Everyone has his or her own ideas about what value and performance mean, some buyers choosing to bolster a low-end system with high-end graphics while others might inadvertently cripple performance platforms with mid-priced parts. Our most recent System Builder Marathon attempted to balance CPU and graphics performance within strict budget limits, but our most controversial build surprised everyone with its 3-way SLI set of moderately-priced performance graphics cards.
Of course, opinions run strong in our comments section. And while we appreciate the feedback and will use it to guide future System Builder Marathon configurations, we did notice a contingent that protested 3-way SLI as “excessive” for our $2,500 budget, suggesting that a pair of Radeon HD 4870 X2s might have been a better option. The primary argument was one of multi-card performance scaling, where the third card provides the least benefit.
While this first half of the argument is certainly worthy of debate, suggesting that two HD 4870 X2’s would be a better choice ignores the fact that two RV770 GPUs comprise each Radeon HD 4870 X2 card. That is to say, one cannot effectively argue against the “poor scaling” of three graphics processors excessive while simultaneously arguing in favor of four.
Tom’s Hardware is no stranger to controversy and its editors need to be prepared to defend their decisions. Knowing that selecting 3-way SLI with GeForce GTX 260 graphics cards would be far more controversial than two GTX 280 cards, we armed ourselves with two GTX 280 units in preparation of today’s SBM addendum. Let's get to the bottom of how these high-horsepower configuration stack up against each other.


I thought that the only problem was that one of the builds performed less than the last one at the same price point. But then again I am used to the comments section of such articles being full of whinners and idiots so its not much of a surprise...
285 SLI is looking interesting
This was a good review, it pointed out to us that buying 3 260s is not so beneficial, Crysis is a great game, but it is not the deciding factor for the purchase of my next video card.
If I really want SLI I would choose as fewer cards as possible. The driver mess and inefficient scaling makes no sense to get more than two cards. GTX 295 > GTX 280 x2 > GTX 260 x3 > 9800GT x4.
A 4870 X2 costs around $500. A GTX 280 $330. $500 x 2 = $1000 > $330 x 2 = $660 so what extra funds are you talking about?
Better to notice the fact the 280x2 consistently beats the 260x3 except in the one game that sets the standards for erratic performance results.
People are always gonna critisize choices made in these articles, and make assumptions grabbed out of thin air without any form of proof or facts to back up their statements, imo this article and the 2500$ build made some really interesting and compelling points and pointed out the advantages and disadvantages of both.
Of course I would love to see a 2 x 260 SLI vs. 3 x 260 SLI to see how much that last card really gives, since that's what we are really talking about how much a third GPU really stacks with the others. But since THG already has the GPU charts (http://www.tomshardware.com/charts/graphics-cards,1.html) which I assume/hope is getting a updated very soon based on a Core I7 machine, all these speculations will basically be answered by this.
This article is about a 2500$ build and what solution is best for this, and as seen in this article the best decision when going for playable framerates is trip SLI core 216 260's, not dual 280's which only boost your already enormous frame rates on other games.
That would be interesting. I know the scaling improved with the i7 for SLI but the comparison Vixe is talking about would be very telling.
Well. It's one of those few well done games that actually benefit from multible GPU's and multible CPU's. I have a high hope from the Futuremark Games Studios Shattered Horison http://www.shatteredhorizon.com/ because they have allso known how to utilice multible cores...
And yoeh you are right that Crysis is not so good game, but as I mentioned, it is one of those few that scale with multicores guite well, so I think that that is the main reason why it reamains as one of the standard test games. I hope that in the future there will be much more games that can really use all that power that you have in your computer (read scale well). It does not make a good or bad, but tells asomething about the programing.
I know, I know... that doesnt give you a good idea of the performance you can expect for the same dollar spent on a 2-way vs a 3-way. And I understand that the point of this article was to deal with that question posed by the readers and thier comments. Based on that I say...
You should be comparing one of two different scenarios...
1) Performance level per dollar spent. i.e. Spend the exact same (or as close as possible) amount on a 2-way and a 3-way SLI or Crossfire setup. that give you the most performance for your purchase and compare them. For this comparison it does not make any difference if the cards have 2 GPUs or 1 GPU. It is about the performance you can squeeze out of 2 Cards vs 3 Cards... not out of 2 GPUs vs 3 GPUs.
2) Cost/Performance difference in 2-way vs 3-way SLI. This may sound like the same thing but it is not. This does not compare the best 2-way you can buy for the same amount as a 3-way. This is taking 3 identical cards. Test performance with all three cards. Test performance with only two cards... and if you really want to do the world a favor, test it with only one card as well. What do you get? A scale that show the amount of performance increase as it scales in proportion to the cost of each card.
Just my $.02,
TechDicky
Which win matters in game play? You could have 100 games that play at 90FPS on a cheaper setup or 100FPS on the more expensive one, and the difference would be inconsequential since both of them play smoothly. Same goes for comparing RTS games, which don't need to be perfectly smooth, and FPS games, were any lag gets you fragged. It's always the FPS games that fall below 40FPS that take precidence.
That kind of comment shows people that don't take RTS seriously... having a stutter of 0.1 second is enough to break havoc on any carefully planned strategy during an RTS game.
Any "good" RTS player will need to execute between 5 to 10 actions per second (300-600 APM) during the key moments of high. And a stutter of 0.1 or 0.2 seconds those conditions means a dead army. Meaning Frames should NEVER EVER drop below 30 FPS, the same applies to First person Shooters.