Multi-gpu not really faster or???

[jstrob]

Hi,

One thing I don't understand in the v-ray benchmark results is that a single 1080 ti is almost as fast as 4 ou even 8 1080 ti. Could it be a mistake in the listing or is V-Ray not able to use all gpu on some systems?

see: https://benchmark.chaosgroup.com/gpu...string=1080+ti

21 sec for a single 1080 ti at the bottom... then 21 for 4x1080 ti.

Then some x4 1080 ti gives 17.774 sec while x8 1080 ti gives only 16,6...

I just bought 8 1080 ti and I'm wondering if I should put them all in the same system (I have 6 gpu slot but was thinking using pci-e splitters) or distribute them in many system and use DR. I would like to know before ordering another power supply and the risers etc.

3	00:16.601	Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz x40, GeForce GTX 1080 Ti 11264MB x8	1
4	00:16.669	Intel(R) Core(TM) i7-6950X CPU @ 3.00GHz x20, GeForce GTX 1080 Ti 11264MB, TITAN X (Pascal) 12288MB	1
5	00:17.592	AMD Ryzen 7 1700X Eight-Core Processor x16, GeForce GTX 1080 Ti 11264MB, Quadro GP100 16291MB x2	1
6	00:17.774	Intel(R) Core(TM) i7-5930K CPU @ 3.50GHz x12, GeForce GTX 1080 Ti 11264MB x4	1
7	00:17.926	Intel(R) Core(TM) i7-6950X CPU @ 3.00GHz x20, GeForce GTX 1080 Ti 11264MB x3, TITAN X (Pascal) 12288MB	1
8	00:18.652	Intel(R) Core(TM) i7-5820K CPU @ 3.30GHz x12, GeForce GTX 1080 8192MB, GeForce GTX 1080 Ti 11264MB x2, GeForce GTX 980 Ti 6144MB, GeForce GTX TITAN X 12288MB x2, TITAN X (Pascal) 12288MB x2	1
9	00:19.194	Intel(R) Core(TM) i7-5960X CPU @ 3.00GHz x16, GeForce GTX 1080 Ti 11264MB x4	1
10	00:19.513	Intel(R) Core(TM) i7-6950X CPU @ 3.00GHz x20, GeForce GTX 1080 Ti 11264MB x2, TITAN X (Pascal) 12288MB x2	1
11	00:19.669	Genuine Intel(R) CPU @ 2.80GHz x32, GeForce GTX 1080 Ti 11264MB x4	1
12	00:19.856	Intel(R) Core(TM) i7-6900K CPU @ 3.20GHz x16, GeForce GTX 1080 Ti 11264MB x4	1
13	00:19.974	Intel(R) Core(TM) i7-6950X CPU @ 3.00GHz x20, GeForce GTX 1080 Ti 11264MB x4	1
14	00:20.831	Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz x56, GeForce GTX 1080 Ti 11264MB x4	1
15	00:21.132	Intel(R) Xeon(R) CPU E5-2687W v4 @ 3.00GHz x48, GeForce GTX 1080 Ti 11264MB x4	1
16	00:21.613	Intel(R) Xeon(R) CPU E5-2683 v3 @ 2.00GHz x56, GeForce GTX 1080 Ti 11264MB	3

[yavor.rubenov]

Hello,

The results at 16th place seem strange a bit - two of the 3 results have comments and the comments mention Quad 1080Ti - so I would guess that there is some issue with the results display.

Best regards,
Yavor

[jstrob]

Ok yes I didn't know we could read comments by hovering mouse over the text bubble. So they are four GPU. I'm still surprised that 8 gpu only gain 1 second but maybe with a more difficult scene it would give a bigger difference.

[varosi]

We had a problem with 1.0.5 version of V-Ray Benchmark and it didn't provided information about GPUs count. This is fixed in 1.0.6. Database at https://benchmark.chaosgroup.com is fixed, too.
So now you may see correct results for multi-GPU configurations. Please, check again.

[KhyeKading]

Hi Strob, I was noticing that too. I'm wondering if at that point the PCIe speed is diminishing enough (probably to x1/x2) to support that number of cards. In a 4 way setup you're maintaining at least x8/x8/x8/x8.

[jstrob]

It may be a problem due to configuration in the bios and or windows related problem. My guess is that we son't have that problem with Linux. Righ now I have only 2 cards in the same system but I have a total of 8 cards that I will try to test test all in the system soon and will compare it to putting the same cards inside of many different system using DR.

[KhyeKading]

It may be a problem due to configuration in the bios and or windows related problem. My guess is that we son't have that problem with Linux. Righ now I have only 2 cards in the same system but I have a total of 8 cards that I will try to test test all in the system soon and will compare it to putting the same cards inside of many different system using DR.

That's very possible. From my internet travels I've seen instructions about hacking registries and bios to get more than 4 cards recognized. Please let us know how your 5+ card solution turns out.

[JezUK]

Glad I've caught this early on - but am wondering does the fault lie in the benchmark software, or is it truly an issue with hardware / bios etc limitations being hit ?

From memory, I was seeing scalable results both from the earlier versions of the benchmark software (jumping from 2 to 4 1080 ti's) and with my own project scenes.

I've just built an open rig frame and purchased all the risers etc so will be following this thread carefully before making any further hardware purchases (my mobo can support up to 7 GPU's).

[savage309]

Usually V-Ray GPU scales okay with multiple GPUs. Could be V-Ray Bench thing. Will check.

Best,
Blago.

[jstrob]

Usually V-Ray GPU scales okay with multiple GPUs. Could be V-Ray Bench thing. Will check.

Best,
Blago.

Cool to know.

And JezUK, cool your open rig! I considered those handy aluminium extrusion and then I remembered I had some electrical panels (kind of metal boxes) lying in my shed. I am building a 8 gpu rig in an enclosed case with those panels. Also using risers including a special one that split one pci into 4.