toad.social

Dr. Moritz LehmannFinally I can "SLI" AMD+Intel+Nvidia <a href="https://mast.hpc.social/tags/GPU" class="mention hashtag" rel="nofollow noopener" target="_blank">#GPU</a>s at home! I simulated this crow in flight at 680M grid cells in 36GB VRAM, pooled together from - 🟥 <a href="https://mast.hpc.social/tags/AMD" class="mention hashtag" rel="nofollow noopener" target="_blank">#AMD</a> Radeon RX 7700 XT 12GB (RDNA3) - 🟦 <a href="https://mast.hpc.social/tags/Intel" class="mention hashtag" rel="nofollow noopener" target="_blank">#Intel</a> Arc B580 12GB (Battlemage) - 🟩 <a href="https://mast.hpc.social/tags/Nvidia" class="mention hashtag" rel="nofollow noopener" target="_blank">#Nvidia</a> Titan Xp 12GB (Pascal) My <a href="https://mast.hpc.social/tags/FluidX3D" class="mention hashtag" rel="nofollow noopener" target="_blank">#FluidX3D</a> <a href="https://mast.hpc.social/tags/CFD" class="mention hashtag" rel="nofollow noopener" target="_blank">#CFD</a> software can pool the VRAM of any combination of any GPUs together via <a href="https://mast.hpc.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a>. <a href="https://mast.hpc.social/tags/Kr%C3%A4henliebe" class="mention hashtag" rel="nofollow noopener" target="_blank">#Krähenliebe</a> <a href="https://mast.hpc.social/tags/birds" class="mention hashtag" rel="nofollow noopener" target="_blank">#birds</a> <a href="https://mast.hpc.social/tags/crow" class="mention hashtag" rel="nofollow noopener" target="_blank">#crow</a> <a href="https://www.youtube.com/watch?v=1z5-ddsmAag" rel="nofollow noopener" translate="no" target="_blank">https://www.youtube.com/watch?v=1z5-ddsmAag</a>

John-Mark GurneyAs usual, getting something like GPU compute that's cross platform working is a message because everyone likes to do their own thing and reinvent the wheel.I would like something that is [modern] macOS and FreeBSD compatible, but doesn't look like that's possible since Apple deprecated OpenCL.(Also, could Apple have picked a less searchable term for their new GPU framework?)It's again looking like the best way to be cross platform is to use JS+browser.Or am I missing some library?<a href="https://flyovercountry.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> <a href="https://flyovercountry.social/tags/GPUCompute" class="mention hashtag" rel="nofollow noopener" target="_blank">#GPUCompute</a> <a href="https://flyovercountry.social/tags/FreeBSD" class="mention hashtag" rel="nofollow noopener" target="_blank">#FreeBSD</a>

karolherbst 🐧 🦀Who is using CL_sRGBA images with <a href="https://chaos.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a>, specifically to write to it (cl_khr_srgb_image_writes)?There is limited hw support for writing to sRGBA images and I'm now curious what even uses that feature.It was apparently important enough to require support for it for OpenCL 2.0, but... that's not telling me much.

Dr. Moritz LehmannIs it possible to run AMD+Intel+Nvidia <a href="https://mast.hpc.social/tags/GPU" class="mention hashtag" rel="nofollow noopener" target="_blank">#GPU</a>s in the same PC? Yes! 🖖😋 Got this RDNA3 chonker for free from 11 bit studios contest! It completes my 36GB VRAM RGB SLI abomination setup: - 🟥 <a href="https://mast.hpc.social/tags/AMD" class="mention hashtag" rel="nofollow noopener" target="_blank">#AMD</a> Radeon RX 7700 XT 12GB - 🟦 <a href="https://mast.hpc.social/tags/Intel" class="mention hashtag" rel="nofollow noopener" target="_blank">#Intel</a> Arc B580 12GB - 🟩 <a href="https://mast.hpc.social/tags/Nvidia" class="mention hashtag" rel="nofollow noopener" target="_blank">#Nvidia</a> Titan Xp 12GB The drivers all work together in <a href="https://mast.hpc.social/tags/Linux" class="mention hashtag" rel="nofollow noopener" target="_blank">#Linux</a> Ubuntu 24.04.2. Backbone is an ASUS ProArt Z790 with i7-13700K and 64GB, PCIe 4.0 x8/x8 + 3.0 x4 - plenty interconnect bandwidth. Finally I can develop and test <a href="https://mast.hpc.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> on all major patforms!

