Strix Halo Home Lab llama.cpp performance page: Qwen3-30B-A3B UD-Q4_K_XL depth sweep (ROCm 7.10 vs Vulkan)

Page documents llama.cpp b6702, ROCm 7.10.0a20251011 gfx1151 and Mesa 25.2.5 RADV, on a Framework Desktop / GMKtec EVO-X2 class Strix Halo box. llama-bench command is printed: -fa 1 -r 1 --mmap 0 -m Qwen3-30B-A3B-UD-Q4_K_XL.gguf, and the same with -d 130560. Tables give pp512 and tg128 at depth 0 and at depth 130560 for ROCm, ROCm+hipBLASLt, and RADV. Distinct from the short-context rows already ingested from the same wiki under strix-halo-homelab-wiki-llama-cpp-performance. Third-party measured, per-request batch-1.