-
Notifications
You must be signed in to change notification settings - Fork 60
Benchmarks UCX DGX2
Benjamin Zaitlen edited this page Dec 18, 2019
·
4 revisions
Merge benchmark
--------------------------
Chunk-size | 100000000
Frac-match | 0.3
Ignore-size | 1.05 MB
Protocol | ucx
Device(s) | 1
rmm-pool | True
tcp | True
ib | False
nvlink | True
==========================
Total time | 146.74 ms
Total time | 332.22 ms
Total time | 146.05 ms
==========================
Worker-Worker Transfer Rates
(w1,w2) | 25% 50% 75% (total nbytes)
--------------------------
Merge benchmark
--------------------------
Chunk-size | 100000000
Frac-match | 0.3
Ignore-size | 1.05 MB
Protocol | ucx
Device(s) | 1,2
rmm-pool | True
tcp | True
ib | False
nvlink | True
==========================
Total time | 1.10 s
Total time | 1.14 s
Total time | 1.15 s
==========================
Worker-Worker Transfer Rates
(w1,w2) | 25% 50% 75% (total nbytes)
--------------------------
(01,02) | 27.54 GB/s 29.48 GB/s 32.95 GB/s (19.20 GB)
(02,01) | 26.48 GB/s 29.35 GB/s 31.84 GB/s (19.20 GB)
Merge benchmark
--------------------------
Chunk-size | 100000000
Frac-match | 0.3
Ignore-size | 1.05 MB
Protocol | ucx
Device(s) | 1,2,3,4
rmm-pool | True
tcp | True
ib | False
nvlink | True
==========================
Total time | 1.62 s
Total time | 1.68 s
Total time | 1.80 s
==========================
Worker-Worker Transfer Rates
(w1,w2) | 25% 50% 75% (total nbytes)
--------------------------
(01,02) | 3.44 GB/s 4.35 GB/s 10.15 GB/s (13.40 GB)
(01,03) | 2.88 GB/s 3.38 GB/s 4.16 GB/s (7.90 GB)
(01,04) | 2.99 GB/s 3.17 GB/s 3.90 GB/s (7.20 GB)
(02,01) | 3.43 GB/s 3.83 GB/s 11.11 GB/s (14.40 GB)
(02,03) | 2.65 GB/s 3.53 GB/s 3.70 GB/s (6.50 GB)
(02,04) | 2.92 GB/s 3.33 GB/s 3.97 GB/s (7.20 GB)
(03,01) | 2.69 GB/s 3.14 GB/s 3.94 GB/s (7.20 GB)
(03,02) | 2.68 GB/s 3.87 GB/s 4.56 GB/s (7.20 GB)
(03,04) | 3.59 GB/s 4.10 GB/s 6.86 GB/s (12.00 GB)
(04,01) | 2.67 GB/s 3.61 GB/s 3.99 GB/s (7.20 GB)
(04,02) | 2.84 GB/s 4.01 GB/s 4.42 GB/s (7.20 GB)
(04,03) | 2.98 GB/s 4.51 GB/s 6.51 GB/s (12.00 GB)
Merge benchmark
--------------------------
Chunk-size | 100000000
Frac-match | 0.3
Ignore-size | 1.05 MB
Protocol | ucx
Device(s) | 0,1,2,3,4,5,6,7
rmm-pool | True
tcp | True
ib | False
nvlink | True
==========================
Total time | 4.65 s
Total time | 4.19 s
Total time | 5.01 s
==========================
Worker-Worker Transfer Rates
(w1,w2) | 25% 50% 75% (total nbytes)
--------------------------
(00,01) | 326.81 MB/s 374.27 MB/s 429.48 MB/s (4.30 GB)
(00,02) | 235.57 MB/s 325.63 MB/s 473.19 MB/s (3.60 GB)
(00,03) | 347.40 MB/s 461.02 MB/s 525.82 MB/s (3.95 GB)
(00,04) | 339.39 MB/s 443.83 MB/s 468.86 MB/s (3.95 GB)
(00,05) | 248.57 MB/s 333.42 MB/s 687.69 MB/s (6.00 GB)
(00,06) | 326.14 MB/s 358.11 MB/s 417.15 MB/s (3.25 GB)