kandidA little bit like <a href="https://chaos.social/tags/fractalFlame" class="mention hashtag" rel="nofollow noopener" target="_blank">#fractalFlame</a> Made with <a href="https://chaos.social/tags/openFrameworks" class="mention hashtag" rel="nofollow noopener" target="_blank">#openFrameworks</a> and <a href="https://chaos.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a>

Janne MorenI wish <a href="https://fosstodon.org/tags/pytorch" class="mention hashtag" rel="nofollow noopener" target="_blank">#pytorch</a> wasn't CUDA/ROCm only :(I know I *can* use the nodes at work, but that's not the point. I want to use my own new toy, not somebody else's.Any DL framework out there with good support for <a href="https://fosstodon.org/tags/Vulkan" class="mention hashtag" rel="nofollow noopener" target="_blank">#Vulkan</a> or <a href="https://fosstodon.org/tags/Opencl" class="mention hashtag" rel="nofollow noopener" target="_blank">#Opencl</a> ?

Dr. Moritz LehmannMy <a href="https://mast.hpc.social/tags/IWOCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#IWOCL</a> 2025 Keynote presentation is online! 🖖🧐 Scaling up <a href="https://mast.hpc.social/tags/FluidX3D" class="mention hashtag" rel="nofollow noopener" target="_blank">#FluidX3D</a> <a href="https://mast.hpc.social/tags/CFD" class="mention hashtag" rel="nofollow noopener" target="_blank">#CFD</a> beyond 100 Billion cells on a single computer - a story about the true cross-compatibility of <a href="https://mast.hpc.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> <a href="https://www.youtube.com/watch?v=Sb3ibfoOi0c&list=PLA-vfTt7YHI2HEFrpzPhhQ8PhiztKhHU8&index=1" rel="nofollow noopener" translate="no" target="_blank">https://www.youtube.com/watch?v=Sb3ibfoOi0c&list=PLA-vfTt7YHI2HEFrpzPhhQ8PhiztKhHU8&index=1</a> Slides: <a href="https://www.iwocl.org/wp-content/uploads/iwocl-2025-moritz-lehmann-keynote.pdf" rel="nofollow noopener" translate="no" target="_blank">https://www.iwocl.org/wp-content/uploads/iwocl-2025-moritz-lehmann-keynote.pdf</a>

Dr. Moritz LehmannI just uploaded the 5000th <a href="https://mast.hpc.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> hardware report to <a href="https://mastodon.gamedev.place/@sascha" class="u-url mention" rel="nofollow noopener" target="_blank">@sascha</a>'s gpuinfo.org database! 🖖🥳 And guess what <a href="https://mast.hpc.social/tags/GPU" class="mention hashtag" rel="nofollow noopener" target="_blank">#GPU</a> I reserved the spot for: <a href="https://mast.hpc.social/tags/Intel" class="mention hashtag" rel="nofollow noopener" target="_blank">#Intel</a> Arc B580 <a href="https://mast.hpc.social/tags/Battlemage" class="mention hashtag" rel="nofollow noopener" target="_blank">#Battlemage</a> 🟦 <a href="https://opencl.gpuinfo.org/displayreport.php?id=5000" rel="nofollow noopener" translate="no" target="_blank">https://opencl.gpuinfo.org/displayreport.php?id=5000</a> I have contributed 4.2% (211) of all entries. 🖖🫡

AdventureTenseIf you're using darktable 5.01, the latest update on the Fedora Linux repo (flatpak), may have just added OpenCL support for Radeon GPUs (at least it did for my RX 6600). The Flathub version doesn't seem to add OpenCL (currently), so it may be a Fedora thing. I have not installed the RPM version so far, so I'm not sure about that package.<a href="https://mapstodon.space/tags/AMD" class="mention hashtag" rel="nofollow noopener" target="_blank">#AMD</a> <a href="https://mapstodon.space/tags/Radeon" class="mention hashtag" rel="nofollow noopener" target="_blank">#Radeon</a> <a href="https://mapstodon.space/tags/Darktable" class="mention hashtag" rel="nofollow noopener" target="_blank">#Darktable</a> <a href="https://mapstodon.space/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a>

