projects / platform / upstream / tensorflow.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: ffe6ede) | patch
[XLA:GPU] Disable multi-streaming by default.
authorJustin Lebar <jlebar@google.com>
Wed, 9 May 2018 00:29:01 +0000 (17:29 -0700)
committerTensorFlower Gardener <gardener@tensorflow.org>
Wed, 9 May 2018 17:35:53 +0000 (10:35 -0700)
commit15879526893886852b64d60b72c40bc6daeda22e
tree79ed2945bace4e38505fea0f243da6272425523etree | snapshot
parentffe6ede215729f99764761c5acf6a3bdebf69cedcommit | diff
[XLA:GPU] Disable multi-streaming by default.

Run all GPU work on one stream by default.  We've found experimentally
that multi-streaming creates significant additional memory pressure on
some models, and we don't have any good benchmarks where multi-streaming
helps on which to tune the stream-assignment heuristics.  So just
disable it for now.

PiperOrigin-RevId: 195903229
tensorflow/compiler/xla/legacy_flags/debug_options_flags.cc diff | blob | history
tensorflow/compiler/xla/service/gpu/hlo_schedule_test.cc diff | blob | history
tensorflow/compiler/xla/service/gpu/stream_assignment_test.cc diff | blob | history
Domain: Machine Learning / ML Framework;