(00,07) | 304.63 MB/s 329.35 MB/s 679.78 MB/s (6.35 GB)
(01,00) | 349.01 MB/s 426.61 MB/s 596.48 MB/s (3.95 GB)
(01,02) | 308.42 MB/s 324.30 MB/s 384.43 MB/s (3.95 GB)
(01,03) | 324.16 MB/s 355.91 MB/s 487.02 MB/s (3.60 GB)
(01,04) | 318.64 MB/s 405.89 MB/s 449.62 MB/s (6.35 GB)
(01,05) | 326.09 MB/s 351.79 MB/s 395.48 MB/s (3.95 GB)
(01,06) | 308.51 MB/s 370.74 MB/s 475.39 MB/s (3.95 GB)
(01,07) | 338.19 MB/s 395.52 MB/s 709.43 MB/s (8.75 GB)
(02,00) | 319.84 MB/s 401.11 MB/s 745.76 MB/s (3.25 GB)
(02,01) | 356.00 MB/s 579.58 MB/s 756.69 MB/s (6.00 GB)
(02,03) | 336.25 MB/s 418.43 MB/s 675.07 MB/s (6.35 GB)
(02,04) | 333.87 MB/s 513.51 MB/s 545.85 MB/s (3.60 GB)
(02,05) | 348.70 MB/s 403.66 MB/s 429.10 MB/s (3.60 GB)
(02,06) | 362.60 MB/s 499.80 MB/s 671.90 MB/s (6.35 GB)
(02,07) | 338.56 MB/s 531.88 MB/s 594.73 MB/s (4.30 GB)
(03,00) | 340.59 MB/s 384.41 MB/s 505.94 MB/s (6.00 GB)
(03,01) | 307.13 MB/s 354.70 MB/s 412.25 MB/s (3.60 GB)
(03,02) | 330.81 MB/s 356.54 MB/s 458.79 MB/s (3.60 GB)
(03,04) | 357.25 MB/s 412.34 MB/s 536.13 MB/s (6.00 GB)
(03,05) | 315.81 MB/s 424.67 MB/s 518.08 MB/s (3.60 GB)
(03,06) | 303.86 MB/s 313.99 MB/s 578.82 MB/s (3.60 GB)
(03,07) | 321.96 MB/s 389.38 MB/s 962.88 MB/s (8.40 GB)
(04,00) | 329.92 MB/s 463.17 MB/s 845.60 MB/s (6.00 GB)
(04,01) | 302.06 MB/s 423.82 MB/s 752.58 MB/s (6.00 GB)
(04,02) | 330.94 MB/s 421.92 MB/s 502.44 MB/s (3.60 GB)
(04,03) | 329.69 MB/s 426.35 MB/s 888.25 MB/s (6.00 GB)
(04,05) | 309.62 MB/s 378.41 MB/s 449.82 MB/s (3.60 GB)
(04,06) | 293.40 MB/s 330.06 MB/s 485.16 MB/s (3.60 GB)
(04,07) | 284.08 MB/s 324.06 MB/s 479.00 MB/s (3.60 GB)
(05,00) | 318.12 MB/s 471.36 MB/s 649.84 MB/s (3.95 GB)
(05,01) | 295.72 MB/s 455.05 MB/s 520.24 MB/s (6.00 GB)
(05,02) | 296.77 MB/s 484.85 MB/s 671.34 MB/s (4.30 GB)
(05,03) | 336.94 MB/s 445.78 MB/s 556.01 MB/s (3.95 GB)
(05,04) | 302.22 MB/s 433.72 MB/s 610.28 MB/s (3.95 GB)
(05,06) | 330.26 MB/s 511.31 MB/s 659.31 MB/s (6.35 GB)
(05,07) | 295.30 MB/s 435.12 MB/s 604.09 MB/s (3.95 GB)
(06,00) | 291.83 MB/s 483.34 MB/s 674.02 MB/s (6.00 GB)
(06,01) | 329.12 MB/s 416.47 MB/s 555.08 MB/s (3.60 GB)
(06,02) | 302.44 MB/s 344.61 MB/s 832.94 MB/s (6.00 GB)
(06,03) | 286.65 MB/s 335.21 MB/s 536.38 MB/s (3.60 GB)
(06,04) | 283.83 MB/s 306.26 MB/s 527.61 MB/s (6.00 GB)
(06,05) | 306.91 MB/s 539.75 MB/s 783.36 MB/s (6.00 GB)
(06,07) | 314.08 MB/s 329.18 MB/s 427.33 MB/s (3.60 GB)
(07,00) | 361.53 MB/s 541.03 MB/s 706.52 MB/s (6.70 GB)
(07,01) | 344.03 MB/s 427.87 MB/s 565.89 MB/s (4.30 GB)
(07,02) | 357.50 MB/s 381.41 MB/s 545.77 MB/s (4.30 GB)
(07,03) | 350.74 MB/s 460.18 MB/s 632.73 MB/s (3.95 GB)
(07,04) | 323.39 MB/s 403.57 MB/s 492.07 MB/s (4.30 GB)
(07,05) | 362.77 MB/s 436.20 MB/s 593.78 MB/s (6.70 GB)
(07,06) | 339.48 MB/s 410.97 MB/s 571.82 MB/s (4.30 GB)