ProfInfer: An eBPF-based Fine-Grained LLM Inference Profiler
#JuliaCon 2026 will have a minisymposium on "Julia, GPUs, and Accelerators" hosted by Julian Samaroo, Tim Besard, and @vchuravy find out more on https://pretalx.com/juliacon-2026/talk/UPRSQL/ and submit your talks through the Call for Proposals https://juliacon.org/2026/cfp/ until Feb 28th!
hm, not sure what I did, but today I seem to be able to use PoCL, Clover and rusticl simultaneously in the same program on Debian Trixie (previously my attempts failed with mysterious segfaults). Using fraktaler-3-3.1 with environment variables RUSTICL_ENABLE=radeonsi RUSTICL_FEATURES=fp64 for my AMD RX 580 GPU.
Seem rusticl is significantly slower than Clover, hopefully Clover is not deleted before rusticl catches up.
Call for Submissions for IWOCL 2026 has been extended until January 25th!
Now is the time to submit on of four types of submissions:
+ Technical Presentations & Case Studies (500 word abstract)
+ Short Research Papers (3-4 page paper)
+ Full Research Papers (8-12 page paper)
+ Posters (1-2 page abstract)
Get your submission in now and join us in for another inspiring workshop at IWOCL 2026!
https://iwocl.org
#OpenCL #SYCL #IWOCL2026
Radxa NX4 system-on-module runs RK3576 with LPDDR5 and 6 TOPS AI
Khronos Group Welcomes Lisuan Technology as Contributor Member
Lisuan Technology is a company dedicated to the research and development of graphics rendering GPUs. Founded by leading figures in the GPU industry, it possesses capabilities in large-scale chip R&D, top-level architecture design, software stack design, and mass production. They are interested in #OpenCL, #OpenGL, #OpenGLES, #Vulkan, #WebGL, #SPIR, and #OpenXR.
https://www.lisuantech.com/
Okay #Gamer #Gamers... A friend's 16 year old is looking to build a PC for #gaming over the holidays... Now, I have a bunch of old PC components lying around... So I have to ask the question... Is an #AMD #Radeon Vega RX 64 or a WX9100 usable for a modern game machine? I remember these as monsters for #OpenCL -- but I don't think I *ever* connected a monitor to them.
Mostly interested in running #Linux for now, but might move over to #Windows eventually...
Finally enabled OpenCL support in Darktable 5.0.1 on Debian 13 with AMD integrated GPU! First, `sudo apt-get install ocl-icd-libopencl1 rocm-opencl-icd`. However, `darktable-cltest -d verbose` shows "failed to compile program `demosaic_ppg.cl'!" and "fatal error: malformed or corrupted AST file: 'could not find file '/usr/lib/llvm-17/lib/clang/17/include/opencl-c-base.h'". Solution: `sudo apt-get install libclang-common-17-dev`.
#darktable #opencl #rocm #debian #amdgpu #linux
Interesting notes here on CLBlast https://github.com/ysh329/OpenCL-101/issues/42 worth running through a translater. #opencl #machinelearning
I need to get better about announcing improvements to the OpenCL Intercept Layer, so: I merged two new features this week that I think are pretty neat. As a reminder, the OpenCL Intercept Layer is an open source tool for debugging and profiling OpenCL applications. It works with most OpenCL implementations and requires no application modifications. #OpenCL
Blog post: Analyzing @darktable OpenCL Memory Use
https://op-co.de/blog/posts/darktable_opencl_memory/
This post outlines how to identify workloads that don't run on the GPU, slowing down the export pipeline, how to understand how much more GPU memory you will need, and how to make available that memory or change your workflow to require less of it.
I ran into troubles upgrading to Fedora43, and switching completely to Wayland. Hence #EGL OpenGL bindings somehow cannot share the context with #OpenCL. At least on 580.95.05 @nvidia drivers. Need to dig deeper, however it pushes me to rewrite my whole-life fun project yet another time. Maybe I switch to #rust
New update for the #OpenCL Hardware Capability viewer released. Mostly a maintenance update to improve Android compatibility incl. support for reading OpenCL caps on newer handhelds (e.g. Pixel devices).
https://opencl.gpuinfo.org/download.php