DeepEP
DeepEP copied to clipboard
feat: remove recordStream of normal mode
Motivation
A possible way to remove record stream of normal mode by holding reference. Issue see #455 .
Test results
Test on downstream training task, no problems encountered. Test on 8 node, passed.
[testing] Running with BF16, without top-k (async=False, previous=False) ... passed
[testing] Running with BF16, with top-k (async=False, previous=False) ... passed
[testing] Running with BF16, without top-k (async=False, previous=False) ... passed
[testing] Running with BF16, with top-k (async=False, previous=False) ... passed
[testing] Running with FP8, without top-k (async=False, previous=False) ... passed
[testing] Running with FP8, with top-k (async=False, previous=False) ... passed
[testing] Running with FP8, without top-k (async=False, previous=False) ... passed
[testing] Running with FP8, with top-k (async=False, previous=False) ... passed
[testing] Running with BF16, without top-k (async=True, previous=False) ... passed
[testing] Running with BF16, with top-k (async=True, previous=False) ... passed
[testing] Running with BF16, without top-k (async=True, previous=False) ... passed
[testing] Running with BF16, with top-k (async=True, previous=False) ... passed
[testing] Running with FP8, without top-k (async=True, previous=False) ... passed
[testing] Running with FP8, with top-k (async=True, previous=False) ... passed
[testing] Running with FP8, without top-k (async=True, previous=False) ... passed
[testing] Running with FP8, with top-k (async=True, previous=False) ... passed
[testing] Running with BF16, without top-k (async=False, previous=True) ... passed
[testing] Running with BF16, with top-k (async=False, previous=True) ... passed
[testing] Running with BF16, without top-k (async=False, previous=True) ... passed
[testing] Running with BF16, with top-k (async=False, previous=True) ... passed
[testing] Running with FP8, without top-k (async=False, previous=True) ... passed
[testing] Running with FP8, with top-k (async=False, previous=True) ... passed
[testing] Running with FP8, without top-k (async=False, previous=True) ... passed
[testing] Running with FP8, with top-k (async=False, previous=True) ... passed
[testing] Running with BF16, without top-k (async=True, previous=True) ... passed
[testing] Running with BF16, with top-k (async=True, previous=True) ... passed
[testing] Running with BF16, without top-k (async=True, previous=True) ... passed
[testing] Running with BF16, with top-k (async=True, previous=True) ... passed
[testing] Running with FP8, without top-k (async=True, previous=True) ... passed
[testing] Running with FP8, with top-k (async=True, previous=True) ... passed
[testing] Running with FP8, without top-k (async=True, previous=True) ... passed
[testing] Running with FP8, with top-k (async=True, previous=True) ... passed
[tuning] SMs 24, NVL chunk 4, RDMA chunk 4, transmit: 2447.00 us, notify: 1994.00 us, BW: 49.24 GB/s (RDMA), 91.70 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 8, transmit: 2336.00 us, notify: 677.52 us, BW: 51.58 GB/s (RDMA), 96.06 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 12, transmit: 2339.00 us, notify: 658.60 us, BW: 51.51 GB/s (RDMA), 95.94 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 16, transmit: 2343.00 us, notify: 895.24 us, BW: 51.43 GB/s (RDMA), 95.77 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 20, transmit: 2357.00 us, notify: 989.53 us, BW: 51.12 GB/s (RDMA), 95.21 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 24, transmit: 2377.00 us, notify: 899.59 us, BW: 50.69 GB/s (RDMA), 94.40 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 28, transmit: 2398.00 us, notify: 919.23 us, BW: 50.25 GB/s (RDMA), 93.58 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 32, transmit: 2428.00 us, notify: 876.37 us, BW: 49.63 GB/s (RDMA), 92.42 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 4, transmit: 2446.00 us, notify: 428.26 us, BW: 49.26 GB/s (RDMA), 91.74 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 