviews here but what I don't understand is why you constantly test at resolution that a majority of people are unable to access(2560+ etc.). Could someone explain this?
Click to expand...

They find the max playable settings. Anything less than what they used, resolution, is going to be easier on the cards and get better frame rates. Then again if your not running high rez, theres no need for this setup.
 
Chameleoki said:
Wow. Very surprised at the troubles you had. Especially with the lack of troubles with the triple GPU tests you did. I would have thought mixing cards would give you troubles, not two of the same.
Click to expand...

Indeed, it seems the complexity of more GPUs, in this case 4, is what caused most of the issues, and I do believe they are software related. Games are just not built with being accelerated by 4 GPUs in mind sadly. I wish more game devs would make their games with scaling from 1, to 2, to 3 to 4 GPUs in mind and gear the game so it works with AFR well.

Crest said:
AMD just needs to stop making terrible driver packages and get their OpenCL support out there. Their 6xxx series are definite winners, but their software side is really really lacking.
Click to expand...

The AMD driver package is not terrible. I will agree with you on AMD's OpenCL support in general. As a proponent of both Direct Compute and OpenCL, they really need to beef up the software support. Also, NVIDIA is ahead of them currently by supporting some DX11 multi-threading code in its drivers for Civilization V and Battlefield: Bad Company 3 both which utilize DX11 driver multi-threading. AMD needs to step up their game in this department.

Krenum said:
Holy shit ! Almost a Kilowatt at full load :eek:
Click to expand...

Indeed, and on our old system the GTX 590 SLI setup would probably be in the 1100 system Watt range. This system seems to be around 200W less than our old system was, at full load Wattages I'm finding out.

Teenyman45 said:
Kyle, had the bios switch been flipped, what do you believe the difference in the 6990's frame rate and power consumption would be? All things considered I doubt it would go much more than 3-4% do to scaling issues but would still consume at least 100-125 watts more power.

For the umpteenth +1 iteration of this ultra high-end graphics test, will you be running overclocked doubled dual GPU tests after Nvidia finishes revising the 590 to be more stable when overclocked?

To Jarman: You do realize that (shameless plug) the [H]'s front page clearly links to a deal going on at the 'egg where Galaxy 580's can be had for a penny under $430 after rebate; so it's only a "mere" $1,720 to get quad 580 power.
Click to expand...

Whatever comes down the pike, we will evaluate. I don't really see how overclocking a Quad-GPU setup will help more than we are seeing though, we are limited by other factors, not GPU horsepower here. Scaling is a big one, it gets worse and worse the more GPUs you add, so overclocking becomes less useful.

Undoubtedly switching the 6990 BIOS will add more power demand, as it also overvolts each GPU. However, the speed increase of 50MHz per GPU won't add anything relevant to the performance already delivered. So it is not really worth it in a Quad-GPU config.

tricky0502 said:
i like the fact that you used an 1155 system.

what i really want to know is how much fun is it to have that much power in front of you?
Click to expand...

Great fun

Omegas said:
So is this 1155 system going to be the main gaming test rig for a while or was this just for the current review?
Click to expand...

It will remain our testing rig for doing anything Multi-GPU related, as well as testing very high-end cards. I have a 3GB GTX 580 on the way that will be tested on this system.

Bologna Tits said:
I really enjoy quality and thoroughness of the reviews here but what I don't understand is why you constantly test at resolution that a majority of people are unable to access(2560+ etc.). Could someone explain this?
Click to expand...

What's the point of spending $1500 on a GPU setup to play it at anything less than a 30" display? Quad-GPU is pointless for resolutions below 1920x1200 or 2560x1600. Also, to push the GPUs, we have to test at high resolutions or you run into CPU bottlenecks, and then at that point your just testing the CPU, not the video cards. We want to test GPU limitations, not CPU ones. Also, highest playable settings, we use what we can use, crank things up as high as they can go and see what kind of gameplay experience is delivered.
 
Man, I have been tempted lately to spend $1.5k on complete upgrades for two boxes! It is amazing (and depressing) what money can buy. Nice article anyway.
 
Thanks for taking the time to review these behemoths, it was an epic showdown for sure.

But you guys really deserve thanks for testing 6970 TriFire, or more importantly, three Cayman GPUs running at 880/5500, which will be interesting to see which side gains dominance in three-way single cards, another important category that doesn't seem to be plagued with driver issues. It just eats up all your slots instead. :p
 
Thank you for your efforts in reviewing these configurations.

It definitely helps in making some decisions that would make The Wife very angry about.
 
