llamaperf
Submit

Qwen 3.5-122B-A10B / AesSedai Q4_K_M / RADV native PP512 only

on AMD Strix Halo 128GB · llama.cpp / Vulkan RADV / build unknown

AMD hardware
Oct 13, 2026
Share
Throughput
259.0 tokens/s prompt processing (prefill)Prompt processing measures input prefill speed; it does not tell you the time to first token.
Quant
AesSedai Q4_K_M (GGUF)
Flash Attention
on
System RAM
128 GB

Summary

User reports Qwen3.5-122B-A10B at 259 t/s prompt processing on a single Radeon 8060S gfx1151 in an AMD Ryzen AI MAX+ 395 with 128 GB unified memory. Setup is the RADV backend with AesSedai Q4_K_M, full GPU offload at ngl 999, batch 2048, 32 threads and flash attention 1. The figure is a PP-only measurement over 512 prompt tokens; no TG result was supplied for this configuration.