8, transmit: 2335.00 us, notify: 646.25 us, BW: 51.60 GB/s (RDMA), 96.10 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 12, transmit: 2327.00 us, notify: 751.40 us, BW: 51.78 GB/s (RDMA), 96.43 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 16, transmit: 2337.00 us, notify: 999.82 us, BW: 51.56 GB/s (RDMA), 96.02 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 20, transmit: 2348.00 us, notify: 944.60 us, BW: 51.32 GB/s (RDMA), 95.57 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 24, transmit: 2367.00 us, notify: 832.81 us, BW: 50.90 GB/s (RDMA), 94.80 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 28, transmit: 2389.00 us, notify: 933.12 us, BW: 50.44 GB/s (RDMA), 93.93 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 32, transmit: 2404.00 us, notify: 913.03 us, BW: 50.12 GB/s (RDMA), 93.34 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 4, transmit: 2439.00 us, notify: 318.43 us, BW: 49.40 GB/s (RDMA), 92.00 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 8, transmit: 2334.00 us, notify: 615.93 us, BW: 51.62 GB/s (RDMA), 96.14 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 12, transmit: 2326.00 us, notify: 779.14 us, BW: 51.80 GB/s (RDMA), 96.47 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 16, transmit: 2338.00 us, notify: 961.88 us, BW: 51.54 GB/s (RDMA), 95.98 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 20, transmit: 2344.00 us, notify: 973.96 us, BW: 51.40 GB/s (RDMA), 95.73 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 24, transmit: 2360.00 us, notify: 910.55 us, BW: 51.05 GB/s (RDMA), 95.08 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 28, transmit: 2381.00 us, notify: 920.34 us, BW: 50.60 GB/s (RDMA), 94.25 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 32, transmit: 2400.00 us, notify: 819.04 us, BW: 50.20 GB/s (RDMA), 93.50 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 4, transmit: 2435.00 us, notify: 308.75 us, BW: 49.48 GB/s (RDMA), 92.16 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 8, transmit: 2332.00 us, notify: 601.73 us, BW: 51.67 GB/s (RDMA), 96.23 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 12, transmit: 2329.00 us, notify: 711.45 us, BW: 51.73 GB/s (RDMA), 96.35 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 16, transmit: 2335.00 us, notify: 946.27 us, BW: 51.60 GB/s (RDMA), 96.10 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 20, transmit: 2347.00 us, notify: 923.43 us, BW: 51.34 GB/s (RDMA), 95.61 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 24, transmit: 2361.00 us, notify: 815.65 us, BW: 51.03 GB/s (RDMA), 95.04 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 28, transmit: 2380.00 us, notify: 868.54 us, BW: 50.63 GB/s (RDMA), 94.29 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 32, transmit: 2396.00 us, notify: 896.06 us, BW: 50.29 GB/s (RDMA), 93.66 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 4, transmit: 2443.00 us, notify: 315.88 us, BW: 49.32 GB/s (RDMA), 91.85 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 8, transmit: 2337.00 us, notify: 633.58 us, BW: 51.56 GB/s (RDMA), 96.02 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 12, transmit: 2327.00 us, notify: 744.71 us, BW: 51.78 GB/s (RDMA), 96.43 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 16, transmit: 2332.00 us, notify: 980.10 us, BW: 51.67 GB/s (RDMA), 96.23 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 20, transmit: 2346.00 us, notify: 972.22 us, BW: 51.36 GB/s (RDMA), 95.65 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 24, transmit: 2360.00 us, notify: 820.88 us, BW: 51.05 GB/s (RDMA), 95.08 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 28, transmit: 2382.00 us, notify: 872.97 us, BW: 50.58 GB/s (RDMA), 94.21 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 32, transmit: 2397.00 us, notify: 906.19 us, BW: 50.27 GB/s (RDMA), 93.62 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 4, transmit: 2440.00 us, notify: 294.39 us, BW: 