AMD Zen 5 (Turin, 32 cores pinned to one NUMA node)

4 records across 1 models (quants: BF16-w8a8).

hardwaremodelquantbackendtok/spp tok/sWttft sctx/tgdateclassflagssource
AMD Zen 5 (Turin, 32 cores pinned to one NUMA node)gpt-oss 20B BF16-w8a8vllm375.74–––1282026-09-30community–source
AMD Zen 5 (Turin, 32 cores pinned to one NUMA node)gpt-oss 20B BF16-w8a8vllm369.79–––1282026-09-30community–source
AMD Zen 5 (Turin, 32 cores pinned to one NUMA node)gpt-oss 20B BF16-w8a8vllm158.21–––1282026-09-30community–source
AMD Zen 5 (Turin, 32 cores pinned to one NUMA node)gpt-oss 20B BF16-w8a8vllm154.23–––1282026-09-30community–source

All rows cite their source. Provenance: community 40018 sourced 1085

← all hardware