Great review as always!

Couldn't be more happier since switching over from the Green Team I absolutely love my 6990 :D
 
dont think ill ever get more than 2 cards. SLI is plently enough for me and since i change cards every year........
 
Rizen said:
Let's be real here, if you are buying a Quad GPU configuration you've already thrown price:performance arguments right out the window.
Click to expand...
You would also have to be unconcerned about the risk of your setup not working on a subset of upcoming (untested) games.

This is pretty much an interesting What If for now.
 
AMD drivers are winning my respect, Nvidia is slowly loosing it. And to think thats the #1 complaint from a Nvidia fanboys perspective. Ever since AMD acquired ATI one improvement after another, AMD it making great strides in more than just the GPU market.... Fusion is the future.
 
DanNeely said:
The problem with a 2x2GB 590 is that the 590 is already power limited, and using 2x as many ram chips or 2x capacity ram chips would increase the amount of power needed for the memory subsystem and leave less behind for the GPUs themselves. As severely throttled as they already are; having to cut their power even more would cause it's own set of problems, and potentially leave performance unchanged but now shader bound.

Beyond that, look at the last page of the review. 3GB 580's are on the maybe list; hopefully we'll be able to see reviews of them soon.
Click to expand...

Ram really does use very little power. In the order of say 10-15 watts for even the Radeons 2gb.
The Fermi cards might use even less due to using more hum-drum, lower speed modules than their ATI counterparts. It really is next to negligible regards power draw.
 
jo0minator said:
dont think ill ever get more than 2 cards. SLI is plently enough for me and since i change cards every year........
Click to expand...

I dont think I will ever have 2 gpus in 1 system let alone 4. Still these articles are not just for my benefit they are for everyone and are always fascinating to read.
 
Action_Parsnip said:
I dont think I will ever have 2 gpus in 1 system let alone 4. Still these articles are not just for my benefit they are for everyone and are always fascinating to read.
Click to expand...

I love these articles as well, I just hope [H] is deriving good benefit and pageviews because of it. I wonder if people are getting high-end video card review fatigue, seems like these threads draw less and less comments after each review. Perhaps it is a good thing, as their questions have been answered with more data.
 
Good Read!

I think performance is inline with what most people thought it would be.

But damn that's a lot of power being consumed... :eek:
 
TwistedAegis said:
I love these articles as well, I just hope [H] is deriving good benefit and pageviews because of it. I wonder if people are getting high-end video card review fatigue, seems like these threads draw less and less comments after each review. Perhaps it is a good thing, as their questions have been answered with more data.
Click to expand...

Well this one does come to the same conclusion the 3-gpu article did, namely:

1) scaling is no better/worse for red or green,
2) top-end nvidia runs into a pretty severe memory deficiencies.

Still enjoyed the read though :)
 
Coincidentally they had several 3gb evgas sitting on the shelf at the San Marcos frys when I bought my 580 the other day, but showed none available on their site
 
Excellent review. It's nice to see some new barriers crossed, such as Metro 2033 with DOF and Eyefinity Crysis with all Enthusiast settings and AA. I'm sure the driver teams are pleased as well. ;)
 
I actually registered to ask these questions:

Would using the monitors in portrait mode make any difference? 3600x1920 is close to 16:9 and might be less problematic for the drivers.

Isn't the benefit of using two of the x2 cards over 3 or 4 cards that you don't lose USB3 and other PCIe functionality?
 
I lean towards Team Red but I was somewhat surprised Team Green didn't do better.
 
I would love to see this non consistent/non repeatable benchmark compared to time demo's just for kicks. pretty sure the results would be a bit different...
 
I stopped reading after the second paragraph......

"What if money wasn't a concern? What if you wanted to plop down all the cash that is required to own the best possible performing video card configuration for gaming with no tweaking? There would only be two options right now that would give you the best possible performance, and that is a Quad-GPU configuration from NVIDIA or AMD. While most gamers enjoy performance with one GPU, sometimes it just isn't enough. NVIDIA's and AMD's solution to this problem is to throw more GPUs at you and hope it does what you need.

Quad 3gb gtx580's vs all = Nvidia winning.
This just seems like another review that favors AMD cards.
If money was not an object ,would anyone here use a gtx590 or 6990 to do the review? I think not unless you were uniformed or just plain biased.

4 3gb gtx580's vs 4 6970's is the ultimate setup. Who buys these cards and does no tweaking?