49.38 GB/s (RDMA), 91.97 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 8, transmit: 2330.00 us, notify: 606.91 us, BW: 51.71 GB/s (RDMA), 96.31 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 12, transmit: 2382.00 us, notify: 775.80 us, BW: 50.58 GB/s (RDMA), 94.21 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 16, transmit: 2338.00 us, notify: 952.89 us, BW: 51.54 GB/s (RDMA), 95.98 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 20, transmit: 2344.00 us, notify: 979.52 us, BW: 51.40 GB/s (RDMA), 95.73 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 24, transmit: 2356.00 us, notify: 829.35 us, BW: 51.14 GB/s (RDMA), 95.25 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 28, transmit: 2383.00 us, notify: 886.39 us, BW: 50.56 GB/s (RDMA), 94.17 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 32, transmit: 2396.00 us, notify: 913.22 us, BW: 50.29 GB/s (RDMA), 93.66 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 4, transmit: 2446.00 us, notify: 300.67 us, BW: 49.26 GB/s (RDMA), 91.74 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 8, transmit: 2333.00 us, notify: 636.02 us, BW: 51.65 GB/s (RDMA), 96.18 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 12, transmit: 2326.00 us, notify: 756.02 us, BW: 51.80 GB/s (RDMA), 96.47 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 16, transmit: 2334.00 us, notify: 954.54 us, BW: 51.62 GB/s (RDMA), 96.14 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 20, transmit: 2343.00 us, notify: 967.82 us, BW: 51.43 GB/s (RDMA), 95.77 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 24, transmit: 2357.00 us, notify: 848.03 us, BW: 51.12 GB/s (RDMA), 95.21 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 28, transmit: 2375.00 us, notify: 1029.00 us, BW: 50.73 GB/s (RDMA), 94.48 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 32, transmit: 2395.00 us, notify: 917.50 us, BW: 50.31 GB/s (RDMA), 93.69 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 4, transmit: 2444.00 us, notify: 292.90 us, BW: 49.30 GB/s (RDMA), 91.82 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 8, transmit: 2335.00 us, notify: 629.24 us, BW: 51.60 GB/s (RDMA), 96.10 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 12, transmit: 2328.00 us, notify: 766.76 us, BW: 51.76 GB/s (RDMA), 96.39 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 16, transmit: 2332.00 us, notify: 944.27 us, BW: 51.67 GB/s (RDMA), 96.23 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 20, transmit: 2341.00 us, notify: 989.01 us, BW: 51.47 GB/s (RDMA), 95.86 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 24, transmit: 2360.00 us, notify: 879.64 us, BW: 51.05 GB/s (RDMA), 95.08 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 28, transmit: 2376.00 us, notify: 966.13 us, BW: 50.71 GB/s (RDMA), 94.44 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 32, transmit: 2403.00 us, notify: 916.29 us, BW: 50.14 GB/s (RDMA), 93.38 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 4, transmit: 2443.00 us, notify: 286.33 us, BW: 49.32 GB/s (RDMA), 91.85 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 8, transmit: 2335.00 us, notify: 624.65 us, BW: 51.60 GB/s (RDMA), 96.10 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 12, transmit: 2327.00 us, notify: 732.50 us, BW: 51.78 GB/s (RDMA), 96.43 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 16, transmit: 2338.00 us, notify: 928.79 us, BW: 51.54 GB/s (RDMA), 95.98 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 20, transmit: 2344.00 us, notify: 967.01 us, BW: 51.40 GB/s (RDMA), 95.73 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 24, transmit: 2357.00 us, notify: 912.57 us, BW: 51.12 GB/s (RDMA), 95.21 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 28, transmit: 2378.00 us, notify: 982.52 us, BW: 50.67 GB/s (RDMA), 94.36 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 32, transmit: 2402.00 us, notify: 884.01 us, BW: 50.16 GB/s (RDMA), 93.42 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 