Yes you are right.
Is seams that at the moment there are only Cluster Physics apps. that are using
the advantage of double precision calculation hardware capability of the AMD cards.
So I guess the message are 'buy NVIDIA cards from now on' at least if you want
to use them in anything else than games.
Yes you are right.
Is seams that at the moment there are only Cluster Physics apps. that are using
the advantage of double precision calculation hardware capability of the AMD cards.
So I guess the message are 'buy NVIDIA cards from now on' at least if you want
to use them in anything else than games.
I think as more of the OpenCL foramt comes out and becomes easier and more of its intricacies are known by more people, Boinc will adapt it and the pure Nvidia card crunchers will be phased out. But as stated Nvidia provided the techs to do the heavy programming and since CUDA only works on Nvidia cards your conclusion is accurate, for now. As you also said those with ATI cards can always crunch for Folding or MilkyWay. Although as has been stated many, many times before, Boinc itself needs to grow up a bit for the gpu and cpu to co-exist at optimal levels.
I think as more of the OpenCL format comes out and becomes easier and more of its intricacies are known by more people, Boinc will adapt it and the pure Nvidia card crunchers will be phased out.
One more time. BOINC will just detect what GPU you have in your system, it is up to the projects if they want to use CUDA, CAL or OpenCL for the applications.
BOINC will not have OpenCL support. It doesn't do any of the crunching.
CUDA, CAL, OpenCL are all ways of utilizing the GPU to do (massively intricate) calculations.
BOINC will just detect what GPU you have in your system, it is up to the projects if they want to use CUDA, CAL or OpenCL for the applications.
BOINC will not have OpenCL support. It doesn't do any of the crunching.
CUDA, CAL, OpenCL are all ways of utilizing the GPU to do (massively intricate) calculations.
Is that to say that it is up to Milkyway@Home to include the AMD GPU apps. in their code library and then BOINC will detect what kind of GPU you have and the right apps could be chosen atomaticaly?
At this moment BOINC detects only Nvidia GPUs. For a future version it will also detect ATI GPUs. And after that whatever else there is available on the GPU market.
Travis at Milkyway will just have to decide upon what to use for the GPU application for the ATI cards, CAL (Brook+) as is in use at this moment in the app made by Cluster Physx, or OpenCL, which is still very much in its infancy.
If it's CAL, it'll be a separate application.
If it's OpenCL, it may only need one application that can cater Nvidia and ATI GPUs at the same time. (And I don't see why not, CPU applications can cater all kinds of CPUs as well).
If it's OpenCL, it may only need one application that can cater Nvidia and ATI GPUs at the same time. (And I don't see why not, CPU applications can cater all kinds of CPUs as well).
I don't think so, with OpenCL we can get source code compatibility, but the compiled apps should be different. The fact that we don't need separate apps for (say) AMD Opterons etc is just a result of the intentional x86 compatibility designed into the AMD CPUs because they have a license agreement with Intel. There's no such agreement between AMD and NVIDIA, and I doubt they are aiming for one.
If it's OpenCL, it may only need one application that can cater Nvidia and ATI GPUs at the same time. (And I don't see why not, CPU applications can cater all kinds of CPUs as well).
I don't think so, with OpenCL we can get source code compatibility, but the compiled apps should be different. The fact that we don't need separate apps for (say) AMD Opterons etc is just a result of the intentional x86 compatibility designed into the AMD CPUs because they have a license agreement with Intel. There's no such agreement between AMD and NVIDIA, and I doubt they are aiming for one.
CU Bikeman
And in fact there are different apps for Mac, Linux and Windows OSes, and some of them even use the same cpu.
What performance of GPU-applications expected? Comparable with MilkyWay GPU or not?
