DeepSeek V4.1 Flash on 8x 8 GB (Zanooda PP8)/Confirmation 1
From CMP Wiki
More actions
| Confirms | DeepSeek V4.1 Flash on 8x 8 GB (Zanooda PP8) |
|---|---|
| Result | Partly reproduced |
| By | @darkmaniac7 |
| Date | 2026-10-27 |
| Cards | 8 GB |
| Numbers | 180 W: one user code 64 / structured 46 / prose 32 tok/s; 8 users 546 / 316 / 285 aggregate; prefill 5,478 tok/s (17k prompt). 150 W: 65 / 50 / 34; 533 / 296 / 284; prefill 5,276; 1M context, KV pool 5.56M tokens. |
Eight cards at Gen2 x16 with P2P, Engram from disk (125 GB host RAM at the time), 1,024-token answers. Prefill, aggregate throughput, 1M context and image input match the author; single-stream decode is about half the author's 117 tok/s. The first run ended with the NVMe dropping off (see the page's warning). Source: @darkmaniac7, 2026-10-27 (lab notes)