4, transmit: 2436.00 us, notify: 357.19 us, BW: 49.46 GB/s (RDMA), 92.12 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 8, transmit: 2336.00 us, notify: 689.41 us, BW: 51.58 GB/s (RDMA), 96.06 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 12, transmit: 2327.00 us, notify: 770.62 us, BW: 51.78 GB/s (RDMA), 96.43 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 16, transmit: 2338.00 us, notify: 943.69 us, BW: 51.54 GB/s (RDMA), 95.98 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 20, transmit: 2347.00 us, notify: 974.51 us, BW: 51.34 GB/s (RDMA), 95.61 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 24, transmit: 2355.00 us, notify: 866.46 us, BW: 51.16 GB/s (RDMA), 95.29 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 28, transmit: 2378.00 us, notify: 929.25 us, BW: 50.67 GB/s (RDMA), 94.36 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 32, transmit: 2401.00 us, notify: 902.59 us, BW: 50.18 GB/s (RDMA), 93.46 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 4, transmit: 2447.00 us, notify: 361.12 us, BW: 49.24 GB/s (RDMA), 91.70 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 8, transmit: 2331.00 us, notify: 672.83 us, BW: 51.69 GB/s (RDMA), 96.27 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 12, transmit: 2327.00 us, notify: 710.19 us, BW: 51.78 GB/s (RDMA), 96.43 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 16, transmit: 2336.00 us, notify: 866.34 us, BW: 51.58 GB/s (RDMA), 96.06 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 20, transmit: 2347.00 us, notify: 964.72 us, BW: 51.34 GB/s (RDMA), 95.61 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 24, transmit: 2356.00 us, notify: 893.31 us, BW: 51.14 GB/s (RDMA), 95.25 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 28, transmit: 2383.00 us, notify: 944.42 us, BW: 50.56 GB/s (RDMA), 94.17 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 32, transmit: 2400.00 us, notify: 957.04 us, BW: 50.20 GB/s (RDMA), 93.50 GB/s (NVL)
[tuning] Best dispatch (FP8): SMs 24, NVL chunk 12, RDMA chunk 12, transmit: 2326.00 us, notify: 779.14 us, BW: 51.80 GB/s (RDMA), 96.47 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 4, transmit: 4488.00 us, notify: 1193.00 us, BW: 52.07 GB/s (RDMA), 96.97 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 8, transmit: 4484.00 us, notify: 1884.00 us, BW: 52.11 GB/s (RDMA), 97.06 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 12, transmit: 4470.00 us, notify: 5016.00 us, BW: 52.28 GB/s (RDMA), 97.36 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 16, transmit: 30180.00 us, notify: 20559.00 us, BW: 7.74 GB/s (RDMA), 14.42 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 20, transmit: 23556.00 us, notify: 41167.00 us, BW: 9.92 GB/s (RDMA), 18.48 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 24, transmit: 18894.00 us, notify: 55680.00 us, BW: 12.37 GB/s (RDMA), 23.03 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 28, transmit: 10955.00 us, notify: 62392.00 us, BW: 21.33 GB/s (RDMA), 39.73 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 32, transmit: 6721.00 us, notify: 73571.00 us, BW: 34.77 GB/s (RDMA), 64.75 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 4, transmit: 4479.00 us, notify: 1135.00 us, BW: 52.17 GB/s (RDMA), 97.16 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 8, transmit: 4481.00 us, notify: 1876.00 us, BW: 52.15 GB/s (RDMA), 97.12 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 12, transmit: 4476.00 us, notify: 1996.00 us, BW: 52.21 GB/s (RDMA), 97.23 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 16, transmit: 4845.00 us, notify: 25621.00 us, BW: 48.23 GB/s (RDMA), 89.82 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 20, transmit: 9222.00 us, notify: 41026.00 us, BW: 25.34 GB/s (RDMA), 47.19 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 24, transmit: 5594.00 us, notify: 56775.00 us, BW: 41.77 GB/s (RDMA), 77.80 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 28, transmit: 