Post #1389003
2026-04-19 09:50 UTC
Replies (3)
-
@stoeps@infosec.exchange 2026-04-19 10:00
@xahteiwi@mastodon.social when I look at the last 6 months and my amd-gpu issues I can feel with him. Multiple kernel panics, freezes or automatic reboots a day. Followed pretty every distro issue tracker and went back to Kernel 6.17.8, now at least 24 hours stable. All newer kernels have issues, downgrade of the amd-gpu firmware helps a bit, but not completely. If the next 24 hours keep stable, I will write a blog post on this. Since AMD announced a 30% performance boost (October 2025) it's a big pale of 💩
-
@masek@infosec.exchange 2026-04-19 15:14
@xahteiwi@mastodon.social ROCm definitely brings back the experimental feeling of Kernels of old 😁
-
@theuni@social.tchncs.de 2026-04-19 18:27
@xahteiwi@mastodon.social I've spend 6 (or maybe nearly 9) months trying to get ollama/llama-cpp/vllm working properly with AMD workstation GPUs (W7900). it's been tiring. Model support is hit and miss, stability is not that, and performance doesn't scale. We had to cut our losses (for now) and spend money with nvidia to unblock us... :/ ... would love to revisit this at some point, but right now ... it's like the readme says.