Benchmark Scene, Check your GPU power!

[RobSteady]

Really strange how close together the 1070 and 1080 are. Right now it seems that you can save a lot of money to get identical results...
Is there room left for optimizing FStorm for the Pascal Cards?

[m17a4y]

Really strange how close together the 1070 and 1080 are. Right now it seems that you can save a lot of money to get identical results...
Is there room left for optimizing FStorm for the Pascal Cards?

The results wont stay forever close like now, so no, you won't save money and get identical results :S

[m17a4y]

I feel like the gtx 1080 is not max optimized yet
*i might be wrong tho*

[RobSteady]

"I feel like the gtx 1080 is not max optimized yet"
That's why I was asking in the first place...

[adams]

If you look at any reviews, benchmarks, synthetic or games there is a clear difference in performance between 1070 and 1080, but not huge. Maybe in compute, upto 20% 1080 faster, generally less 15-18%. There is that variation of speed alone in the range of 1070 results on benchmark. I think it's not too far from sounding correct (times we get 1070v1080), and other things have influence benchmark/render times: hardware/system/apps/clocks etc.

[mitviz]

well so far on facebook the user who posted the similar time to the 1080 he had the 1070 ran the test again and got less time than the 1080, which made sense so maybe its something with fstorm and individual pcs

[AnconeusRex]

4:01:35 with the Quadro K2000m
Killin' it

[RobSteady]

137:52:12 on GeForce 2 Ultra

[Alexevp]

[deC9r]

Gainward Phoenix GTX1070 @ 2101Mhz / 4399Mhz
Rendertime: 11:38
Fstorm 0.3.1 / Driver 373.06

Gainward Phoenix GTX1070 @ 2101Mhz / 4399Mhz
Rendertime: 11:10
Fstorm 0.2.5 / Driver 372.54

Gainward Phoenix GTX1070 @ 2101Mhz / 4399Mhz
Rendertime: 10:44
Fstorm 0.2.4 / Driver 372.54

Gainward Phoenix GTX1070 @ 2101Mhz / 4399Mhz
Rendertime: 11:55
Fstorm 0.2.1b / Driver 368.81

[deC9r]

Temps are 42°C

[mitviz]

42 c is probaly at idle

[deC9r]

No! It's 42°C at 100% while rendering with fstorm while 25°C room temperature. In games it's max. 59°C. The card has a tripple slot cooling solution / best aftermarket cooling solution which is currently available aside from watercooling.

[PeP]

GTX 1080 + GTX 1070 + 0.2.6a = 6 min 20 segs

Now i have to test each one

[wastzzz]

2x gigabyte 1070 G1
6min 14sec

[paoloverona]

tested again with new version of Fstorm, near 7% speed drop:

pre V3: 13min 14secs with one Asus GTX 980Ti Strix (no overclocked)
with V3: 14min 08sec

[jesuskroma]

Hello everyone, my 2x GTX 970 SLI managed to render time of 9:50min. I'm very impressed.

[deC9r]

Gainward Phoenix GTX1070 @ 2101Mhz / 4399Mhz
Rendertime: 11:15
Fstorm 0.3.3 / Driver 373.06

Gainward Phoenix GTX1070 @ 2101Mhz / 4399Mhz
Rendertime: 11:38
Fstorm 0.3.1 / Driver 373.06

Gainward Phoenix GTX1070 @ 2101Mhz / 4399Mhz
Rendertime: 11:10
Fstorm 0.2.5 / Driver 372.54

Gainward Phoenix GTX1070 @ 2101Mhz / 4399Mhz
Rendertime: 10:44
Fstorm 0.2.4 / Driver 372.54

Gainward Phoenix GTX1070 @ 2101Mhz / 4399Mhz
Rendertime: 11:55
Fstorm 0.2.1b / Driver 368.81

[deC9r]

Nice speed improvent you did there!

[m17a4y]

Gainward Phoenix GTX1070 @ 2101Mhz / 4399Mhz
Rendertime: 11:15
Fstorm 0.3.3 / Driver 373.06

Asus ROG STRIX GTX1080 A8G @ 2050Mhz
Rendertime: 11:28
Fstorm 0.3.3 / Driver 373.06

[m17a4y]

Asus ROG STRIX GTX1080 A8G @ 2088Mhz / 4959Mhz
Rendertime: 11:07
Fstorm 0.3.3 / Driver 373.06

[mitviz]

dam its really not worth getting the 1080

[cdb]

dec9r, look at my previous page post

i got 3.00 with 4x titanx pascal fs 0.2.6a production and active shade both the same times lastest driver
4.06min with 4x980ti maxwell fs 0.2.4 production and 7.57 same config on activeshade


it's quite fast, however i was expecting a HUGE difference between the two generation..not only one minute

also 4xtitan is 20 secs faster than the 1070 but with almost the double of cuda cores and larger bandwidth, clock and whatnot

[deC9r]

Pascal is just an overclocked maxwell with a new memory controller... Btw 3 minutes with 4 new pascal cards is rather pathetic. I guess they are running at stock which is quite low.

Clockspeed > Cuda cores at least for pascal and fstorm.

I bet 4 gtx1070 running at 2100 / 4400 would beat or be as fast as 4 titan x.

Ttitan x stock -> 10,9 TFLOP | 3584 Cudacores | 10 Gbps | 480 GB/s
Gtx 1070 @ 2100/4400 -> 8,06 TFLOP | 1920 Cudacores | 8,8 Gbps | 297GB/s

Therefore I would say the Titan gets bootlenecked by the memory bandwith which should have been at least 512 GB/s in the first place. Yeah I know, they added new compression but obviously it's not as good as they thought for computing.

Btw I used the boost clock to calculate the TFLOPs. If the Titan can't hold 1531mhz through a render, TFLOP would be lower...

[cdb]

we should start a new thread for that, i'll post it later or tomorrow with new stats as well such as clocks etc

[cadmium]

I ran the benchmark on my older gpus, and I found something strange:

gtx 750ti: 48min
gtx 680: 57 min

All with the latest drivers and 3.4 fstorm version.
The gtx 680 is supposed to be a lot faster, than the 750ti!
Running any other benchmarks on my systems I get that the 680 is at least 1.5 times as fast!

What's going on?

Also it seems like the pascal cards have some performance issues. (1070vs1080)
Karba, could you look into this!

I saw a 780ti here with 48min which is the same as my 750ti!

I can imagine you're quite busy.., but there could be significant performance gains!

[Miguel1900]

Another GTX 1060 SC: 17min 53sec stock clocks @ 60ºC 50% fan speed. FStorm v0.3.5

However memory clock downclock by itself from 2003mhz to 1901mhz. Maybe memory clocks are not as important as core clocks, but maybe there could be a little improvements. (Memory clocks only downclock here. Even in graphic stress tests reaching 75ºC it keeps 2003mhz)

PS: Maybe we should use the same version of FStorm when doing the benchmark.