Lukas WeidingerI’m thinking of <a href="https://gruene.social/tags/compiling" class="mention hashtag" rel="nofollow noopener" target="_blank">#compiling</a> <a href="https://gruene.social/tags/darktable" class="mention hashtag" rel="nofollow noopener" target="_blank">#darktable</a> from source so that it’s better optimized for my processor. Anybody experience with its potential? <a href="https://gruene.social/tags/question" class="mention hashtag" rel="nofollow noopener" target="_blank">#question</a> <a href="https://gruene.social/tags/followerpower" class="mention hashtag" rel="nofollow noopener" target="_blank">#followerpower</a>I’m generally ok with how fast the flatpak runs on my i7-1255 laptop. However, with such an iterative workflow, I feel that one has much to gain with slight improvements via <a href="https://gruene.social/tags/opencl" class="mention hashtag" rel="nofollow noopener" target="_blank">#opencl</a> and AVX.

Dr. Moritz LehmannWhat an honor to start the <a href="https://mast.hpc.social/tags/IWOCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#IWOCL</a> conference with my keynote talk! Nowhere else you get to talk to so many <a href="https://mast.hpc.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> and <a href="https://mast.hpc.social/tags/SYCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#SYCL</a> experts in one room! I shared some updates on my <a href="https://mast.hpc.social/tags/FluidX3D" class="mention hashtag" rel="nofollow noopener" target="_blank">#FluidX3D</a> <a href="https://mast.hpc.social/tags/CFD" class="mention hashtag" rel="nofollow noopener" target="_blank">#CFD</a> solver, how I optimized it at the smallest level of a single grid cell, to scale it up on the largest <a href="https://mast.hpc.social/tags/Intel" class="mention hashtag" rel="nofollow noopener" target="_blank">#Intel</a> <a href="https://mast.hpc.social/tags/Xeon6" class="mention hashtag" rel="nofollow noopener" target="_blank">#Xeon6</a> <a href="https://mast.hpc.social/tags/HPC" class="mention hashtag" rel="nofollow noopener" target="_blank">#HPC</a> systems that provide more memory capacity than any <a href="https://mast.hpc.social/tags/GPU" class="mention hashtag" rel="nofollow noopener" target="_blank">#GPU</a> server. 🖖😃

Dr. Moritz LehmannJust arrived in wonderful Heidelberg, looking forward to present the keynote talk at <a href="https://mast.hpc.social/tags/IWOCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#IWOCL</a> tomorrow!! See you there! 🖖😁 <a href="https://www.iwocl.org/" rel="nofollow noopener" translate="no" target="_blank">https://www.iwocl.org/</a> <a href="https://mast.hpc.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> <a href="https://mast.hpc.social/tags/SYCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#SYCL</a> <a href="https://mast.hpc.social/tags/FluidX3D" class="mention hashtag" rel="nofollow noopener" target="_blank">#FluidX3D</a> <a href="https://mast.hpc.social/tags/GPU" class="mention hashtag" rel="nofollow noopener" target="_blank">#GPU</a>

Giuseppe BilottaI'm liking the class this year. Students are attentive and participating, and the discussion is always productive.We were discussing the rounding up of the launch grid in <a href="https://fediscience.org/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> to avoid the catastrophic performance drops that come from the inability to divide the “actual” work size by anything smaller than the maximum device local work size, and were discussing on how to compute the “rounded up” work size.The idea is this: given the worksize N and the local size L, we have to round N to the smallest multiple of L that is not smaller than N. This effectively means computing D = ceili(N/L) and then using D*L.There are several ways to compute D, but on the computer, working only with integers and knowing that integer division always rounded down, what is the “best way”?D = N/L + 1 works well if N is not a multiple of L, but gives us 1 more than the intended result if N *is* a multiple of L. So we want to add the extra 1 only if N is not a multiple. This can be achieved for example withD = N/L + !!(N % L)which leverages the fact that !! (double logical negation) turns any non-zero value into 1, leaving zero as zero. So we round *down* (which is what the integer division does) and then add 1 if (and only if) there is a reminder to the division.This is ugly not so much because of the !!, but because the modulus operation % is slow.1/n

