Wormhole n150 tt-metal: Mamba-2.8B batch-32 tok/s per user
14.1 tok/s per user at batch 32; 451.2 tok/s aggregate
Fields
id
benchmarks/ttmetal-n150-mamba-2p8b-b32-tpsuMetric
tok_s_per_user
Value
14.1
Unit
tok/s per user at batch 32; 451.2 tok/s aggregate
Model
Mamba-2.8B
Methodology
VENDOR demo in tenstorrent/tt-metal models/README.md (matrix last updated 2025-12-08). Mamba 2.8B, single n150 (Wormhole), batch 32, TT-Metalium v0.59.0-rc38. ttft 35 ms. t/s/u 14.1 is 1/inter-token-latency of the first generated token (per user); aggregate t/s is 451.2 = 14.1 * 32 concurrent requests, NOT a per-request rate. Mamba is the one model in the table not pinned to input sequence length 128. Per-user figure recorded; aggregate noted in unit.
Measured by
vendor
Reproducible
yes
notes
Vendor demo numbers, not a vLLM serving trace. The README states t/s = t/s/u * batch and that vLLM-server performance may differ. LLM rows use input sequence length 128 except Mamba, which accepts any length. Starred Blackhole figures carry the vendor's own caveat that the software is still under active optimization.
Notes
Vendor demo numbers, not a vLLM serving trace. The README states t/s = t/s/u * batch and that vLLM-server performance may differ. LLM rows use input sequence length 128 except Mamba, which accepts any length. Starred Blackhole figures carry the vendor's own caveat that the software is still under active optimization.