Help me understand Firestrike results: graphics score vs. physics

I have built a desktop out of new and ebay parts. The goal was to build a power desktop for gaming and general use with breaking the bank (specifically, MY bank!)

Gigabyte X79-UD3
Xeon E5-1650 V1
4x 8GB DDR3-2133

The system started out with a GTX 690 with is basically two 680 GPUs on one physical card. I added a second 690 later. I liked the performance of the two 690s in games that supported SLI, but not all games did. The 2GB VRAM per GPU was limiting as well.

With this system I am seeing some interesting Firestrike graphics and physics scores with Quad SLI (two 690s) vs. the single 1070. Can someone explain to me the what these differences mean?

2x GTX 690
Graphics Score 26,511
Physics Score 10,274
Combined Score 4,915
FS Score 15, 813
http://www.3dmark.com/fs/10918817

1x GTX 1070
Graphics Score 19,244
Physics Score 16,019
Combined Score 7,858
FS Score 16,376
http://www.3dmark.com/fs/10991810

 

Why is your xeon at 5.6GHz? Is that real?

Physics is usually heavily cpu dependant not gpu, so your results are kinda weird.

Graphics the quad gpu set up should win. I'm not 100% sure why the 1070 has such a higher physics score. It could be because it's newer, sli or vram could have been hurting it or something too not sure.

 

The Xeon was running at 4.5GHz at the time. I saw it reported as 5.6GHz as well and no, it is not real. I wish!

 

I'm thinking your quad sli is eating into your Physics score, maybe there's a large CPU overhead to power quad sli. Impressive GPU score for the quad sli, but in benchmarks you get the best scaling & from what I've read quad sli support in gaming is woeful! Also, that 2GB VRAM for the GTX 690's will really hurt your gaming performance, new games just need more than 2GB of VRAM - those cards are really dated now - the single GTX 1070 is worlds away a better solution for today's gaming! I would sell the 690's and stick with the GTX 1070 that you have.

 

Your comments about the 690s in Quad are spot on. I saw minimal improvements in WoW. It crushed MechWarrior Online. The 2GB VRAM required me to lower texture quality or it would stutter badly. Fun experiment though.

Physics is primarily a function of the CPU?

 

Yes, that's right, the 3DMark physics tests hammer the CPU while keeping GPU load at quite low levels - but maybe there's a constant baseline CPU overhead required to drive quad sli even at low GPU loads. I mean I've not googled it, but to me it seems possible.

 

I googled Firestrike graphics and physics scores, but not "does quad SLI take more CPU". I just did this and found this thread talking about GTX 680's in SLI. I am going to check this by putting the Quad SLI in another motherboard to test.
https://forums.geforce.com/default/topic/528038/sli-and-cpu-combination/

I bought the 690 on a lark. Then I found another at a great price so I said, "Why not?" I can get most if not all of my money back on these.

I bet Quad SLI would run better with something like an i7-6700K due to the better IPC, but I only have $150 in the motherboard and CPU.

 

Well, be interesting to see what you find with your testing!

 

When you updated your CPU with the GTX 1070, your physics score increased. The Physics score is basically the cpu score.

 

We stuffed the two GTX 690's in my son's machine which has a 6700K @ 4.6GHz. We couldn't get the same overclock on the 690's for some reason. The 6700K beat the E5-1650 physics by a good margin.

Firestrike 16,372
http://www.3dmark.com/fs/11939942

 

We were talking about your Physics score before weren't we, and postulating that quad sli may lower it? You haven't compared like for like on your sons machine, but your Physics score is lower than mine, mine's about 9% higher at 4.6Ghz, but I do have faster RAM which I've lowered the timings a lot too. Maybe quad sli does create a CPU overhead even in the Physics test.

 

Yeah, he is wanting faster RAM.