GPUOpen🧐 AMD Radeon GPU Analyzer (RGA) is our performance analysis tool for <a href="https://mastodon.gamedev.place/tags/DirectX" class="mention hashtag" rel="nofollow noopener" target="_blank">#DirectX</a>, <a href="https://mastodon.gamedev.place/tags/Vulkan" class="mention hashtag" rel="nofollow noopener" target="_blank">#Vulkan</a>, SPIR-V, <a href="https://mastodon.gamedev.place/tags/OpenGL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenGL</a>, & <a href="https://mastodon.gamedev.place/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a>. ✨As well as updates for AMD RDNA 4, there's enhancements to the ISA view UI, using the same updated UI as RGP ✨More detail: <a href="https://gpuopen.com/learn/rdna-cdna-architecture-disassembly-radeon-gpu-analyzer-2-12/?utm_source=mastodon&utm_medium=social&utm_campaign=rdts" rel="nofollow noopener" translate="no" target="_blank">https://gpuopen.com/learn/rdna-cdna-architecture-disassembly-radeon-gpu-analyzer-2-12/?utm_source=mastodon&utm_medium=social&utm_campaign=rdts</a> (🧵5/7)

Dr. Moritz LehmannHere's my <a href="https://mast.hpc.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> implementation: <a href="https://github.com/ProjectPhysX/FluidX3D/blob/master/src/kernel.cpp#L1924-L1993" rel="nofollow noopener" translate="no" target="_blank">https://github.com/ProjectPhysX/FluidX3D/blob/master/src/kernel.cpp#L1924-L1993</a>

Dr. Moritz Lehmann<a href="https://mast.hpc.social/tags/FluidX3D" class="mention hashtag" rel="nofollow noopener" target="_blank">#FluidX3D</a> <a href="https://mast.hpc.social/tags/CFD" class="mention hashtag" rel="nofollow noopener" target="_blank">#CFD</a> v3.2 is out! I've implemented the much requested <a href="https://mast.hpc.social/tags/GPU" class="mention hashtag" rel="nofollow noopener" target="_blank">#GPU</a> summation for object force/torque; it's ~20x faster than <a href="https://mast.hpc.social/tags/CPU" class="mention hashtag" rel="nofollow noopener" target="_blank">#CPU</a> <a href="https://mast.hpc.social/tags/multithreading" class="mention hashtag" rel="nofollow noopener" target="_blank">#multithreading</a>. 🖖😋 Horizontal sum in <a href="https://mast.hpc.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> was a nice exercise - first local memory reduction and then hardware-supported atomic floating-point add in VRAM, in a single-stage kernel. Hammering atomics isn't too bad as each of the ~10-340 workgroups dispatched at a time does only a single atomic add. Also improved volumetric <a href="https://mast.hpc.social/tags/raytracing" class="mention hashtag" rel="nofollow noopener" target="_blank">#raytracing</a>! <a href="https://github.com/ProjectPhysX/FluidX3D/releases/tag/v3.2" rel="nofollow noopener" translate="no" target="_blank">https://github.com/ProjectPhysX/FluidX3D/releases/tag/v3.2</a>

Dr. Moritz LehmannMy OpenCL-Benchmark now uses the dp4a instruction on supported hardware (<a href="https://mast.hpc.social/tags/Nvidia" class="mention hashtag" rel="nofollow noopener" target="_blank">#Nvidia</a> Pascal, <a href="https://mast.hpc.social/tags/Intel" class="mention hashtag" rel="nofollow noopener" target="_blank">#Intel</a> <a href="https://mast.hpc.social/tags/Arc" class="mention hashtag" rel="nofollow noopener" target="_blank">#Arc</a>, <a href="https://mast.hpc.social/tags/AMD" class="mention hashtag" rel="nofollow noopener" target="_blank">#AMD</a> RDNA, or newer) to benchmark INT8 tghroughput. dp4a is not exposed in <a href="https://mast.hpc.social/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> C, but can still be used via inline PTX assembly and compiler pattern recognition. Even Nvidia's compiler will turn the emulation implementation into dp4a, but in some cases does so with a bunch of unnecessary shifts/permutations on inputs, so better use inline PTX directly. 🖖🧐 <a href="https://github.com/ProjectPhysX/OpenCL-Benchmark/releases/tag/v1.8" rel="nofollow noopener" translate="no" target="_blank">https://github.com/ProjectPhysX/OpenCL-Benchmark/releases/tag/v1.8</a>