Edit: So I said to myself , "I wonder how noise and heat levels were"? Of coarse they were left out because the 6990 sounds like a airplane and in crossfire runs hot as the sun.
BUT, they were sure to put the gtx590 losing in power consumption. :( sad day indeed.
 
Last edited:
for 2k you could get 4 3GB GTX 580s and probably blow another 500 or so on water cooling htem and then yes you'd have the ultimate setup: Price here

http://www.google.com/products/cata...=X&ei=3a3aTZ-RLIj4swPirJWMDA&ved=0CDoQ8wIwBA#

But if Physx and 3d gaming isn't so much what you're into and you're for whatever reason liking 4 6970s with 2GB of framebuffer - because of my own non-nvidia bias I'd probably go that route because yeah i'm biased but its the difference between preferring a camaro to a charger or something
 
happymedium said:
If money was not an object ,would anyone here use a gtx590 or 6990 to do the review? I think not unless you were uniformed or just plain biased.
Click to expand...
Thanks to the BIOS switch, the 6990 can be as fast as 6970 CF. And as long as Nvidia keeps claiming that the GTX 590 is the fastest graphics card, it is fully justified to use it in all reviews. Even if it is only to show how ridiculous that claim is.
 
happymedium said:
I stopped reading after the second paragraph......

"What if money wasn't a concern? What if you wanted to plop down all the cash that is required to own the best possible performing video card configuration for gaming with no tweaking? There would only be two options right now that would give you the best possible performance, and that is a Quad-GPU configuration from NVIDIA or AMD. While most gamers enjoy performance with one GPU, sometimes it just isn't enough. NVIDIA's and AMD's solution to this problem is to throw more GPUs at you and hope it does what you need.

Quad 3gb gtx580's vs all = Nvidia winning.
This just seems like another review that favors AMD cards.
If money was not an object ,would anyone here use a gtx590 or 6990 to do the review? I think not unless you were uniformed or just plain biased.

4 3gb gtx580's vs 4 6970's is the ultimate setup. Who buys these cards and does no tweaking?

Edit: So I said to myself , "I wonder how noise and heat levels were"? Of coarse they were left out because the 6990 sounds like a airplane and in crossfire runs hot as the sun.
BUT, they were sure to put the gtx590 losing in power consumption. :( sad day indeed.
Click to expand...

Thought you stopped reading after the second paragraph?
 
Quartz-1 said:
I actually registered to ask these questions:

Would using the monitors in portrait mode make any difference? 3600x1920 is close to 16:9 and might be less problematic for the drivers.

Isn't the benefit of using two of the x2 cards over 3 or 4 cards that you don't lose USB3 and other PCIe functionality?
Click to expand...

Yes.

and

It would all depend on the motherboard you were using.
 
great article. When money doesn't matter THAT is when the fun starts. :) Seriously though the issues you had and the amount of raw performance of both setups is insane!
 
happymedium said:
Mabe you missed my edit. I went on to read a little more out of sheer curiosity.

"Edit: So I said to myself , "I wonder how noise and heat levels were"? Of coarse they were left out because the 6990 sounds like a airplane and in crossfire runs hot as the sun.
BUT, they were sure to put the gtx590 losing in power consumption. sad day indeed. "

So no response to anything else I posted in post #69?
Click to expand...

I as a new user don't want to be polemic in my very first posts here. BUT, you know who is the guy that just answered you? Is the guy that expends hours doing those "biased" reviews, the numbers are there for you to take your own conclusions my friend. If you think what is writen is bs movealong! I totally agree with you in your thoughs about what really is the ultimate setup "if money doesn't matter" but also you seen to be very concerned about how "they were sure to put the gtx590 losing in power consumption" witch is a FACT, and do a little research before saying things about the 6990s. They are way louder sure they are, but just look for some thermal camera shots comparing the silent GTX590 vs the airplane 6990 and see where the sun is really burning friend. They are biased, you say! But Nvidia recomending reviewers to focus about how silent the card is, since it couldn't beat the 6990 isn't misleading right? Being quieter while keeping the pcb and other components way hotter isn't either right? Now let me do a biased interpretation of the review for the AMD side. "W8, why didn't they overclock the cards to see which one would hold the maximum OC??? Just because the GTX 590 can't overclock well! Who in an ultimate setup wouldn't overclock??? They are NVIDIA biased" see? My friend you are much more biased than you clain people that actually work and put real numbers (so I can take my own conclusions) out there.

EDIT: typos and missing prepositions, lol.
 
Last edited:
You must log in or register to reply here.
Back
Top