experiment=ch14_protocol_wire_sync run_id=20260710T161632Z timestamp_utc=2026-07-10T16:25:59.886814+00:00 hostname=sg-fuxc2-260708-12652-default0-0 world_size=4 algorithms=Ring,Tree protocols=LL,LL128,Simple sizes=4B..1GiB_factor2 channels=12 cycles=5 iterations=20 main_rows=1740 alignment_rows=160 alignment_cycles=10 alignment_protocol=Ring+LL128 unalign_float_elements=0,1,2,3 practical_tie_percent=5.0 nsys_profiles=6 resume_existing_raw=1 private_nsys_dir=/root/nccl-learning/private/ch14_protocol/20260710T161632Z nccl_tests_commit=5bcd45d2ee8365e37895e118c9d11b0ebc15aa69 nccl_source_commit=178b6b759074597777ce13438efb0e0ba625e429 command_01=NCCL_ALGO=Ring NCCL_PROTO=LL NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum command_02=NCCL_ALGO=Ring NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum command_03=NCCL_ALGO=Ring NCCL_PROTO=Simple NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum command_04=NCCL_ALGO=Tree NCCL_PROTO=LL NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum command_05=NCCL_ALGO=Tree NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum command_06=NCCL_ALGO=Tree NCCL_PROTO=Simple NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum command_07=NCCL_ALGO=Tree NCCL_PROTO=Simple NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum command_08=NCCL_ALGO=Tree NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum command_09=NCCL_ALGO=Tree NCCL_PROTO=LL NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum command_10=NCCL_ALGO=Ring NCCL_PROTO=Simple NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum command_11=NCCL_ALGO=Ring NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum command_12=NCCL_ALGO=Ring NCCL_PROTO=LL NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 4 -e 1G -f 2 -g 4 -w 5 -n 20 -N 5 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum alignment_command_01=NCCL_ALGO=Ring NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 1M -e 64M -f 64 -g 4 -w 5 -n 20 -N 10 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum alignment_command_02=NCCL_ALGO=Ring NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 1M -e 64M -f 64 -g 4 -w 5 -n 20 -N 10 -c 1 -I 0 -z 0 -u 1 -C 0 -a 3 -d float -o sum alignment_command_03=NCCL_ALGO=Ring NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 1M -e 64M -f 64 -g 4 -w 5 -n 20 -N 10 -c 1 -I 0 -z 0 -u 2 -C 0 -a 3 -d float -o sum alignment_command_04=NCCL_ALGO=Ring NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 1M -e 64M -f 64 -g 4 -w 5 -n 20 -N 10 -c 1 -I 0 -z 0 -u 3 -C 0 -a 3 -d float -o sum alignment_command_05=NCCL_ALGO=Ring NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 1M -e 64M -f 64 -g 4 -w 5 -n 20 -N 10 -c 1 -I 0 -z 0 -u 3 -C 0 -a 3 -d float -o sum alignment_command_06=NCCL_ALGO=Ring NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 1M -e 64M -f 64 -g 4 -w 5 -n 20 -N 10 -c 1 -I 0 -z 0 -u 2 -C 0 -a 3 -d float -o sum alignment_command_07=NCCL_ALGO=Ring NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 1M -e 64M -f 64 -g 4 -w 5 -n 20 -N 10 -c 1 -I 0 -z 0 -u 1 -C 0 -a 3 -d float -o sum alignment_command_08=NCCL_ALGO=Ring NCCL_PROTO=LL128 NCCL_MIN_NCHANNELS=12 NCCL_MAX_NCHANNELS=12 /root/nccl-learning/third_party/nccl-tests/build/all_reduce_perf -b 1M -e 64M -f 64 -g 4 -w 5 -n 20 -N 10 -c 1 -I 0 -z 0 -u 0 -C 0 -a 3 -d float -o sum