Qwen3.6 35B on a 780M iGPU

On a ThinkPad T14 Gen 5 (AMD 8840U, Radeon 780M, 64GB DDR5 5600 MT/s), Qwen3.6 35B-A3B GGUF was tested using llama.cpp's Vulkan backend.

The llama-bench configuration was as follows.

• -hf AesSedai/Qwen3.6-35B-A3B-GGUF:Q6_K
• -fa 1
• -ub 1024
• -b 1024
• -p 1024 -n 128 -mmp 0

The benchmark results were pp1024 282.40 t/s and tg128 20.74 t/s. According to the output, Vulkan detected the AMD Radeon 780M Graphics (RADV PHOENIX), and it ran with 99 layers offloaded.

The author noted that running at Q6 required increasing GTT and adjusting the hang timeout, and stated that it also worked normally with full context.