#NVIDIA Bringing #CUDA To #RISCV
NVIDIA's drivers and CUDA software stack are predominantly supported on x86_64 and AArch64 systems but in the past was supported on IBM POWER. This week at the RISC-V Summit China event, NVIDIA's Frans Sijstermans announced that CUDA will be coming to RISC-V.
#AMD for their part with the upstream #opensource #AMDKFD kernel compute driver can already build on RISC-V and the #ROCm user-space components can also be built on RISC-V.
https://www.phoronix.com/news/NVIDIA-CUDA-Coming-To-RISC-V

New #ZLUDA 5 Preview Released For #CUDA On Non-NVIDIA #GPU
For now this ability to run unmodified CUDA apps on non-#NVIDIA GPUs is focused on #AMD GPUs of the #Radeon RX 5000 series and newer, which is AMD Radeon GPUs with #ROCm. Besides CUDA code samples, GeekBench has been one of the early targets for testing.
https://www.phoronix.com/news/ZLUDA-5-preview.43

It's convenient that I can use #LLMs to help me learn how to use LLMs because I'm pretty sure I wouldn't be able to figure it out any other way.

I want to use my local #Ollama models with #Copilot in #VSCode, but I have an #AMD #GPU so apparently I need to install something called the #ROCm (Radeon Open Compute Platform) via the Windows 11 HIP SDK?

And maybe all this doesn't work in #WSL, so I'll have to reinstall it in #Ubuntu there if I want to use it in one of those workspaces?

#ZLUDA Making Progress In 2025 On Bringing #CUDA To Non-NVIDIA #GPU
ZLUDA #opensource effort that started half-decade ago as drop-in CUDA implementation for #Intel GPUs and then for several years was funded by ##AMD as a CUDA implementation for #Radeon GPUs atop #ROCm and then open-sourced but then reverted has been continuing to push along a new path since last year. Current take on ZLUDA is a multi-vendor CUDA implementation for non-NVIDIA GPUs for #AI workloads & more.
https://www.phoronix.com/news/ZLUDA-Q2-2025-Update

Engineering Supercomputing Platforms for Biomolecular Applications

#CUDA #ROCm #Biology #Biomolecules #MolecularDynamics #HPC #Physics #Package

https://hgpu.org/?p=29954

Any #Linux #kernel ,
#graphics or #GPU people out there?

I'm trying to understand the relationship between the #amgdpu driver shipped with the kernel; and the "andgpu-dkms" driver that comes with #ROCm .

Specifically, with a recent enough kernel, do we really need to install the ROCm version of the driver? Does the ROCm version contain stuff the general driver does not? Or is the ROCm stack (esp. libhsa) tightly tied to a very specific version of the driver?

#AMD splits #ROCm toolkit into two parts – ROCm #AMDGPU drivers get their own branch under Instinct #datacenter #GPU moniker
The new #datacenter Instinct driver is a renamed version of the #Linux AMDGPU driver packages that are already distributed and documented with ROCm. Previously, everything related to ROCm (including the amdgpu driver) existed as part of the ROCm software stack.
https://www.tomshardware.com/pc-components/gpus/amd-splits-rocm-toolkit-into-two-parts-rocm-amdgpu-drivers-get-their-own-branch-under-instinct-datacenter-gpu-moniker

100 CPU threads & 240GB RAM to make @risc_v #AI @amd #ROCm and #t2linux https://www.twitch.tv/videos/2421181919

ffs, why does their docker only support Navi 31 and not Navi 32?
https://hub.docker.com/r/rocm/pytorch

I just wish both #Nvidia and #AMD would stop with that whole licensing bullshit around #CUDA and #ROCm and just include that damn stuff in the default driver.
I just want to run #Codestral on my local machine so I can use it with non-public code. Will be troublesome enough to cram it into 16gb VRAM.
#computer #Linux #AI

The Shamrock code: I- Smoothed Particle Hydrodynamics on GPUs

#SYCL #ROCm #CUDA #PTX #OpenMP #MPI #Astrophysics #Physics #Package

https://hgpu.org/?p=29827

AI rabbit hole ... I've been playing with Ollama and some stability diffusion tools on my MacBook Pro M2 Max and my Linux desktop ... the desktop is way faster and only has an RX6800 in it, so of course I'm now thinking about an Rx7900XTX ... (I don't do Nvidia cards) ...

Anyone have experience with this upgrade? Is going from 16gb of VRAM to 24gb going to make a massive difference?

Using radeontop I can see it's using all 16gb at some points, but not consistently ... and I'm not sure if that's an issue or a feature. I believe #rocm still has some issues.

I also uploaded the slides for my talk in the #hpc devroom at #fosdem about the #programming models in #ROCm.
Video is reviewed too and waiting to be released to the public. You can get the slides and the video (once released) at https://fosdem.org/2025/schedule/event/fosdem-2025-5143-programming-models-with-the-rocm-compiler/

Just how deep is #Nvidia's #CUDA moat really?
Not as impenetrable as you might think, but still more than Intel or AMD would like
It's not enough just to build a competitive part: you also have to have #software that can harness all those #FLOPS — something Nvidia has spent the better part of two decades building with its CUDA runtime, while competing frameworks for low-level #GPU #programming are far less mature like AMD's #ROCm or Intel's #OneAPI.
https://www.theregister.com/2024/12/17/nvidia_cuda_moat/ #developers

I'm quite excited by this output. Darknet normally displays information on the NVIDIA CUDA-capable GPU when you run "darknet --version".

Check out these new changes I'm working on.

- rsmi_num_monitor_devices()
- rsmi_dev_name_get()
- rsmi_dev_memory_total_get()

Had an opportunity to reboot my homelab server. Big kernel update and all that. This seems to have fixed my broken ROCm.

So I’m back to GPU accelerating my AI workloads.

The version of the kernel, though, might include the new AMD AI stuff, so I need to investigate this.