Gpu crunching does happen much faster than cpu crunching but trying to equate its performance here with its performance over on MilkyWay is not possible right now. That can only happen when the app finally comes out. There is a ton of testing neeeded to make sure it will even work here and then a ton of testing to make SURE it works here, etc etc. Now I a nobody here, so I am just a regular ol user here just like everyone else and for all I know they will introduce the gpu app tomorrow. BUT I would guess that Einstein, like alot of the other projects, are not putting their full resources towards gpu crunching until things settle down a bit more. CUDA only supports Nvidia cards while OpenGL supports all video cards, BUT OpenGL is not a mature product yet. Here is the wiki for OpenGL http://en.wikipedia.org/wiki/Opengl
Yes you are right. Is seams
)
Yes you are right.
Is seams that at the moment there are only Cluster Physics apps. that are using
the advantage of double precision calculation hardware capability of the AMD cards.
So I guess the message are 'buy NVIDIA cards from now on' at least if you want
to use them in anything else than games.
RE: Yes you are right. Is
)
I think as more of the OpenCL foramt comes out and becomes easier and more of its intricacies are known by more people, Boinc will adapt it and the pure Nvidia card crunchers will be phased out. But as stated Nvidia provided the techs to do the heavy programming and since CUDA only works on Nvidia cards your conclusion is accurate, for now. As you also said those with ATI cards can always crunch for Folding or MilkyWay. Although as has been stated many, many times before, Boinc itself needs to grow up a bit for the gpu and cpu to co-exist at optimal levels.
RE: I think as more of the
)
One more time. BOINC will just detect what GPU you have in your system, it is up to the projects if they want to use CUDA, CAL or OpenCL for the applications.
BOINC will not have OpenCL support. It doesn't do any of the crunching.
CUDA, CAL, OpenCL are all ways of utilizing the GPU to do (massively intricate) calculations.
RE: BOINC will just detect
)
Is that to say that it is up to Milkyway@Home to include the AMD GPU apps. in their code library and then BOINC will detect what kind of GPU you have and the right apps could be chosen atomaticaly?
I thoght that BOINC only detected NVIDIA GPU's.
At this moment BOINC detects
)
At this moment BOINC detects only Nvidia GPUs. For a future version it will also detect ATI GPUs. And after that whatever else there is available on the GPU market.
Travis at Milkyway will just have to decide upon what to use for the GPU application for the ATI cards, CAL (Brook+) as is in use at this moment in the app made by Cluster Physx, or OpenCL, which is still very much in its infancy.
If it's CAL, it'll be a separate application.
If it's OpenCL, it may only need one application that can cater Nvidia and ATI GPUs at the same time. (And I don't see why not, CPU applications can cater all kinds of CPUs as well).
RE: If it's OpenCL, it
)
I don't think so, with OpenCL we can get source code compatibility, but the compiled apps should be different. The fact that we don't need separate apps for (say) AMD Opterons etc is just a result of the intentional x86 compatibility designed into the AMD CPUs because they have a license agreement with Intel. There's no such agreement between AMD and NVIDIA, and I doubt they are aiming for one.
CU
Bikeman
RE: RE: If it's OpenCL,
)
And in fact there are different apps for Mac, Linux and Windows OSes, and some of them even use the same cpu.
What performance of
)
What performance of GPU-applications expected? Comparable with MilkyWay GPU or not?
RE: What performance of
)
Gpu crunching does happen much faster than cpu crunching but trying to equate its performance here with its performance over on MilkyWay is not possible right now. That can only happen when the app finally comes out. There is a ton of testing neeeded to make sure it will even work here and then a ton of testing to make SURE it works here, etc etc. Now I a nobody here, so I am just a regular ol user here just like everyone else and for all I know they will introduce the gpu app tomorrow. BUT I would guess that Einstein, like alot of the other projects, are not putting their full resources towards gpu crunching until things settle down a bit more. CUDA only supports Nvidia cards while OpenGL supports all video cards, BUT OpenGL is not a mature product yet. Here is the wiki for OpenGL
http://en.wikipedia.org/wiki/Opengl
But OpenGL is not OpenCL and
)
But OpenGL is not OpenCL and viceversa.
Tullio