llamaperf

Step-3.5-Flash / b8118 Vulkan / PP512 TG128 at prepared depth8000

on AMD Radeon 8060S gfx1151 (Strix Halo; CPU SKU and installed RAM unknown) · llama.cpp

Oct 13, 2026
Throughput
24.6 t/s gen · 160.5 tokens/s prompt processing (prefill)Prompt processing is how fast the input is read, not how long you wait for the first word. How to measure time to first token.
Quant
IQ3_M (GGUF)
KV cache
Q8
Flash Attention
on
MTP (Multi-Token Prediction)
off

Summary

User reports Step-3.5-Flash IQ3_M at 24.6 t/s generation and 160.5 t/s prompt processing on a single Radeon 8060S gfx1151 device. Setup is llama.cpp build b8118 with IQ3_M, q8_0 KV cache, -ngl 999, -fa 1, -ub 2048 -b 2048 and depth 8000, run on the Vulkan/RADV BLAS arm. The headline pair comes from that arm only, with no cross-pairing; the ROCm/HIP arm measured 16.69 t/s TG128 and 87.66 t/s PP512. PP512 at depth 8000 is not an 8000-token prefill, and TG128 is 128 synthetic timed decode steps.

Compare this run

Medians of plain runs: one device, one request, no speculative decoding, the whole model in the device's memory. The report above is one setup, so read it beside them and not as one of them.

Every Step-3.5-Flash report →