7591.00 us, notify: 40542.00 us, BW: 30.78 GB/s (RDMA), 57.33 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8, RDMA chunk 32, transmit: 6151.00 us, notify: 59678.00 us, BW: 37.99 GB/s (RDMA), 70.75 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 4, transmit: 4490.00 us, notify: 1114.00 us, BW: 52.04 GB/s (RDMA), 96.93 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 8, transmit: 4486.00 us, notify: 1854.00 us, BW: 52.09 GB/s (RDMA), 97.01 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 12, transmit: 4553.00 us, notify: 2653.00 us, BW: 51.32 GB/s (RDMA), 95.58 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 16, transmit: 5091.00 us, notify: 28888.00 us, BW: 45.90 GB/s (RDMA), 85.48 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 20, transmit: 5379.00 us, notify: 29969.00 us, BW: 43.44 GB/s (RDMA), 80.91 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 24, transmit: 5604.00 us, notify: 57656.00 us, BW: 41.70 GB/s (RDMA), 77.66 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 28, transmit: 5759.00 us, notify: 40969.00 us, BW: 40.58 GB/s (RDMA), 75.57 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12, RDMA chunk 32, transmit: 5769.00 us, notify: 46458.00 us, BW: 40.51 GB/s (RDMA), 75.44 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 4, transmit: 4491.00 us, notify: 1155.00 us, BW: 52.03 GB/s (RDMA), 96.90 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 8, transmit: 4490.00 us, notify: 1827.00 us, BW: 52.04 GB/s (RDMA), 96.93 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 12, transmit: 4476.00 us, notify: 1944.00 us, BW: 52.21 GB/s (RDMA), 97.23 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 16, transmit: 5261.00 us, notify: 22370.00 us, BW: 44.42 GB/s (RDMA), 82.72 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 20, transmit: 5183.00 us, notify: 20125.00 us, BW: 45.09 GB/s (RDMA), 83.97 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 24, transmit: 5808.00 us, notify: 41073.00 us, BW: 40.23 GB/s (RDMA), 74.93 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 28, transmit: 5831.00 us, notify: 24971.00 us, BW: 40.07 GB/s (RDMA), 74.64 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16, RDMA chunk 32, transmit: 5839.00 us, notify: 44763.00 us, BW: 40.02 GB/s (RDMA), 74.53 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 4, transmit: 4492.00 us, notify: 1147.00 us, BW: 52.02 GB/s (RDMA), 96.88 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 8, transmit: 4483.00 us, notify: 1882.00 us, BW: 52.13 GB/s (RDMA), 97.08 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 12, transmit: 4468.00 us, notify: 2183.00 us, BW: 52.30 GB/s (RDMA), 97.40 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 16, transmit: 6100.00 us, notify: 22332.00 us, BW: 38.31 GB/s (RDMA), 71.34 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 20, transmit: 5698.00 us, notify: 30857.00 us, BW: 41.01 GB/s (RDMA), 76.38 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 24, transmit: 5681.00 us, notify: 32847.00 us, BW: 41.13 GB/s (RDMA), 76.61 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 28, transmit: 6789.00 us, notify: 19764.00 us, BW: 34.42 GB/s (RDMA), 64.10 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20, RDMA chunk 32, transmit: 6262.00 us, notify: 33909.00 us, BW: 37.32 GB/s (RDMA), 69.50 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 4, transmit: 4487.00 us, notify: 1140.00 us, BW: 52.08 GB/s (RDMA), 96.99 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 8, transmit: 4487.00 us, notify: 1804.00 us, BW: 52.08 GB/s (RDMA), 96.99 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 12, transmit: 4560.00 us, notify: 1890.00 us, BW: 51.24 GB/s (RDMA), 95.44 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 16, transmit: 5213.00 us, notify: 17296.00 us, BW: 44.83 GB/s (RDMA), 83.48 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 20, transmit: 5446.00 us, notify: 17394.00 us, BW: 42.91 GB/s (RDMA), 79.91 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 