Like John Clark I also run some ATI 48XX and 58XX cards, and would also be interested in using them for Einstein. Perhaps if the developers could gauge the amount of interest in that, it might help to determine timescales.
Waiting for Godot & salvation :-)
Why do doctors have to practice?
You'd think they'd have got it right by now
There won't be an App using proprietary ATI CAL / Stream, and OpenCL development is underway as fast as the bugs in the ATI OpenCL libraries / drivers / compilers allow.
Good to know that it's actively being pursued. I've got an HD4670 that I'd move completely over to Einstein once there's a Beta app that needs my testing help.
There won't be an App using proprietary ATI CAL / Stream, and OpenCL development is underway as fast as the bugs in the ATI OpenCL libraries / drivers / compilers allow.
BM
GPU-Grid is facing a similar situation: OpenCL is far to buggy to be used by now. At least for more complex software.
There won't be an App using proprietary ATI CAL / Stream, and OpenCL development is underway as fast as the bugs in the ATI OpenCL libraries / drivers / compilers allow.
Because CUDA
* was first
* is more stable than both, CAL / Stream and OpenCL
* is still being developed (AMD will probably want to switch to OpenCL only, as CAL / Stream hasn't really taken off yet)
Personally I prefer ATI / AMD hardware and am crunching on 2 ATIs by now. But on the software side CUDA is far superior to anything else currently on the markent, on has to give the Green team credit for that.
I've found that the 6.14.* series of BOINC versions will be the first ones that offer significant support for GPU workunits in OpenCL. None ready for even alpha testing yet.
Among other things, one problem they will have to address is to provide a standard mapping from the way BOINC identifies multiple GPUs on the same computer to the way OpenCL identifies them. Milkyway@Home is currently using their own version of this mapping, which appears to work only if:
The computer does not have both an AMD/ATI GPU and an Nvidia GPU.
Milkyway@Home gets the entire set of GPUs to choose from, or there is only one
GPU.
There are no two graphics cards on the computer different enough that
workunits for one of them will not also work on all the others.
Available in source code form, mainly for Linux. It appears to be possible to compile it for Windows also, but with this method rather poorly documented.
Some BOINC projects are already using OpenCL for CPU workunits, such as Milkyway@Home and Poem@Home. Milkyway@Home appears to be the only one offering any GPU OpenCL workunits, though.
Like John Clark I also run
)
Like John Clark I also run some ATI 48XX and 58XX cards, and would also be interested in using them for Einstein. Perhaps if the developers could gauge the amount of interest in that, it might help to determine timescales.
Waiting for Godot & salvation :-)
Why do doctors have to practice?
You'd think they'd have got it right by now
There won't be an App using
)
There won't be an App using proprietary ATI CAL / Stream, and OpenCL development is underway as fast as the bugs in the ATI OpenCL libraries / drivers / compilers allow.
BM
BM
Good to know that it's
)
Good to know that it's actively being pursued. I've got an HD4670 that I'd move completely over to Einstein once there's a Beta app that needs my testing help.
RE: There won't be an App
)
GPU-Grid is facing a similar situation: OpenCL is far to buggy to be used by now. At least for more complex software.
MrS
Scanning for our furry friends since Jan 2002
RE: There won't be an App
)
So why the decision to use proprietary CUDA?
Because CUDA * was first * is
)
Because CUDA
* was first
* is more stable than both, CAL / Stream and OpenCL
* is still being developed (AMD will probably want to switch to OpenCL only, as CAL / Stream hasn't really taken off yet)
Personally I prefer ATI / AMD hardware and am crunching on 2 ATIs by now. But on the software side CUDA is far superior to anything else currently on the markent, on has to give the Green team credit for that.
MrS
Scanning for our furry friends since Jan 2002
OpenCL 1.1 ATI support You
)
OpenCL 1.1 ATI support
You must use supported Compilers also...
WWW of Polish National Team - Join! Crunch! Win!
I've found that the 6.14.*
)
I've found that the 6.14.* series of BOINC versions will be the first ones that offer significant support for GPU workunits in OpenCL. None ready for even alpha testing yet.
Among other things, one problem they will have to address is to provide a standard mapping from the way BOINC identifies multiple GPUs on the same computer to the way OpenCL identifies them. Milkyway@Home is currently using their own version of this mapping, which appears to work only if:
The computer does not have both an AMD/ATI GPU and an Nvidia GPU.
Milkyway@Home gets the entire set of GPUs to choose from, or there is only one
GPU.
There are no two graphics cards on the computer different enough that
workunits for one of them will not also work on all the others.
I've found a program said to
)
I've found a program said to offer SOME help for translating CUDA to OpenCL, but not to do the full job. Named Swan.
http://www.cse.scitech.ac.uk/disco/mew20/presentations/GPU_MattHarvey.pdf
Available in source code form, mainly for Linux. It appears to be possible to compile it for Windows also, but with this method rather poorly documented.
Some BOINC projects are already using OpenCL for CPU workunits, such as Milkyway@Home and Poem@Home. Milkyway@Home appears to be the only one offering any GPU OpenCL workunits, though.
RE: AMD and Academic
)
http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-newsArticle&ID=1573883
Hopefully this will spur things on and everyone can get away from proprietary anything.