Benchmark Scene, Check your GPU power!

[igen]

I try to make collect benchmark list from member here for more easier to compare between VGA, hope this forum have benchmark page

[Korsa]

Benchmark results have been transferred into the table. We collected all results since August 1. We didn't use results until that date due to constant speed improvements. Please, have a look at the first page.

[Chakib]

Benchmark results has been transferred into the table. Please, have a look at the first page.

Thanks for the table !
Hmm my GTX970's 16:45 yesterday test (oc to 110%) is faster than 2xGTX780, good to know even with that nasty 3Go vram limit.
I guess I'll go for the GTX1070 cause 1060 is just not worth it at least for me.

[Sylk]

don't show any table :/

Edit: about igen post

[igen]

I know my table not real right to compare until stable but at least I can more easier to make decision which VGA I will choose for upgrade

2x GTX 1070 still more faster then 2x GTX 1080??

[Sylk]

What the meaning of the gray lines in table ?

Edit: oh ok i see, gray for multiGPU.

[RobSteady]

05:36

FStorm 0.2.6e
2 x 980Ti + 1 x Titan Z

This version seems to be a little bit slower than previous ones (05:11).

[adams]

12:37

EVGA GTX1070 FTW (2138mhz core, 4398mhz mem)
FStormRender v0.2.6a

The table times column should be relevant to a version number of FStormRender as a lot of variation version number to render speed.

[gabrielefx]

3:13
4xGTX1080s Power limit 120%

3:11
4XGTX1080s Power limit 120% + core clock 200Mhz

[Sylk]

OC is clearly useless. It just eat up your GPU.

[adams]

Shows on multi gpu setup something bottle necks system hardware or code loses efficiency or both, as no real change 1070 vs 1080 multi GPU speed times.
1 gpu I reduce render time by nearly 1.5m by OC.
Would be interesting experiment see where the point of diminishing returns is, eg. Someone with multiple gpu sysytem, test time 1 gpu OC and no OC, then 2 the same , then 3 same ,then 4 same.

[Sylk]

You right Adams.
I did an OC benchmark. Gain 3min20s (29:51 > 26.33).

See my next post below.

[RobSteady]

4 x 1080 = 8GB = 3160 € = 3:13
4 x 1070 = 8GB = 2000 € = 3:25

mmhhh...

[Karba]

4 x 1080 = 8GB = 3160 € = 3:13
4 x 1070 = 8GB = 2000 € = 3:25

mmhhh...

Yes, this is strange.

[mitviz]

4 x 1080 = 8GB = 3160 € = 3:13
4 x 1070 = 8GB = 2000 € = 3:25

mmhhh...

very strange, not possible

[antonyebl]

doesn't make sense relative to CUDA cores count, even if one disregards slightly higher base clock speed of gtx1080;

4x 2560 cores (gtx1080)
vs
4x 1960 cores (gtx1070)

in those terms 1080 should be ~1.3x faster (which +- holds for single 1080 vs 1070 i.e. 13min vs 11min >> ~1.2x)

just ran the benchmark again with the 0.2.6e, stock gpu clock @1531Mhz but while rendering card auto-boosts to 1898Mhz, time about the same as with previous benchmarked release (~13min)

[deC9r]

My 1070 runs at stock which is app. 1500/4000mhz 13-14mins and oced 2100/4400mhz around 10-11mins.
There is something limiting the 1080, probably it's a combination of gpu clockspeed and the new vram controller for GDDR5X.