24, transmit: 6398.00 us, notify: 28617.00 us, BW: 36.52 GB/s (RDMA), 68.02 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 28, transmit: 5749.00 us, notify: 9062.00 us, BW: 40.65 GB/s (RDMA), 75.70 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24, RDMA chunk 32, transmit: 5908.00 us, notify: 23289.00 us, BW: 39.55 GB/s (RDMA), 73.66 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 4, transmit: 4485.00 us, notify: 1142.00 us, BW: 52.10 GB/s (RDMA), 97.03 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 8, transmit: 4488.00 us, notify: 1796.00 us, BW: 52.07 GB/s (RDMA), 96.97 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 12, transmit: 4469.00 us, notify: 1943.00 us, BW: 52.29 GB/s (RDMA), 97.38 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 16, transmit: 5193.00 us, notify: 12090.00 us, BW: 45.00 GB/s (RDMA), 83.80 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 20, transmit: 5436.00 us, notify: 11975.00 us, BW: 42.99 GB/s (RDMA), 80.06 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 24, transmit: 5700.00 us, notify: 22701.00 us, BW: 41.00 GB/s (RDMA), 76.35 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 28, transmit: 5801.00 us, notify: 10570.00 us, BW: 40.28 GB/s (RDMA), 75.02 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28, RDMA chunk 32, transmit: 5847.00 us, notify: 13908.00 us, BW: 39.97 GB/s (RDMA), 74.43 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 4, transmit: 4485.00 us, notify: 1130.00 us, BW: 52.10 GB/s (RDMA), 97.03 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 8, transmit: 4481.00 us, notify: 1811.00 us, BW: 52.15 GB/s (RDMA), 97.12 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 12, transmit: 4466.00 us, notify: 1993.00 us, BW: 52.32 GB/s (RDMA), 97.45 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 16, transmit: 5549.00 us, notify: 10151.00 us, BW: 42.11 GB/s (RDMA), 78.43 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 20, transmit: 5326.00 us, notify: 8242.00 us, BW: 43.87 GB/s (RDMA), 81.71 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 24, transmit: 6027.00 us, notify: 17828.00 us, BW: 38.77 GB/s (RDMA), 72.21 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 28, transmit: 6265.00 us, notify: 5787.00 us, BW: 37.30 GB/s (RDMA), 69.46 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32, RDMA chunk 32, transmit: 5896.00 us, notify: 14804.00 us, BW: 39.63 GB/s (RDMA), 73.81 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 4, transmit: 4491.00 us, notify: 1121.00 us, BW: 52.03 GB/s (RDMA), 96.90 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 8, transmit: 4491.00 us, notify: 1794.00 us, BW: 52.03 GB/s (RDMA), 96.90 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 12, transmit: 4469.00 us, notify: 1983.00 us, BW: 52.29 GB/s (RDMA), 97.38 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 16, transmit: 5303.00 us, notify: 9594.00 us, BW: 44.07 GB/s (RDMA), 82.07 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 20, transmit: 5758.00 us, notify: 12371.00 us, BW: 40.58 GB/s (RDMA), 75.58 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 24, transmit: 5762.00 us, notify: 14330.00 us, BW: 40.55 GB/s (RDMA), 75.53 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 28, transmit: 5914.00 us, notify: 6499.00 us, BW: 39.51 GB/s (RDMA), 73.59 GB/s (NVL)
[tuning] SMs 24, NVL chunk 36, RDMA chunk 32, transmit: 5840.00 us, notify: 10584.00 us, BW: 40.01 GB/s (RDMA), 74.52 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 4, transmit: 4489.00 us, notify: 1130.00 us, BW: 52.06 GB/s (RDMA), 96.95 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 8, transmit: 4490.00 us, notify: 1854.00 us, BW: 52.04 GB/s (RDMA), 96.93 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 12, transmit: 4474.00 us, notify: 1940.00 us, BW: 52.23 GB/s (RDMA), 97.27 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 16, transmit: 4871.00 us, notify: 8777.00 us, BW: 47.97 GB/s (RDMA), 89.34 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 20, transmit: 5407.00 us, notify: 3117.00 us, BW: 43.22 GB/s (RDMA), 80.49 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 24, transmit: 5619.00 us, notify: 12994.00 us, BW: 41.59 GB/s (RDMA), 77.45 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 28, transmit: 5537.00 us, notify: 2889.00 us, BW: 42.20 GB/s (RDMA), 78.60 GB/s (NVL)
