Home › Evidence › Records › cfg-0058

cfg-0058

backfilled
citable URL: https://halobench.com/records/cfg-0058/ — this address never moves; the anchor /records/#cfg-0058 keeps resolving
modelNVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_M-00001-of-00003.gguf · UD-Q4_K_M
engineggml-org/llama.cpp 3653e6d · rocm · host aihydra (igpu)
flags-ngl 999 -fa 1 -b 2048 -ub 512 -ctk q8_0 -ctv q8_0 -t 16
templatenot recorded at test time
treeupstream — stock
aged evidence — reconstructed from the archive. cfg-0057 with quantised KV — the second half of the perf-matrix-gtt120 re-test (2026-08-14, amdgpu.gttsize=122880). Halving the KV footprint via q8_0 did not avoid the OOM either: see run-0235. Same recording gaps as cfg-0057 — no memory block, machine-asserted build only.

Capability basis

unmeasured — performance numbers on this config stand on a guard alone, not an established capability

Runs on this config (1)

rundatekindsuitemetrics
run-02352026-08-14performanceperf-matrix-gtt120@1oom=true