9 ms·

...from 2016. back in grad school, I did a comparison between the (then new) XeonPhi and GPUs to run my own N-Body engine using the same OpenCL implementation -

...from 2016. back in grad school, I did a comparison between the (then new) XeonPhi and GPUs to run my own N-Body engine using the same OpenCL implementation - the rest is history: GPUs still rule today for high intensity computational needs. Look at Machine Learning: GPUs are so good at SIMD, that if your domain can benefit from faster computation, /i you should make it SIMD i/, if it's not already. ;)

klelatti 6y ago
It's sad that OpenCL development seems to have lost its way. I have some OpenCL workloads where a running on a CPU is faster and some where GPUs are much quicker. The beauty of OpenCL is that I can choose between the two with zero code changes. But CUDA and ML have steamrollered everything!
- lasagnaphil 6y ago
  I really hope SYCL (https://en.m.wikipedia.org/wiki/SYCL https://en.m.wikipedia.org/wiki/SYCL) can gain some traction (since Intel also jumped on the GPU competition and are backing a substantial amount of it)
- MayeulC 6y ago
  I heard of Hallide quite recently, and find it promising: https://halide-lang.org/ https://halide-lang.org/ Not only you can decide on CPU vs GPU, you have more control over the kind of parallelism and intermediate results.
- DrNosferatu 6y ago
  Indeed, OpenCL runs on everything - if you can set it up... :/ Actually, at the time, what I did was a 3-way comparison between GPU, XeoPhi and classic CPU (was it 8 cores?) running the same OpenCL code. Anyways, the "many-core" specimen, XeonPhi, benched at the geometric mean between classic CPU and GPU - so closer to classic CPU! (we were using beta hardware, our intel specialist - nice, helpful guy actually - went mute after I got those results :D ) On top of that, reimplementing the same algo in CUDA gave ~10% advantage over OpenCL on NVidia hardware. I wonder how it would fare on a FPGA of the same price/TDP? And more relevant for 2021: - I wonder if we would get the same kind of results using Vulkan? - Speaking of which, does anyone know of any good books / tutorials / runnable examples of Vulkan in a compute-oriented application? Cheers!
  - DrNosferatu 6y ago
    PS: Does anyone have any experience with ArrayFire? https://arrayfire.com/ https://arrayfire.com/ (I believe it's from the people that did the [GPU]Jacket for Matlab)

klelatti 6y ago

It's sad that OpenCL development seems to have lost its way. I have some OpenCL workloads where a running on a CPU is faster and some where GPUs are much quicker. The beauty of OpenCL is that I can choose between the two with zero code changes. But CUDA and ML have steamrollered everything!

lasagnaphil 6y ago
I really hope SYCL (https://en.m.wikipedia.org/wiki/SYCL https://en.m.wikipedia.org/wiki/SYCL) can gain some traction (since Intel also jumped on the GPU competition and are backing a substantial amount of it)
MayeulC 6y ago
I heard of Hallide quite recently, and find it promising: https://halide-lang.org/ https://halide-lang.org/ Not only you can decide on CPU vs GPU, you have more control over the kind of parallelism and intermediate results.
DrNosferatu 6y ago
Indeed, OpenCL runs on everything - if you can set it up... :/ Actually, at the time, what I did was a 3-way comparison between GPU, XeoPhi and classic CPU (was it 8 cores?) running the same OpenCL code. Anyways, the "many-core" specimen, XeonPhi, benched at the geometric mean between classic CPU and GPU - so closer to classic CPU! (we were using beta hardware, our intel specialist - nice, helpful guy actually - went mute after I got those results :D ) On top of that, reimplementing the same algo in CUDA gave ~10% advantage over OpenCL on NVidia hardware. I wonder how it would fare on a FPGA of the same price/TDP? And more relevant for 2021: - I wonder if we would get the same kind of results using Vulkan? - Speaking of which, does anyone know of any good books / tutorials / runnable examples of Vulkan in a compute-oriented application? Cheers!
- DrNosferatu 6y ago
  PS: Does anyone have any experience with ArrayFire? https://arrayfire.com/ https://arrayfire.com/ (I believe it's from the people that did the [GPU]Jacket for Matlab)