Wim Vander Schelden

Switching back to 570.133.20 solves the issue, so I guess this is a regression: `NVRM: loading NVIDIA UNIX Open Kernel Module for x86_64 570.133.20 Release Build` works fine

> Any fan monitor/control software? Not that I'm aware of

Is the plan to remove all the Copilot branded stuff from future releases of VS Codium? The reason I use Codium instead of Code is to be free of all...

I just retested and it didn't give the same error as before, but it now crashes during execution. However, I retested with the ROCm build for llama.cpp and the same...

Ran again, on a clean Ubuntu 24.04 setup with a ROCm 7.0.1 install & ZLUDA built from source, getting the same error as on my other system (which is Fedora...

Running with `export AMD_LOG_LEVEL=4 HIP_LAUNCH_BLOCKING=1 AMD_SERIALIZE_KERNEL=3 AMD_SERIALIZE_COPY=3` gave me the following: ``` ... :3:hip_device.cpp :656 : 33585513589 us: [pid:68741 tid: 0x7ed36a757000] hipGetDevicePropertiesR0600 ( 0x7ffc06033a60, 0 ) :3:hip_device.cpp :658 : 33585513607...

I've captured the trace logs for this here [zluda-llama-cpp-gemma3.tar.gz](https://github.com/user-attachments/files/22861746/zluda-llama-cpp-gemma3.tar.gz) As I mentioned on Discord - I've built ZLUDA and llama.cpp against ROCm 7.0.1, if this is complicating things, let me...

Hi, First of all, many thanks for the hard work, and apologies for the delay in validating this fix. Unfortunately I am now experiencing segfaults on my laptop (Fedora 43,...

Back on ROCm 7.1 with the latest version of ZLUDA built from source, I get no crashes but llama-bench gets stuck with 100% load on a single CPU core and...