Right now I think the current GPU-linux requirements are:
- OpenCL 1.1 capable graphics card (this means GeForce 4xx and above)
- minimum 1 gig of memory on the graphics card
- 64 bit linux
Can someone please confirm this or correct me?
Btw: what AMD card is the minimum to meet this requirements? Radeon HD 5xxx?
Thanks,
Stephan
PS: This puts two of my crunchers out of the race. Sad, but ...
> https://einsteinathome.org/host/11710709
> https://einsteinathome.org/host/2069906
PPS: Is it possible to get CUDA workunits again? Or OpenCL 1.0
Copyright © 2024 Einstein@Home. All rights reserved.
Stephan Goll wrote:Right now
)
The answers are, (i) Yes (I believe but could be wrong), (ii) Yes but ~0.75GB for the beta test app, (iii) Yes (I believe), (iv) Why don't you check here.
Anything's possible :-).
The Devs mentioned that they were committed to OpenCL and that pressures of their workload tended to mitigate against having the time to develop a CUDA version. However my impression is that the door wasn't completely closed and that maybe with enough volunteer pressure and the lack of any progress on a more efficient nvidia opencl app version or nvidia implementation of opencl (wherever the problem lies - I wouldn't know for sure), it could happen at some future point.
I've voted with my feet :-). I have a number of GTX650s that were doing fine with BRPx but very poorly with FGRPB1G. I'm running a few just to see how things progress but the bulk have been shut down or replaced with AMD R7 370s which are doing very well.
Cheers,
Gary.
Gary Roberts wrote:Why
)
I'm not that familiar with the AMD stuff. So thanks for helping me. Really.
--
Stephan
PS: lspci tells me: 00:01.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Richland [Radeon HD 8670D]
But the X server says it's ARUBA aka northern island. Sometime AMD is a bit confusing (nVidia too, but that's another story).
OpenCL1.1 and above old
)
OpenCL1.1 and above
old hosts can run SETI maybe
Gary Roberts wrote: Yes but
)
Just need to be a little careful about the measure of GPU memory size, the requirement is the size reported available for OpenCL - which will be less than the physical size of the card.
I have two 768MB GPUs - that are too small, clinfo shows 708MB available, so if you have a card with physical 0.81 GB or more, you may be able to run the beta app - run clinfo for Global Memory size.
freestman wrote:old hosts can
)
or primegrid or collatz.