There has been new applications for NVIDIA GPUs released. The apps are for NVIDIA with CUDA 5.5 and Compute Capability 2.0 and the application versions are for linux and windows. Minimum recommended versions of NVIDIA drivers are 325.15 for linux and 326.80 Beta for windows.

Radim Vančo (Kyong)

Measured times for several cards (480 credit wu):

GTX 780Ti ~ 2250 s
GTX Titan ~ 1250 s (v101.10) (recommended CUDA double precision disabled)
GTX 780 ~ 1400 s (v101.10)
GTX 770 ~ 3700 s
GTX 760 ~ 1300 s (v101.10)
GTX 750Ti ~ 3900 s (v101.12)

GTX 680 ~ 2900 s (v101.10)
GTX 670 ~ 5000 s
GTX 660Ti ~ 3500 s (v101.10)
GTX 650Ti ~8400 s

GTX 580 ~4200 s
GTX 570 ~4200 s
GTX 560Ti ~5500 s
GTX 550Ti ~12000 s
GT 610 ~61000 s

GTX 460 ~9400 s

GTX 660M ~9000 s (v101.10)
GTX 540M ~32000 s
.
.
.
in progress.

Hello !

Nice ! A new GPU app.

Would you please be so kind to confirm the app name to be
used in order to write a short app_config.xml, as the
0.01 CPU allocation is not enough at all.

Thank You !

Kind Regards,

Philippe

Hello !

Nice ! A new GPU app.

Would you please be so kind to confirm the app name to be
used in order to write a short app_config.xml, as the
0.01 CPU allocation is not enough at all.

Thank You !

Kind Regards,

Philippe

period_search


I think that 0.01 (1%) should be ok. On my PC's CPU usage is about 0.003 (0.3%). Try 0.05 first and let us know.

Thanks
Z

Hello !

Thank You

I will reload some for my GTX770 + GTX570M=> 2 more for your stats ;)

Thank You

Kind Regards

Philippe

I will reload some for my GTX770 + GTX570M=> 2 more for your stats ;)

I'm interesting in GTX770.

Thanks

This is wonderful!

Thanks Radim & HA-SOFT!

I see very low CPU usage & high GPU load. Perfect!

---

terencewee*
Sicituradastra.

Wanted to send you this message via PM ...

Hello,

I have another question : how can I force receiving AVX CPU UT's ?

Thank You

Philippe

NB Will leave the GTX 770 on the project tonight, if this helps ;)

Only way is to use app_info, I think. On Intel Haswell cpus (4xxx) the speedup of AVX is about 40% comparing to SSE3.

@Phil1966:

I'd posted a sample app_info.xml over there.

---

terencewee*
Sicituradastra.

Now you have to add gpu part if you want to crunch on gpu also.

Thank You Very Much !
Philippe

It's more than 2 hours that I try to get some WU for GPU only (CPUs are fulload on Rosetta Challenge) but still "got 0 new tasks".
Tried to reset and detach/attach the project also.

There's some trick to get wus?

TIA
Riccardo

There's some trick to get wus?

Look at your boinc log.

Maybe you have the same problem as I had:
http://asteroidsathome.net/boinc/forum_thread.php?id=234

There's some trick to get wus?

Look at your boinc log.

Maybe you have the same problem as I had:
http://asteroidsathome.net/boinc/forum_thread.php?id=234

Uhm... I'm on Win7/64 and at the attach time the only eligible message in the log seem to be this:
"29/12/2013 21:30:08 | Asteroids@home | Tasks for CPU are available, but your preferences are set to not accept them"

So maybe apps && WU are for Linux only at the time? In fact after disabling CPU fetch the data folder is empty.
About the trick, maybe I'll try it after the chellange end.

Thank you very much, anyway.

Riccardo

Congratulations guys!
-- im so happy for this and look fwd to TURBO performance boost in next weeks :)


-- PS.: i have AMD graphics only :-/

P.

Yes, very low CPU usage, that's nice :) Well done!

If one has enough memory on the card one should be able to run 2 or more tasks simultaneously?

I have 2 tasks completed and verified and I see the credits awarded are the same as for CPU tasks. That suggests the GPU tasks and CPU tasks are the same size, is that correct?

---

BOINC FAQ Service
Official BOINC wiki
Installing BOINC on Linux

GPU and CPU crunch the same tasks. Same size and same credit.

Parallel tasks bring no profit, I think. Occupancy of card is 100% for CC3.X and 70% for CC2.X.

I do not expect performace boost from GPU comparing to CPU app. The most efficient configuration is Intel Haswell with AVX app. Only GTX Titan is faster (for one task) but needs more power.

Could we get an OpenCL-App for AMD GPUs too ?

What about AVX-App on current AMD CPUs ?

Has to be asked... Mac?
I have a 680MX that would be nice to run. Happy to be an app_info'd test case if you need one, and willing to try compilation if needed (tho i guess i'd need the nvidia cuda sdk, which i don't have).

My windows machines thank you and add a... 430... and a 760 to the mix. will report the pain and glorious (or otherwise - these would be DP float calculations right?) times of those when I've had a few tasks.

Cheers, Andrew