Files
mscclpp/test/deploy/setup.sh
Changho Hwang bb7b85a810 2-node AllReduce improvements (#118)
* Added `get()` interfaces to `SmChannel`
* Improved 2-node (8 gpus/node) AllReduce: algbw 139GB/s for 1GB (kernel
3) and 99GB/s for 48MB (kernel 4)
* Fixed a FIFO perf bug
* Several fixes & validations in mscclpp-test

---------

Co-authored-by: Binyang Li <binyli@microsoft.com>
Co-authored-by: Saeed Maleki <saemal@microsoft.com>
2023-07-07 07:05:46 +00:00

18 lines
533 B
Bash

set -e
mkdir -p /root/.ssh
mv /root/mscclpp/sshkey.pub /root/.ssh/authorized_keys
chown root:root /root/.ssh/authorized_keys
mv /root/mscclpp/config /root/.ssh/config
chown root:root /root/.ssh/config
chmod 400 /root/mscclpp/sshkey
chown root:root /root/mscclpp/sshkey
nvidia-smi -pm 1
for i in $(seq 0 $(( $(nvidia-smi -L | wc -l) - 1 ))); do
nvidia-smi -ac $(nvidia-smi --query-gpu=clocks.max.memory,clocks.max.sm --format=csv,noheader,nounits -i $i | sed 's/\ //') -i $i
done
mkdir -p /var/run/sshd
/usr/sbin/sshd -p 22345