Alauddin Maulana Hirzan 💻Other things I have tested with FreeBSD: OpenCL with Discrete GPU via PyOpenCL lib<a href="https://mastodon.bsd.cafe/tags/FreeBSD" class="mention hashtag" rel="nofollow noopener" target="_blank">#FreeBSD</a> <a href="https://mastodon.bsd.cafe/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> <a href="https://mastodon.bsd.cafe/tags/PyOpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#PyOpenCL</a>

Alauddin Maulana Hirzan 💻aaah, nothing can beat the feel of beefed up FreeBSD with working dGPU. 1. OpenCL ✓ 2. OBS RenderD129 ✓ Thanks to <a href="https://mastodon.bsd.cafe/@vermaden" class="u-url mention" rel="nofollow noopener" target="_blank">@vermaden</a> for pointing my fault.<a href="https://mastodon.bsd.cafe/tags/FreeBSD" class="mention hashtag" rel="nofollow noopener" target="_blank">#FreeBSD</a> <a href="https://mastodon.bsd.cafe/tags/amdgpu" class="mention hashtag" rel="nofollow noopener" target="_blank">#amdgpu</a> <a href="https://mastodon.bsd.cafe/tags/opencl" class="mention hashtag" rel="nofollow noopener" target="_blank">#opencl</a>

Benjamin Carr, Ph.D. 👨🏻‍💻🧬<a href="https://hachyderm.io/tags/NVIDIA" class="mention hashtag" rel="nofollow noopener" target="_blank">#NVIDIA</a> <a href="https://hachyderm.io/tags/GeForce" class="mention hashtag" rel="nofollow noopener" target="_blank">#GeForce</a> <a href="https://hachyderm.io/tags/RTX5090" class="mention hashtag" rel="nofollow noopener" target="_blank">#RTX5090</a> <a href="https://hachyderm.io/tags/Linux" class="mention hashtag" rel="nofollow noopener" target="_blank">#Linux</a> <a href="https://hachyderm.io/tags/GPU" class="mention hashtag" rel="nofollow noopener" target="_blank">#GPU</a> Compute Performance <a href="https://hachyderm.io/tags/Benchmarks" class="mention hashtag" rel="nofollow noopener" target="_blank">#Benchmarks</a> When taking geo mean across 60+ benchmarks of <a href="https://hachyderm.io/tags/CUDA" class="mention hashtag" rel="nofollow noopener" target="_blank">#CUDA</a> / <a href="https://hachyderm.io/tags/OptiX" class="mention hashtag" rel="nofollow noopener" target="_blank">#OptiX</a> / <a href="https://hachyderm.io/tags/OpenCL" class="mention hashtag" rel="nofollow noopener" target="_blank">#OpenCL</a> / <a href="https://hachyderm.io/tags/Vulkan" class="mention hashtag" rel="nofollow noopener" target="_blank">#Vulkan</a> Compute, the GeForce RTX 5090 was delivering 1.42x the performance of GeForce <a href="https://hachyderm.io/tags/RTX4090" class="mention hashtag" rel="nofollow noopener" target="_blank">#RTX4090</a>. On performance-per-Watt GeForce RTX 5090 tended to deliver similar power efficiency to the RTX 4080/4090 graphics cards. GeForce RTX 5090 Founders Edition was running cooler than many of the other Founders Edition graphics cards tested. <a href="https://www.phoronix.com/review/nvidia-geforce-rtx5090-linux" rel="nofollow noopener" translate="no" target="_blank">https://www.phoronix.com/review/nvidia-geforce-rtx5090-linux</a>

Recent searches

Search options

Administered by:

Server stats:

#opencl