Advanced search

Message boards : Graphics cards (GPUs) : GTX770 and OpenCL

Author Message
Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 324
Credit: 72,394,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 40241 - Posted: 24 Feb 2015 | 12:29:39 UTC
Last modified: 24 Feb 2015 | 12:43:30 UTC

Hello: My GTX 770 supports OpenCL and is also registered with BOINC see copy:

"mar 24 feb 2015 10:05:41 CET | | CUDA: NVIDIA GPU 0: GeForce GTX 770 (driver version 346.35, CUDA version 7.0, compute capability 3.0, 2048MB, 1992MB available, 3693 GFLOPS peak)
mar 24 feb 2015 10:05:41 CET | | CUDA: NVIDIA GPU 1 (ignored by config): GeForce GT 640 (driver version 346.35, CUDA version 7.0, compute capability 3.0, 2048MB, 1894MB available, 692 GFLOPS peak)
mar 24 feb 2015 10:05:41 CET | | OpenCL: NVIDIA GPU 0: GeForce GTX 770 (driver version 346.35, device version OpenCL 1.1 CUDA, 2048MB, 1992MB available, 3693 GFLOPS peak)
mar 24 feb 2015 10:05:41 CET | | OpenCL: NVIDIA GPU 1 (ignored by config): GeForce GT 640 (driver version 346.35, device version OpenCL 1.1 CUDA, 2048MB, 1894MB available, 692 GFLOPS peak)
mar 24 feb 2015 10:05:41 CET | | NVIDIA library reports 2 GPUs
mar 24 feb 2015 10:05:41 CET | | No ATI library found "

But it is not recognized to run OpenCL GPUGRID tasks in other projects works without problem,the same in Ubuntu 14.10 in Windows 8.1 POEM for example. Greetings.

Profile MJH
Project administrator
Project developer
Project scientist
Send message
Joined: 12 Nov 07
Posts: 696
Credit: 27,266,655
RAC: 0
Level
Val
Scientific publications
watwat
Message 40242 - Posted: 24 Feb 2015 | 13:22:14 UTC - in response to Message 40241.



But it is not recognized to run OpenCL GPUGRID tasks in other projects works without problem,the same in Ubuntu 14.10 in Windows 8.1 POEM for example. Greetings.


Because our application is for AMD GPUs.

Matt

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 324
Credit: 72,394,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 40243 - Posted: 24 Feb 2015 | 14:19:52 UTC - in response to Message 40242.
Last modified: 24 Feb 2015 | 14:20:24 UTC


Because our application is for AMD GPUs.

Matt


Hello: Thank you, if you already know and understand.

I think it would be good for the project, as in others, the tasks were for AMD OpenCL GPU and NVIDIA since both support it. Greetings.

Profile MJH
Project administrator
Project developer
Project scientist
Send message
Joined: 12 Nov 07
Posts: 696
Credit: 27,266,655
RAC: 0
Level
Val
Scientific publications
watwat
Message 40244 - Posted: 24 Feb 2015 | 14:59:46 UTC - in response to Message 40243.


I think it would be good for the project, as in others, the tasks were for AMD OpenCL GPU and NVIDIA since both support it. Greetings.


Quite the opposite. Allowing that application to run on NVIDIA GPUs will mean it will compete against our primary application for resources, with the result that our overall capacity will be reduced.

Matt

eXaPower
Send message
Joined: 25 Sep 13
Posts: 280
Credit: 1,449,568,667
RAC: 89
Level
Met
Scientific publications
watwatwatwatwatwatwatwat
Message 40255 - Posted: 25 Feb 2015 | 12:59:13 UTC - in response to Message 40244.


I think it would be good for the project, as in others, the tasks were for AMD OpenCL GPU and NVIDIA since both support it. Greetings.


Quite the opposite. Allowing that application to run on NVIDIA GPUs will mean it will compete against our primary application for resources, with the result that our overall capacity will be reduced.

Matt

Factoring in recent GROMACS application additions: OpenMP SSE/AVX CPU/AMD CL HSA - will Intel's Iris OpenCL iGPU (Haswell/Broadwell/Skylake) compute 1TeraSPFLOPS/256!G/DP) become a future GPUGRID resource?

?? GPUGRID Resource Efficiency/(ns/day)(hour/ns) performance > CUDA > ?Intel? HSA > CPU only > OPENCL HSA ??

App mdrun-917/VERSION 5.1-dev utilizing AMD GPU (Tahiti/Hawaii) OpenCL 50~ns/day.

SSE2 CPU computes ~1.5 to 4.0~ns. for [1] thread (4ns a thread = 256AVXcode) Will CPUMD only paths reappear before the Summer season?

GM204 GPU 6.5CUDA ACEMD task (70katoms/5e7step) achieves 300ns/day.
100ns/day for the resourceful GTX750 with a GTX960 producing 200~. GK107 provides 40ns/day.

(ns. Variation with interchangeable Total atom/Step length)

Looking at the methods: is [1] MDrun 917 task - step length 200k (400.0 ps.) 12-25x smaller than prior 2.5/5.0e7 (5k ps.) CPUMD?

Profile MJH
Project administrator
Project developer
Project scientist
Send message
Joined: 12 Nov 07
Posts: 696
Credit: 27,266,655
RAC: 0
Level
Val
Scientific publications
watwat
Message 40257 - Posted: 25 Feb 2015 | 14:47:30 UTC - in response to Message 40255.

Intel iGPUs - probably not, they aren't very well-suited for compute
CPUMD - no, that's unlikely to make a re-appearance. The performance simply wasn't good enough. We're going in the direction of a hybrid CPU-GPU application for AMD GPUs.

Matt

Post to thread

Message boards : Graphics cards (GPUs) : GTX770 and OpenCL