From a036a1be471704b1bf5ba32a67ff5ca61ec948d0 Mon Sep 17 00:00:00 2001 From: Esther Quansah Date: Fri, 24 Jun 2022 21:54:32 +0300 Subject: [PATCH 1/2] update about default cpus --- changelog/2022-06-24.md | 14 ++++++++++++++ 1 file changed, 14 insertions(+) create mode 100644 changelog/2022-06-24.md diff --git a/changelog/2022-06-24.md b/changelog/2022-06-24.md new file mode 100644 index 00000000..fa5b51b4 --- /dev/null +++ b/changelog/2022-06-24.md @@ -0,0 +1,14 @@ +## :warning: June 24, 2022 + +**CLI version: 0.8.65 ** + +This release includes an important update to how CPU and memory are allocated to experiments. + +Prior to this release, Grid would set the default number of CPUs to 1 when creating runs and not explictly specifying `--cpus`. + +We recently discovered an issue with runs where setting `--cpus` 1 would also reduce the memory, causing lots of OOM issues. + +So we've updated this behavior to set `--cpus` to 0 by default. This applies when creating runs with GPUs as well. By setting `--cpus` to 0, the backend will allocate all available CPU and memory to the experiment. + + +--- From 242f2d950f34d02723cd97e0568305d5c48b6add Mon Sep 17 00:00:00 2001 From: Esther Quansah Date: Fri, 24 Jun 2022 21:58:39 +0300 Subject: [PATCH 2/2] Update changelog/2022-06-24.md --- changelog/2022-06-24.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/changelog/2022-06-24.md b/changelog/2022-06-24.md index fa5b51b4..5ecd44a2 100644 --- a/changelog/2022-06-24.md +++ b/changelog/2022-06-24.md @@ -6,7 +6,7 @@ This release includes an important update to how CPU and memory are allocated to Prior to this release, Grid would set the default number of CPUs to 1 when creating runs and not explictly specifying `--cpus`. -We recently discovered an issue with runs where setting `--cpus` 1 would also reduce the memory, causing lots of OOM issues. +We recently discovered an issue with runs where setting `--cpus` to 1 would also reduce the memory, causing lots of OOM issues. So we've updated this behavior to set `--cpus` to 0 by default. This applies when creating runs with GPUs as well. By setting `--cpus` to 0, the backend will allocate all available CPU and memory to the experiment.