By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Tachikoma on 19 February 2013
Edit Thread
disolitude said:
Tachikoma said:
disolitude said:
CGI-Quality said:

So, all-in-all, a 60% jump over the 680, but also behind the, still, best-of-the-NVIDIA-best GTX 690.

The prospect of tri SLI for these cards makes these more appealing to me than a 690. By itself 690 isn't enough to drive 3 screens. It's no secret that 690 quad SLI doesn't scale well but tri SLI witht hese cards apears to give very impressive scaling. With these cards in tri SLI, 3D vision surround @60 fps per eye or 120 fps surround may be possible...


?I'd estimate quad doesn't scale because of CPU bottlenecks, not GPU's fault really - I'm running three 3gb GTX580's in tri-SLI and for quite a lot of stuff, games were faster with only two cards, because the added cpu cycles of running a third card impacts the cpu enough to cause the speed to actually drop below that of the GTX580 in situations where the GPU's were running at the maximum capacity of the supporting CPU.

ive been posting on nvidias forums for 5 years now and quad gpu has never scaled well and not because of a cpu bottleneck. its fine for synthetic benchmarks but drivers dont work properly for most games. check out some quad benchmarks on the net. some games have lower frame rates for quad sli vs 2 way sli.

The problem with that is when someone does quad SLI, they do it with high end cards, or at least pretty-damn-good ones, so the difference between performance gain from 2-way to 3-way or 4-way is much less than the jump from single to 2-way, CPU bottlenecking will always be a contributing factor, as the more GPU's you add the higher the load on the CPU just to run them, more cycles needed to support the GPUs means cpu runs slower for everything else (the game), impacting the performance - that is why synthetic benchmarks that measure the GPU alone show much better results than 'real-world' games.

Putting it bluntly, it is actually a CPU bottleneck, that's why the performance gain of high end cards beyond standard sli is not as high - take older cards and SLI them with the very latest CPU and you'll notice the scaling is much, much better than when the cards were originally reviewed - only problem is review sites generally don't go and re-test older GPU's with newer CPU's to check this - It is however the case.

Of course, the solution is to run the games at much higher resolutions than the standard 1080p/1600x1200 reviewers usually use, but then you find yourself being bottlenecked by the video memory for extreme resolutions or multi-monitor setups.

Take this for example : http://www.tomshardware.com/reviews/geforce-gtx-480-3-way-sli-crossfire,2622-11.html

What we see here is extremely high resolution and settings allowing the third GPU to make a much bigger difference using AA, when you tax the GPU even more with 4xMSAA, the jump in power is even higher in the switch to 2-way as it is from single to 2-way, this is because in these situations the GPU is running at it's full capacity and not being held back by the CPU, in lower resolutions the difference between SLI and tri-sli is much, much less because the maximum performance is being held back by the CPU.