native PP512 and separate TG128

on AMD Radeon 8060S integrated GPU (Strix Halo; installed RAM unreported) · llama.cpp b9360 / 6b4e4bd58 / Vulkan RADV

Oct 14, 2026

Throughput

22.2 t/s gen · 241.4 tokens/s prompt processing (prefill)Prompt processing measures input prefill speed; it does not tell you the time to first token.

Quant

UD-IQ4_XS (GGUF)

KV cache

Flash Attention

Summary

User reports Step 3.7 Flash at 22.2 t/s generation and 241.4 t/s prompt processing on an AMD Radeon 8060S (RADV STRIX_HALO) iGPU. Setup is llama-bench with UD-IQ4_XS, q8_0 KV cache for both types, FA 1, ngl 999, batch 2048 and eight CPU threads. At d0 the run measured PP512 241.44 t/s and TG128 22.19 t/s; generation fell to 18.23 t/s at d65536.