From 310c07206d6b14dcfda4935e6fef79264c09d00e Mon Sep 17 00:00:00 2001 From: Susan Lindsey Date: Tue, 18 Feb 2025 14:45:58 -0600 Subject: [PATCH] Revert "Vista - integrate Ian's changes" This reverts commit 560f948edcb971fbaac3e78e140526ddd2ce15ba. --- docs/hpc/vista.md | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/docs/hpc/vista.md b/docs/hpc/vista.md index 86d322e..f900af8 100644 --- a/docs/hpc/vista.md +++ b/docs/hpc/vista.md @@ -375,7 +375,7 @@ Follow these steps to configure MPS on Vista for optimized multi-process workflo 2. **Launch MPS Control Daemon** - Use `ibrun` to start the MPS daemon across all allocated nodes. This ensures one MPS control process per node, targeting GPU 0: + Use `ibrun` to start the MPS daemon across all allocated nodes. This ensures one MPS control process per node: ```job-script # Launch MPS daemon on all nodes @@ -433,7 +433,8 @@ MPS is particularly effective for workloads characterized by: You may verify performance gains for your use case using the following command to monitor the node that your job is running on (e.g., `c608-052`): ```cmd-line -login1$ nvidia-smi dmon --gpm-metrics=3,12 -s u +login1$ ssh c608-052 +c608-052$ nvidia-smi dmon --gpm-metrics=3,12 -s u ``` The side-by-side plots in Figure 1 illustrate the performance enhancement obtained by running two GPU processes simultaneous on a single Hopper node with MPS. The GPU performance improvement is ~12%, compared to no improvement without MPS. Also, the setup cost on the CPU (about 12 seconds) is completely overlapped, resulting in in a 1.2x total improvement for 2 simultaneous Amber executions. Even better performance is expected for applications which don't load the GPU as much as Amber.