Page 5 of 13

Re: FAHBench (OpenMM 5.1)

Posted: Thu Jun 06, 2013 2:26 am
by mdk777
don't disagree.
but in the OP, he did phrase it as an open question for PG to comment on.

They have not, so he is just refining his calculations as best he can with the limited information he has. :wink:

Re: FAHBench (OpenMM 5.1)

Posted: Thu Jun 06, 2013 2:40 am
by EXT64
Well - as for the 2.5x there is also some uncertainty as to what it applies to. Best case or Worst case? (We know performance can vary a lot depending on what forces and physics are turned on). Old Gromacs (pre-4.6) or new? OpenMM 5.0 or 5.1? (both have seen substantial perf increases in recent revisions). I would not recommend using the number as fact and then accusing PG of being unfair with points if the math doesn't line up (I am not saying you all are - but someone will). At this point all I would be confident in saying is that Gromacs is much faster than core 17 running with opencl on the CPU.

Re: FAHBench (OpenMM 5.1)

Posted: Thu Jun 06, 2013 6:30 pm
by Napoleon
I appreciate the healthy criticism regarding my spreadsheet, it is well deserved. I'd like to remind you about the disclaimer I put into the spreadsheet itself, though:
HOWEVER, keep in mind that ultimately
THIS SPREADSHEET IS FOR AMUSEMENT VALUE ONLY !!!
I never intended it to be gospel, nor expected anyone to take it as such. It's just a few facts a picked up here and there, then mashed them up to create one big factoid. As bruce pointed out, it merrily dismisses way too many subtle factors to be truly accurate.

Let's just say I got the urge to brush up my mad spreadsheet skillz, and call it day, shall we? I apologize for drifting off topic, which is posting OpenMM 5.1 FAHBench results instead of speculating (too much) on them.

Re: FAHBench (OpenMM 5.1)

Posted: Thu Jun 06, 2013 7:15 pm
by AndyE
Napoleon,
There was no intention to criticise your spreadsheet, I just wanted to know where the factor was coming from.
Andy

Re: FAHBench (OpenMM 5.1)

Posted: Tue Jun 11, 2013 4:53 pm
by 7im
7im wrote:Asus GT430 @ 730 MHz core, 900 MHz memory (stock)
WinXP 32-bit, Nvidia 314.22 drivers

OpenCL Explicit SP: 3.91663 ns/day
OpenCL Implicit SP: 16.9768 ns/day

Intel Core2 Duo (E8400) @ 3.0 GHz (stock)
WinXP 32-bit

Explicit SP: 0.7763 ns/day
Implicit SP: 0.929 ns/day

Got CUDA running on the GT 430, here are the fahbench numbers.

Explicit Solvent: 4.76 ns/day
Implicit Solvent: 20.6 ns/day

Re: FAHBench (OpenMM 5.1)

Posted: Wed Jun 12, 2013 6:16 pm
by PantherX
Updated the first post with Veix's information.

7im, are those results Single Precision or Double Precision?

Re: FAHBench (OpenMM 5.1)

Posted: Wed Jun 12, 2013 6:20 pm
by 7im
Sorry, GT430 only does SP.

Re: FAHBench (OpenMM 5.1)

Posted: Wed Jun 12, 2013 10:37 pm
by Napoleon
GT430 is DP capable, but it's so very, very, very slow:
Image

Re: FAHBench (OpenMM 5.1)

Posted: Wed Jun 12, 2013 11:38 pm
by 7im
Oops, didn't see that option. Here are the DP cuda numbers, after about 10% each... GT 430 again.

Explicit Solvent DP: 0.9343 ns/day
Implicit Solvent DP: 1.72 ns/day

Re: FAHBench (OpenMM 5.1)

Posted: Fri Jun 21, 2013 1:23 am
by Zagen30
I know there's already 780 results from Anandtech's review, but I figured I'd share mine anyway:

GTX 780 SC (967 MHz core that boosted to 1124 MHz during the FAHBench run, 3005 MHz memory)

OpenCL SP:

Explicit: 43.898 ns/day
Implicit: 189.014 ns/day

Re: FAHBench (OpenMM 5.1)

Posted: Tue Jun 25, 2013 4:45 pm
by Veix
AMD HD 5770 (850/1200) 1GB

OpenCL Single Precision
Explicit: 7.22122 ns/day
Implicit: 35.9727 ns/day

Re: FAHBench (OpenMM 5.1)

Posted: Wed Jul 03, 2013 9:51 pm
by AZBrandon
ASUS GTX 760 2GB (GTX760-DC2OC-2GD5) stock settings: 1006 mhz (versus stock 980mhz), 6000mhz VRAM

In FAHBench it boosts to 1175mhz. Completed a live WU without error yesterday although the car squeals, like every single GPU I've tested with FAH in the last 4 years or so. Anandtech shows the Explicit single/double at 29.3 & 4.5 which is roughly the same as my results:

Explicit single: 30.09
Explicit double: 4.57
Implicit single: 115.32
Implicit double: 6.18

Like everyone has said, it's in the same ballpark as the GTX 660 Ti.

Re: FAHBench (OpenMM 5.1)

Posted: Sun Jul 07, 2013 4:00 pm
by Kurtis200200
(Evga) GeForce GTX 660 using Nvidia's 320.49 WHQL driver
Windows 7 home premium x64 SP1
Intel Core 2 Quad (Q6600; 4 cores at 2.4GHz)
4GB of RAM

All values below were obtained using the "Stress Test" and "Verify Accuracy" options on fahbench 1.2.0

OpenCL Explicit Single: 22.7358 ns/day
OpenCL Explicit Double: 3.70971 ns/day
OpenCL Implicit Single: 97.7558 ns/day
OpenCL Implicit Double: 5.18135 ns/day
CUDA Explicit Single: 32.75 ns/day
CUDA Explicit Double: 4.53586 ns/day
CUDA Implicit Single: 104.382 ns/day
CUDA Implicit Double: 6.21154 ns/day

Re: FAHBench (OpenMM 5.1)

Posted: Tue Jul 09, 2013 12:29 am
by 7im
EVGA GeForce GTX 760 using NV 320.49 @ OC/Boost from 980 to 1175 MHz
Windows Vista Premium SP2
Intel Core 2 Duo 6400 @ 2.13 GHz
4 GB Ram
x16 slot wired as a x1 slot (this and a slow CPU hurts performance a bit)

OpenCL Explicit Single: 24.401 ns/day
OpenCL Explicit Double: 4.38031 ns/day
OpenCL Implicit Single: 101.646 ns/day
OpenCL Implicit Double: 6.11951 ns/day

Re: FAHBench (OpenMM 5.1)

Posted: Thu Jul 18, 2013 11:33 pm
by JimF
PowerColor HD 7790 @1030 MHz core, 1500 MHz memory
Win7 64-bit
AMD 13.6 Beta 2 drivers

OpenCL Explicit SP: 21.3729 ns/day
OpenCL Implicit SP: crashed