If this is your first visit, be sure to
check out the FAQ by clicking the
link above. You may have to register
before you can post: click the register link above to proceed. To start viewing messages,
select the forum that you want to visit from the selection below.
Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite
Wasn't OpenCL deprecated by Apple?
What happened to Vulkan Compute?
1. They were depreciated, but still work on apple.
2. what about it? vulkan compute and opencl are two different APIs. while both can be used for compute they have different strengths and weaknesses.
Yes, and then they realized they were big enough to force their own API on hardware vendors and software developers. Helped by the fact that their hardware ecosystem is closed, meaning they had much more leverage over the hardware vendors they worked with.
Chicken & egg problem. For software developers to support it, it needs to be sufficiently ubiquitous and implemented well enough that they don't get flooded with support requests by their users.
Hopefully, when Mesa has an up-to-date, fully-conformant implementation, that can finally start to happen. Even if performance isn't quite at the level of vendor-provided options, there are undeniable benefits from merely having ubiquitous, out-of-the-box support.
2. what about it? vulkan compute and opencl are two different APIs. while both can be used for compute they have different strengths and weaknesses.
Vulkan compute is targeted towards different apps than OpenCL. Vulkan requires much lower numerical precision than OpenCL, which seems based on the needs of HPC & scientific users. Of course, implementations are free to provide more than the minimum level of precision, but this could affect portability of Vulkan compute apps.
Vulkan is also much harder for app developers to use. You could use some middleware layer, of course, but I don't know if Vulkan addresses the issues of "performance portability" even as well as OpenCL does.
I use OpenCL (and only OpenCL) for all the inhouse GPGPU tools I write. I cannot use CUDA because not all my GPUs are from NVIDIA, and because I cannot depend on a vendor. The other GPGPU alternatives I've heard of don't seem to be ready to be used... they look a bit experimental, with not a good support in all platforms. With OpenCL, you can code today, and it works.
Comment