DGX 4480 - Nightly Benchmark run 4480-20210204

[quasiben]

Benchmark history

Raw Data

<Client: 'tcp://127.0.0.1:34013' processes=10 threads=10, memory=540.94 GB>
Distributed Version: 2021.01.1+14.ga0d60e7b
simple
5.475e-01 +/- 4.27e-02
shuffle
2.256e+01 +/- 1.009e+00
rand_access
1.041e-02 +/- 1.329e-03
anom_mean
9.949e+01 +/- 8.054e-01

Raw Values

simple
[0.58701086 0.52487779 0.50652409 0.51683784 0.58730984 0.53589463
0.62814784 0.47800493 0.56864429 0.54135227]
shuffle
[23.61287832 23.79913616 21.81883407 21.01972342 21.4097302 22.95353341
21.96969128 23.51331043 21.81375837 23.70815349]
rand_access
[0.00773001 0.00953388 0.01086974 0.01190329 0.01062369 0.01146913
0.00869823 0.01183629 0.01141453 0.01006341]
anom_mean
[ 98.67475772 99.0814085 99.3625977 98.47209287 99.49914122
98.5962007 100.6324234 99.86433983 101.00307035 99.75229931]

Dask Profiles

• Shuffle
  Profile
• Random Access
  Profile
• Simple
  Profile
• Anom
  Mean

Scheduler Execution Graph

[jakirkham]

As PR ( dask/distributed#4480 ) appears to be based on commit ( dask/distributed@08ea968 ) as opposed to master latest, this should be compared to the results in issue ( #98 )

[jakirkham]

More recent run using the same PR based off master instead in issue ( #103 )