FAHBench (OpenMM 5.1)

[toTOW]

Results from my GTX 980M

OpenCL - DP:
Explicit: 3.50513 ns/day
Implicit: 7.30925 ns/day

OpenCL - SP:
Explicit: 50.727 ns/day - GPU is only used at 80%
Implicit: 216.631 ns/day

CUDA test is not working : Error loading library D:\FAHBench/OpenMMCUDA.dll: 126

[bruce]

It works for me with FAHBench_1_2_0

[7im]

toTOW, which version of the CUDA SDK do you have installed?

[toTOW]

None ... it's been a long time since the runtimes have been included in regular drivers (both for ATI and NV). And before, they were packaged with the client/program that needed them.

From BOINC client point of view :

21-May-2015 08:19:22 [---] CUDA: NVIDIA GPU 0: GeForce GTX 980M (driver version 352.86, CUDA version 7.5, compute capability 5.2, 4096MB, 3066MB available, 3461 GFLOPS peak)
21-May-2015 08:19:22 [---] OpenCL: NVIDIA GPU 0: GeForce GTX 980M (driver version 352.86, device version OpenCL 1.2 CUDA, 4096MB, 3066MB available, 3461 GFLOPS peak)

From FAH client point of view (I didn't notice that OpenCL detection was missing in FAH client) :

15:49:09: GPU 0: NVIDIA:5 GM204 [GeForce GTX 980M]
15:49:09: CUDA: 5.2
15:49:09: CUDA Driver: 7050

[7im]

Okay, what version of CUDA does the FAHBench page say is needed? The original defunct one, or the page I made to keep it alive on the FAH web site.

[toTOW]

Results from my R9 290X

OpenCL - DP:
Explicit: 9.86929 ns/day
Implicit: 22.6624 ns/day

OpenCL - SP:
Explicit: 57.2944 ns/day
Implicit: 190.162 ns/day

Edit : results from FahBench version using OpenMM 6.2 : https://github.com/pandegroup/fahbench/ ... /tag/2.0.1

OpenCL - DP:
Explicit: 7.3753 ns/day
Implicit: 13.5304 ns/day

OpenCL - SP:
Explicit: 42.2597 ns/day
Implicit: 123.6936 ns/day

The performance issue is clearly seen on these benchmarks.

[Trendle]

Some SP results comparing the latest Catalyst driver to pre-Omega release.

Radeon HD 7970

Catalyst 15.4 @1110 MHz



Catalyst 14.9 @1100 MHz (1110 MHz not stable)

Catalyst 13.12 @1100 MHz (1110 MHz not stable)

[DarthBaggins]

Going to have to try this out, want to see the results of my 5930k and 970 with them both oc'd

[foldy]

A german computer magazine tested the best NVIDIA and AMD GPUs with FahBench 1.2.0 (OpenMM 5.1) on Windows 8.1
http://www.computerbase.de/2015-06/amd- ... e-fahbench

Nvidia GeForce GTX Titan X
Nvidia GeForce GTX 980 Ti
Nvidia GeForce GTX 980
AMD Radeon R9 Fury X
AMD Radeon R9 390X
AMD Radeon R9 290X

Please add to list.

[DarkFoss]

Here's my FuryX results Windows 8.1 x64 AMD 15.7 driver I did not run any DP. AMD FX 8350 Sapphire Fury X nothing overclocked.
2.0.2 run with the following commands since the gui didn't work :
FAHBench-cmd --device-id 0 --platform OpenCL --precision single --solvent implicit
FAHBench-cmd --device-id 0 --platform OpenCL --precision single --solvent explicit

Code: Select all

FahBench 1.0
Fiji
Opencl-SP
Explicit Solvent 34.9122 ns/day.
Implicit Solvent 197.035 ns/day.

FahBench 1.2
Fiji
Opencl-SP
Explicit Solvent 60.2216 ns/day
Implicit Solvent 253.573 ns/day

FahBench 2.0.2 Windows-x64-nocuda
Fiji
Opencl SP
Explicit 46.2156
Implicit 163.2585

[foldy]

Anandtech tested top Nvidia and AMD GPUs
http://www.anandtech.com/show/9421/the- ... re-asus/16

Nvidia GeForce GTX 980 Ti
Nvidia GeForce GTX 980
AMD Radeon R9 Fury X
AMD Radeon R9 Fury

Please add to list.

[Hughesy26]

My results - all OpenCL single precision

GTX 480: Explicit: 19.4645 Implicit: 83.2494
i5 4690k: Explicit: 3.01561 Implicit: 3.78837

GTX 840M: 4.4245 Implicit: 23.21
i5 4210M: 1.41124 Implicit: 1.52605

Bizarrely my desktop (top 2 results) gets the same PPD as my laptop

[Ascetus]

OpenCL double precision

[Ascetus]

OpenCL SP


Double precision in FAH is pointless?

[toTOW]

Yes, it's not used in FAH.