verifiedmedium · 0.75updated 2026-10-15

EPYC 9554 (Genoa, 12ch DDR5) QwQ-32B Q4_K_M decode

13.9 tok/s decode, single-stream

Fields

benchmarks/comm-epyc-9554-qwq-32b-q4-decode

Metric

decode_tok_s

Value

13.9

Unit

tok/s decode, single-stream

Model

QwQ-32B Q4_K_M

Methodology

llama.cpp llama-bench on EPYC 9554, 12x DDR5-5600 fully populated, --numa distribute

Measured by

third_party

Thermal state

unknown

notes

ahelpme.com measured table. Fully populated 12-channel host required - see channel-population gotcha.

ahelpme.com measured table. Fully populated 12-channel host required - see channel-population gotcha.