How can it be possible to go from just over 9000 points in 3Dmark (13) to this - http://www.3dmark.com/3dm/1182124? When dropping from X16 X16 SLI to X16 X4? Could it be my newer (and faster) SSD coming into play?
I wonder if it's cooling - you've separated the cards, no? The SSD may have something to do with it too I guess - although it would surely have limited impact?
Extra cooling would mean potentially higher and more frequent GPU boosts, plus potentially less CPU throttling - both could make a big difference in the benchmark.
With the cards packed together (well, this close...) And with the duct fan at 50% I was getting temps of 83c max on the top card. I wasn't best pleased about that (I thought Kepler were cool cucumbers) so I added these, then turned the duct fan down to 20% so I couldn't hear it at all, then I was seeing 79C top card. Oddly anything over 40% on the actual card fans (pair of 80mm or something) made no difference, so I was able to tweak everything so that it was quite well behaved sound wise.. Now running them like this. Has made literally no difference to that. In fact, really heavy stuff will make the cards run to 80c still. As mentioned yes, I've had to run the cards X16 X4 as it was the only way to fit the Revodrive and to do so I had to hack using HyperSLI as SLI does not run on anything less than X8 lanes. So yeah, only difference apart from the 'massive' (he says without having a clue what he's talking about really) drop in bandwidth the only difference is the Revodrive. Both runs were done on a freshly installed version of Win 8, here is the X16 X16 run. 3Dmark 11 has dropped.. Here is the X16 X16 run. And the X16 X4 run. http://www.3dmark.com/3dm11/7121909 Though let's face it, it's not exactly hurt anything and the power on offer is still pretty barbaric. Which makes me wonder why Nvidia are so bloody shirty (no offence bra ) about allowing SLI to run on X4. I mean yeah, quad SLI Titan ? I can see it causing issues. But why rob yourself of extra GPU sales over 300 points in a benchmark :S
Nah it's about right dude. Is a Titan, that I would bet is overclocked, with a massively superior CPU to mine. I can't overclock my cards any more than they are stock because one of them doesn't like it. I can't even get 40mhz extra out of that card.
Whilst i'm not an expert on 3D Mark, the only way that i can see that a SSD (even vs a HDD) would make any difference is on loading times - so, unless it's dumping textures & whatnot out of the gfx cards' memory & loading new ones in the middle of each individual test, i can see no conceivable reason why changing the SSD could cause a higher score with a b/m that's only testing gfx card & cpu performance. Otherwise, there's a margin of error between runs with any testing & if you've updated the drivers then this could easily cover the variance. Oh, & it's always possible that you had something running in the background on the previous run that wasn't on this latest one - so you had more free cpu resources this time around.
I've run it obsessively and I never scored that high. Best I got before was around 9100. You may be onto something though.. It may be texture streaming to the virtual paging file due to a lack of vram.
As said, i'm not an expert on 3DMark but, whether it's swapping as you're saying or loading new textures (& whatnot) from the source files, that's the only way that the SSD could sensibly have an effect... [OT it'd be sequential r/ws that'd be giving the boost here 'if' that's the cause.] Otherwise, the last part of my thinking was that i assumed you did a fresh install on the Revo... Hence there being fewer things installed thus far & so, at least potentially, fewer background processes running.
http://www.3dmark.com/3dm11/7122246 ^^ single GTX 670 on 3dmark 11 (for comparison) http://www.3dmark.com/fs/785261 ^^ 3dmark firestrike
My guess is increased boost speed. Nvidia cards can be very fussy about it, and the extra space may be just enough for it to keep a high boost. The SSD is unlikely to be a factor. With 8GB of RAM, all of the textures will be cached in memory anyway.
Well after studying the results earlier the major improvement was the combined result. So I can only imagine that uses the hard drive some how :S Catzilla has dropped from 15300 to 13800 which is about the result you would expect.
If its PCI-E 3.0 x4, no difference. If it's PCI-E 2.0 x4, expect at most a 10% drop in bandwidth. Because I run 3 PCI-E cards (A GT210, a 670 and a X-Fi PCI-E) my 670 runs sadly at x4 speeds. Speaking of which I need to change that...