[tuning] SMs 24, NVL chunk 40, RDMA chunk 32, transmit: 5886.00 us, notify: 9079.00 us, BW: 39.70 GB/s (RDMA), 73.94 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 4, transmit: 4488.00 us, notify: 1131.00 us, BW: 52.07 GB/s (RDMA), 96.97 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 8, transmit: 4484.00 us, notify: 1840.00 us, BW: 52.11 GB/s (RDMA), 97.06 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 12, transmit: 4531.00 us, notify: 1940.00 us, BW: 51.57 GB/s (RDMA), 96.05 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 16, transmit: 5233.00 us, notify: 9009.00 us, BW: 44.65 GB/s (RDMA), 83.16 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 20, transmit: 5412.00 us, notify: 6282.00 us, BW: 43.18 GB/s (RDMA), 80.41 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 24, transmit: 5853.00 us, notify: 11069.00 us, BW: 39.92 GB/s (RDMA), 74.35 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 28, transmit: 5663.00 us, notify: 3839.00 us, BW: 41.26 GB/s (RDMA), 76.85 GB/s (NVL)
[tuning] SMs 24, NVL chunk 44, RDMA chunk 32, transmit: 5866.00 us, notify: 16774.00 us, BW: 39.84 GB/s (RDMA), 74.19 GB/s (NVL)
[tuning] Best dispatch (BF16): SMs 24, NVL chunk 32, RDMA chunk 12, transmit: 4466.00 us, notify: 1993.00 us, BW: 52.32 GB/s (RDMA), 97.45 GB/s (NVL)
[tuning] SMs 24, NVL chunk 1, RDMA chunk 8, transmit: 4655.00 us, notify: 1395.00 us, BW: 50.20 GB/s (RDMA), 93.49 GB/s (NVL)
[tuning] SMs 24, NVL chunk 1, RDMA chunk 12, transmit: 4714.00 us, notify: 1691.00 us, BW: 49.57 GB/s (RDMA), 92.32 GB/s (NVL)
[tuning] SMs 24, NVL chunk 1, RDMA chunk 16, transmit: 4730.00 us, notify: 1644.00 us, BW: 49.40 GB/s (RDMA), 92.01 GB/s (NVL)
[tuning] SMs 24, NVL chunk 1, RDMA chunk 20, transmit: 4800.00 us, notify: 1688.00 us, BW: 48.68 GB/s (RDMA), 90.67 GB/s (NVL)
[tuning] SMs 24, NVL chunk 1, RDMA chunk 24, transmit: 4895.00 us, notify: 1647.00 us, BW: 47.74 GB/s (RDMA), 88.91 GB/s (NVL)
[tuning] SMs 24, NVL chunk 1, RDMA chunk 28, transmit: 4968.00 us, notify: 2011.00 us, BW: 47.04 GB/s (RDMA), 87.60 GB/s (NVL)
[tuning] SMs 24, NVL chunk 1, RDMA chunk 32, transmit: 5151.00 us, notify: 6059.00 us, BW: 45.37 GB/s (RDMA), 84.49 GB/s (NVL)
[tuning] SMs 24, NVL chunk 2, RDMA chunk 8, transmit: 4691.00 us, notify: 1446.00 us, BW: 49.81 GB/s (RDMA), 92.77 GB/s (NVL)
[tuning] SMs 24, NVL chunk 2, RDMA chunk 12, transmit: 4745.00 us, notify: 1604.00 us, BW: 49.25 GB/s (RDMA), 91.72 GB/s (NVL)
[tuning] SMs 24, NVL chunk 2, RDMA chunk 16, transmit: 4755.00 us, notify: 1746.00 us, BW: 49.14 GB/s (RDMA), 91.52 GB/s (NVL)
[tuning] SMs 24, NVL chunk 2, RDMA chunk 20, transmit: 4838.00 us, notify: 1680.00 us, BW: 48.30 GB/s (RDMA), 89.95 GB/s (NVL)
[tuning] SMs 24, NVL chunk 2, RDMA chunk 24, transmit: 4925.00 us, notify: 1840.00 us, BW: 47.45 GB/s (RDMA), 88.37 GB/s (NVL)
[tuning] SMs 24, NVL chunk 2, RDMA chunk 28, transmit: 4993.00 us, notify: 3131.00 us, BW: 46.80 GB/s (RDMA), 87.16 GB/s (NVL)
[tuning] SMs 24, NVL chunk 2, RDMA chunk 32, transmit: 5034.00 us, notify: 6706.00 us, BW: 46.42 GB/s (RDMA), 86.45 GB/s (NVL)
[tuning] SMs 24, NVL chunk 3, RDMA chunk 8, transmit: 4750.00 us, notify: 1343.00 us, BW: 49.20 GB/s (RDMA), 91.62 GB/s (NVL)
[tuning] SMs 24, NVL chunk 3, RDMA chunk 12, transmit: 4823.00 us, notify: 1552.00 us, BW: 48.45 GB/s (RDMA), 90.23 GB/s (NVL)
[tuning] SMs 24, NVL chunk 3, RDMA chunk 16, transmit: 4793.00 us, notify: 1626.00 us, BW: 48.75 GB/s (RDMA), 90.80 GB/s (NVL)
[tuning] SMs 24, NVL chunk 3, RDMA chunk 20, transmit: 4888.00 us, notify: 1623.00 us, BW: 47.81 GB/s (RDMA), 89.03 GB/s (NVL)
[tuning] SMs 24, NVL chunk 3, RDMA chunk 24, transmit: 4946.00 us, notify: 1502.00 us, BW: 47.25 GB/s (RDMA), 87.99 GB/s (NVL)
