Since BRP is roughly 20 times the performance on a GPU that it is on a CPU, apparently the data is somewhat conducive to being split across the resources of a GPU to exploit its parallelism. Are there any plans going forward to accomplish the same thing on CPUs? Most everyone here has at least one dual-core machine in their herd, and I think that it would be possible to do a multithreaded app similar to how Aqua does. That would get the processing times down by roughly a factor of the number of cores, and likely spur an addition of beasty computers to the project if accomplished. Of course, an OpenCL version of the app would be able to utilize the resources of Cuda cores, ATI stream processors, and processor cores without having to do much in the way of code changes according to my admittedly small understanding of OpenCL. Just a thought and I certainly understand if it's been considered and abandoned for any particular reason.
CPU: The parallelization factor is less than one, i.e. running an app on 2 cores wouldn't give you a speedup of 2, but of about 1.2-1.5. Thus in the end running 2 single core tasks in parallel is much faster than running 2 dual-core tasks one after the other.
An OpenCL version of the BRP App is being worked on. However, OpenCL is not strongly supported by ATI, so outside the Mac world (pushed and supported by Apple) we encounter all kinds of difficulties with drivers, libraries and such. And for NVidia I guess the CUDA App will always be a little faster than the OpenCL version.
I just re-released the Mac OS application 1.04 under plan class BRP3SSE. This makes it possible for people that can run the CUDA App to opt-out the CPU version the same way as for the Windows and Linux BRP App, i.e. in E@H preferences.
RE: We released a new BRP3
)
Thank-you!
Since BRP is roughly 20 times
)
Since BRP is roughly 20 times the performance on a GPU that it is on a CPU, apparently the data is somewhat conducive to being split across the resources of a GPU to exploit its parallelism. Are there any plans going forward to accomplish the same thing on CPUs? Most everyone here has at least one dual-core machine in their herd, and I think that it would be possible to do a multithreaded app similar to how Aqua does. That would get the processing times down by roughly a factor of the number of cores, and likely spur an addition of beasty computers to the project if accomplished. Of course, an OpenCL version of the app would be able to utilize the resources of Cuda cores, ATI stream processors, and processor cores without having to do much in the way of code changes according to my admittedly small understanding of OpenCL. Just a thought and I certainly understand if it's been considered and abandoned for any particular reason.
CPU: The parallelization
)
CPU: The parallelization factor is less than one, i.e. running an app on 2 cores wouldn't give you a speedup of 2, but of about 1.2-1.5. Thus in the end running 2 single core tasks in parallel is much faster than running 2 dual-core tasks one after the other.
An OpenCL version of the BRP App is being worked on. However, OpenCL is not strongly supported by ATI, so outside the Mac world (pushed and supported by Apple) we encounter all kinds of difficulties with drivers, libraries and such. And for NVidia I guess the CUDA App will always be a little faster than the OpenCL version.
BM
BM
I just re-released the Mac OS
)
I just re-released the Mac OS application 1.04 under plan class BRP3SSE. This makes it possible for people that can run the CUDA App to opt-out the CPU version the same way as for the Windows and Linux BRP App, i.e. in E@H preferences.
BM
BM