Benchmarking Qwen3.5 122B on ROCmFP4
RedParaglider · reddit · 2026-07-16
This Reddit post shares benchmark results for the ROCmFP4 / iMatrix version of Qwen3.5 122B-A10B:
- Notes that this is based on the charlie12345/ROCmFPX solution, and the author mentions that native llama.cpp does not currently support it directly.
- Key metrics include:
- 122B total parameters, 10B activated parameters
- 60.70 GiB VRAM usage
- Decode 28.505 tok/s, marked as approximately 36.89% faster
- 13.47GB smaller than the original version
- The main value of this information is demonstrating the quantization and inference optimization effects of Qwen LLMs within the AMD / ROCm ecosystem.
Related event: Local Benchmark: Qwen3.5 122B ROCmFP4 Quantization(2 posts)→
More from Infra
- NVIDIA publishes Vera CPU architecture details before AMD’s AI event — ryanshrout · 2026-07-22
- oMLX 0.5.2 adds Mac menu-bar stats, low-bit decode kernels, and faster downloads — awnihannun · 2026-07-22
- Strangeworks launches Aura to turn enterprise ops into production optimization systems — whurley · 2026-07-22
- Graph workload 854.graph500 enters SPEC CPU 2026 as a new CPU benchmark — Prof_DavidBader · 2026-07-22
- HilbertRaum open-sources a fully local AI chat and document analysis app for private use — Vladowski · 2026-07-22
- Hybrid and local inference are emerging as a response to AI energy and token costs — dmitry140 · 2026-07-22