Tenstorrent Blackhole p150 2-card data-parallel Llama-3.1-8B aggregate (vendor b32)
1190.0 tok/s aggregate, 2x p150 data-parallel, batch-32
Fields
id
benchmarks/comm-tt-blackhole-p150-2x-llama31-8b-aggMetric
tps_aggregate
Value
1,190
Unit
tok/s aggregate, 2x p150 data-parallel, batch-32
Model
Llama-3.1-8B
Methodology
Vendor tt-metal model matrix, 2x p150 DP, batch-32 decode, ISL=128
Measured by
vendor
Thermal state
unknown
notes
VENDOR-MEASURED. Single p150 was 1059 t/s agg -> 2x DP scales ~1.12x not 2x. Blackhole software marked under active dev.
Notes
VENDOR-MEASURED. Single p150 was 1059 t/s agg -> 2x DP scales ~1.12x not 2x. Blackhole software marked under active dev.