Qwen3.8 Flash CIRU Orca v5.0, 256000-depth cached append

on AMD Strix Halo 128GB · CIRU v5 native HIP · 262,144 ctx

Oct 11, 2026

Throughput

56.5 t/s gen · 655.2 tokens/s prompt processing (prefill)Prompt processing measures input prefill speed; it does not tell you the time to first token.

Quant

CIRU v5 compact indexed affine (experts 4.5175 bpw) (GGUF)

KV cache

F16

MTP (Multi-Token Prediction)

System RAM

128 GB

Use cases

long-context

Summary

User reports Qwen3.8-Flash-CIRU-STRIX-Orca v5.0 at 56.5 t/s generation and 655.2 t/s prompt processing on one AMD Ryzen AI Max+ 395 / Radeon 8060S system with 128 GiB shared unified memory. Setup is CIRU v5 native HIP with compact indexed-affine experts at 4.5175 bpw, F16 target and draft attention cache, a matching Q8 Orca MTP drafter at depth 4, one slot and 262144-token configured context. The run reused all but four tokens of a 256000-token prefix, then appended 2048 tokens and generated 512 output tokens; MTP accepted 408 of 408 proposed tokens. Full append-plus-generation wall time was 12.409 seconds, with no repeated 256K result.