[tuning] SMs 24, NVL chunk 3, RDMA chunk 28, transmit: 5019.00 us, notify: 1826.00 us, BW: 46.56 GB/s (RDMA), 86.71 GB/s (NVL)
[tuning] SMs 24, NVL chunk 3, RDMA chunk 32, transmit: 5191.00 us, notify: 7244.00 us, BW: 45.02 GB/s (RDMA), 83.84 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 8, transmit: 4743.00 us, notify: 1483.00 us, BW: 49.27 GB/s (RDMA), 91.76 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 12, transmit: 4801.00 us, notify: 1788.00 us, BW: 48.67 GB/s (RDMA), 90.65 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 16, transmit: 4881.00 us, notify: 1681.00 us, BW: 47.87 GB/s (RDMA), 89.16 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 20, transmit: 4871.00 us, notify: 1716.00 us, BW: 47.97 GB/s (RDMA), 89.34 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 24, transmit: 4992.00 us, notify: 1542.00 us, BW: 46.81 GB/s (RDMA), 87.18 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 28, transmit: 5070.00 us, notify: 1819.00 us, BW: 46.09 GB/s (RDMA), 85.84 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4, RDMA chunk 32, transmit: 5135.00 us, notify: 2063.00 us, BW: 45.51 GB/s (RDMA), 84.75 GB/s (NVL)
[tuning] SMs 24, NVL chunk 5, RDMA chunk 8, transmit: 4824.00 us, notify: 1526.00 us, BW: 48.44 GB/s (RDMA), 90.22 GB/s (NVL)
[tuning] SMs 24, NVL chunk 5, RDMA chunk 12, transmit: 4816.00 us, notify: 1359.00 us, BW: 48.52 GB/s (RDMA), 90.37 GB/s (NVL)
[tuning] SMs 24, NVL chunk 5, RDMA chunk 16, transmit: 4937.00 us, notify: 1597.00 us, BW: 47.33 GB/s (RDMA), 88.15 GB/s (NVL)
[tuning] SMs 24, NVL chunk 5, RDMA chunk 20, transmit: 4943.00 us, notify: 1700.00 us, BW: 47.27 GB/s (RDMA), 88.04 GB/s (NVL)
[tuning] SMs 24, NVL chunk 5, RDMA chunk 24, transmit: 4990.00 us, notify: 1815.00 us, BW: 46.83 GB/s (RDMA), 87.21 GB/s (NVL)
[tuning] SMs 24, NVL chunk 5, RDMA chunk 28, transmit: 5088.00 us, notify: 1960.00 us, BW: 45.93 GB/s (RDMA), 85.53 GB/s (NVL)
[tuning] SMs 24, NVL chunk 5, RDMA chunk 32, transmit: 5232.00 us, notify: 5364.00 us, BW: 44.66 GB/s (RDMA), 83.18 GB/s (NVL)
[tuning] SMs 24, NVL chunk 6, RDMA chunk 8, transmit: 4866.00 us, notify: 1636.00 us, BW: 48.02 GB/s (RDMA), 89.44 GB/s (NVL)
[tuning] SMs 24, NVL chunk 6, RDMA chunk 12, transmit: 4911.00 us, notify: 1478.00 us, BW: 47.58 GB/s (RDMA), 88.62 GB/s (NVL)
[tuning] SMs 24, NVL chunk 6, RDMA chunk 16, transmit: 4864.00 us, notify: 1553.00 us, BW: 48.04 GB/s (RDMA), 89.47 GB/s (NVL)
[tuning] SMs 24, NVL chunk 6, RDMA chunk 20, transmit: 4988.00 us, notify: 1658.00 us, BW: 46.85 GB/s (RDMA), 87.25 GB/s (NVL)
[tuning] SMs 24, NVL chunk 6, RDMA chunk 24, transmit: 5068.00 us, notify: 2561.00 us, BW: 46.11 GB/s (RDMA), 85.87 GB/s (NVL)
[tuning] SMs 24, NVL chunk 6, RDMA chunk 28, transmit: 5273.00 us, notify: 3051.00 us, BW: 44.32 GB/s (RDMA), 82.53 GB/s (NVL)
[tuning] SMs 24, NVL chunk 6, RDMA chunk 32, transmit: 5165.00 us, notify: 1862.00 us, BW: 45.24 GB/s (RDMA), 84.26 GB/s (NVL)
[tuning] SMs 24, NVL chunk 7, RDMA chunk 8, transmit: 4917.00 us, notify: 1648.00 us, BW: 47.52 GB/s (RDMA), 88.51 GB/s (NVL)
[tuning] SMs 24, NVL chunk 7, RDMA chunk 12, transmit: 4961.00 us, notify: 1617.00 us, BW: 47.10 GB/s (RDMA), 87.72 GB/s (NVL)
[tuning] SMs 24, NVL chunk 7, RDMA chunk 16, transmit: 4955.00 us, notify: 1527.00 us, BW: 47.16 GB/s (RDMA), 87.83 GB/s (NVL)
[tuning] SMs 24, NVL chunk 7, RDMA chunk 20, transmit: 5005.00 us, notify: 1530.00 us, BW: 46.69 GB/s (RDMA), 86.95 GB/s (NVL)
[tuning] SMs 24, NVL chunk 7, RDMA chunk 24, transmit: 5098.00 us, notify: 3796.00 us, BW: 45.84 GB/s (RDMA), 85.37 GB/s (NVL)
[tuning] SMs 24, NVL chunk 7, RDMA chunk 28, transmit: 5166.00 us, notify: 1930.00 us, BW: 45.23 GB/s (RDMA), 84.24 GB/s (NVL)
[tuning] SMs 24, NVL chunk 7, RDMA chunk 32, transmit: 5214.00 us, notify: 2544.00 us, BW: 44.82 GB/s (RDMA), 83.47 GB/s (NVL)
[tuning] Best combine: SMs 24, NVL chunk 1, RDMA chunk 8, transmit: 4655.00 us, notify: 1395.00 us, BW: 50.20 GB/s (RDMA), 93.49 GB/s (NVL)