{"model":"deepseek-4.1-flash","title":"PP8 brought up - 2.75x the KV pool of PP6","status":"positive","summary":"Layout-resolver intersection fix makes PP8 work: 9.35M-token KV pool (2.75x PP6), 8.9x concurrency @1M; 512k aggregate still ~60 tok/s @c32 -\u003e step-cost bound. Leads: vLLM #56120 (SM80 Triton knobs), SGLang #38646 (NVFP4 sparse-MLA, 384 B/tok, SM120-only).","occurred_at":"2026-09-12T15:18:00Z","config":{"serving":"TP1xPP8 (Zanooda 13-patch stack), fp8_ds_mla","engine":"zanooda/vllm-sm80-ds41f:v41-sm80 + KV-layout intersection fix"},"benchmarks":[{"metric":"kv_pool_tokens","value":9350000,"unit":"tokens","context":"PP8 (2.75x PP6)"},{"metric":"agg_decode_512k_tok_s","value":60,"unit":"tok/s","context":"PP8 c32 @512k - step-cost bound"}]}