experiment=ch10_latency_bandwidth_model
run_id=20260710T151843Z
timestamp_utc=2026-07-10T15:19:35.739681+00:00
hostname=sg-fuxc2-260708-12652-default0-0
cycles=10
iterations=30
sizes=1B..1GiB factor2
datatype=int8
world_size=4
cpu_affinity=0,2,4,6 (NUMA0)
performance_command=taskset -c 0,2,4,6 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 1 -e 1G -f 2 -g 4 -d int8 -o sum -z 0 -C 0 -a 3 -w 5 -n 30 -N 10 -c 1 -I 0
tuning_command=NCCL_DEBUG=INFO NCCL_DEBUG_SUBSYS=TUNING taskset -c 0,2,4,6 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 1 -e 1G -f 2 -g 4 -d int8 -o sum -z 0 -C 0 -a 3 -w 0 -n 1 -N 1 -c 0 -I 0
nccl_tests_commit=5bcd45d2ee8365e37895e118c9d11b0ebc15aa69
nccl_source_commit=178b6b759074597777ce13438efb0e0ba625e429