MultiGPU Update: Two-GPU Options in Depth
by Derek Wilson on February 23, 2009 7:30 AM EST- Posted in
- GPUs
Who Scales: How Much?
To calculate this scaling data, we simply looked at percent performance improvement of two cards over one. With perfect scaling we would see 100%, while no improvement is 0% and a negative performance improvement means that the multiGPU solution actually produced worse numbers than the single card. There's a lot of data here, so we'll break it down a bit before we present it all.
It is possible to see more than 100% scaling in some tests for different reasons. Fluctuations in benchmark performance can contribute to just over 100% situations, and some times optimizations to enable better multiGPU performance can cut some work out enabling higher performance than would otherwise have been possible. In one of the cases we test today we have a situation where single GPU performance is limited at some framerate while multiple GPUs aren't hindered by the same limit. This artificially inflates the scaling percent.
When looking at games that scale overall, we end up seeing both Radeon HD 4870 configurations (512MB and 1GB) performing worse than we expected. Granted, the 4870 1GB looks better if we only take 2560x1600 into account, but even then the Radeon HD 4850, GeForce GTX 260 and GTX 280 beat out the 4870 1GB in terms of average performance improvement (when performance improves). When we add in CPU limited cases, the 4870 cards look even worse. Consistently, most of the ways we attempted to analyze the magnitude of performance improvement (averages, geometric means, per game, across games where call cards scaled, etc.), the Radeon HD 4850 and GeForce GTX 260 (and sometimes the GTX 280) did pretty well, while the Radeon HD 4870 cards came in pretty low on the list with the 1GB often looking worse because it hit harder CPU limits at lower resolutions.
Hitting CPU or system limits does speak more to value than desirability from a performance standpoint, but it's still important to look at all the cases. Configurations with lower baseline single GPU performance will have more headroom to scale, but these might not always scale enough to be playable even if they scale well. So it's important to take both value and absolute performance data into account when looking at scaling.
We've put all this data on our benchmark pages with the performance data to make it easier to see in context. There just isn't one good way to aggregate the data or we would talk about it here. Depending on the type of analysis we try to do, we could present it in ways that favor AMD and NVIDIA and since there really isn't a "correct" way to do it we've decided to just present the data per game and leave it at that.
Facebook
Twitter
RSS
95 Comments
View All Comments
makdaddy626 - Monday, February 23, 2009 - link
I really appreciate the article and all the research and work that went into it. Kudos to you for it.A small suggestion would be to take into account a minimal playable frame rate in the value and performance per dollar data, where a ZERO would be substituted for the frame rate in instances where a card failed to reach a playable rate for a given game/resolution. I feel this would more accurately reflect the value of the card(s) as 15 FPS in most games presents no value.
Mastakilla - Monday, February 23, 2009 - link
I agreeMinimum framerates should be more important then average ones even...
Interesting article though! I didn't know the 4850x2 was so competitive...
Only in Crysis it does worse the the 285, which I had in mind for my new pc...
That does make me wonder if the 285 might be a more future proof investment...
makdaddy626 - Monday, February 23, 2009 - link
Yes, But I meant "minimum" in the sense of what the game needs to be played at, even you measure "average" - I just don't think it's fair to say that the 9800 GTX shows the highest performance per dollar on the Crysis Warhead 2560x1600 chart when it turns in frame rates of 13.5. To me that is ZERO value for the money because it's not playable. Someone wanting to play at those settings would be wasting EVERY dollar spent on the 9800 GTX.7Enigma - Tuesday, February 24, 2009 - link
Completely agree. Statistics mean nothing when not taken in a proper context. Zero, NA, or just leaving it blank would be much better. Someone looking to use that card would then click on a lower resolution and see if it is a viable choice. It would reduce the amount of data that needed to be compared as the reader of the article, and make caveats like in the explaination section about comparing between cards/resolutions etc. almost moot.Spivonious - Monday, February 23, 2009 - link
The framerate charts are all but worthless if you're focusing on how performance scales. Why not some line graphs with all three resolutions shown and card models along the x-axis? Then the reader could see how performance is affected by the lower memory of some models and how it scales linearly with the higher-end cards.7Enigma - Monday, February 23, 2009 - link
I would have to agree with this. I always enjoyed the broken line graphs that show multiple resolutions and framerates in the same graph. It makes comparisons very easy and more importantly allows EVERYONE to see their particular resolution without having to click on a link for a separate graph.It's fine to keep your specialized performance/$ and % increase from a single card the way you have it as I understand what you mean about not comparing between resolutions but for the general frame rate comparisons I preferred the old way.
One thing I failed to see which I have seen in previous reviews with X-fire/SLI (or was it with tri/quad setups?) is the stuttering that can be present. I thought it was an Anand article but could have been from another site.
SiliconDoc - Wednesday, March 18, 2009 - link
The charts are designed to autopop to 2650x - and we all know the red ragers have screamed the ati cards are great there.EVERY CHART pops to the favored ati $2,000.00 monitor resolution.
It's not an accident, Derek made them.
C'DaleRider - Monday, February 23, 2009 - link
Derek, if you want to impress, and this article does with its research, please invest in some writing manuals and learn some grammar.Take this sentence:
"This unique card really shined and held it's own all the way up to 2560x1600."
Your use of "IT'S" in this instance is incorrect. IT'S is a contraction for IT IS, not a possessive word, which is ITS.
Or take this passage, "It's very surprising to us that AMD hasn't pushed this configuration and that Sapphire are the only manufacturer to have put one of these out there."
Sapphire is a company or organization, I realize that. But in this instance, you're referring to a group in its singular fashion, or as a single unit. That context is seen by the only manufacturer in the sentence.
That sentence should have read. "It's very surprising to us that AMD hasn't pushed this configuration and that Sapphire IS the only manufacturer to have put one of these out there."
Here's the rule for that (taken from both MLA and APA handbooks): If the action of the verb is on the group as a whole, treat the noun as a singular noun. If the action of the verb in on members of the group as individuals, treat the noun as a plural noun.
This means companies, such as Microsoft, IBM, Sapphire, Ford, etc., when being referred to the company as a whole collective, single entity, has to have a singular verb.
But, if you are referring to pieces of the whole, such as "the engineers of Ford are.....", or "The programmmers at Microsoft are.....".
Please invest in some proper English grammar texts and take time to read and learn from them. Your error laden grammar you write and use is quite distracting and detracts from what is supposed to be a professionally run hardware site.
Hire a proofreader or good copy editor if learning proper grammar is too difficult.
Otherwise, this was a great article!
The0ne - Monday, February 23, 2009 - link
I don't really mind Anandtech articles as much in terms of presentation, spelling and graphics. Other sites such as Ars Technica, x-bit labs, and so forth are much worst. The first is first since they've started writing articles concerning everything, it seems.If I did mind, I say stick to the general guidelines writing manuals, procedures, pamphlets, technical docs, etc. But being online, this isn't the case and won't ever be. Again, I don't mind as much because I do the same thing myself where I hardly pay attention to spelling or grammar when writing online. It's only when I write short stories or for work that I pay attention. Strange but comfortsure does make one do these things :)
And yes, I do write all sorts of articles daily.
oldscotch - Monday, February 23, 2009 - link
"Your error laden grammar you write and use is quite distracting and detracts from what is supposed to be a professionally run hardware site."That should read "The error laden grammar you use is quite distracting..." or just "Your error laden grammer is quite distracting..."
"Your error laden grammam you write and use..." is redundant.
Perhaps you should learn some grammar yourself before criticizing others about theirs.