[Aida-compute-users] AIDA DGX-2 Service again online, now in Phase 3!

[Joel Hedlund]

Hi!

We are proud to announce that the AIDA DGX-2 Service is again online after this week's service stop, now in Phase 3: secure enough for use with sensitive personal data! Terms of use have been updated to reflect this (PIs: please reply I agree to that other email if you still haven't!) but from a user perspective we expect all to be just the same. Put yourself into the booking sheet <https://docs.google.com/spreadsheets/d/1wA7H3Uh53ADVYptiQWXROnMD67HvPOAwSvW20EnzlFM/edit#gid=2127795854&range=D57> and try it out!

In this service stop we have done firmware and OS upgrades, and moved the systems to a high-reliability hospital server room, and done major networking installations to make this relocation possible. This was a major cross-organizational undertaking that in practice went surprisingly smooth. We reserved a week for this service stop, but managed to complete the necessary dgx-2 work in shorter than expected time.

As you may have seen in chat <https://zulip.aida.medtech4health.se/#narrow/stream/13-dgx2/subject/service.20stop.202020-11-09/near/7033>, work is still ongoing in (supposedly!) unrelated parts of the network, which shouldn't affect us (famous last words!). Also, depending on Nvidia feedback we may need to plan for another (short) service stop for additional firmware or kvm image upgrades.

However we expect to now have preempted the need for the recurring service stops necessitated by semiregular tests of hospital emergency power. (And as you may have seen, that one on Dec 1 has been removed from the booking sheet :-)

Some notes: Spawned VMs will still use latest version Nvidia OS KVM image (v4.3). However, as Nvidia OS v4.3 is deprecated, it is recommended to use the following sequence of commands to upgrade your VMs to latest v4.x minor release (eg v4.6):

sudo apt update
sudo apt install -y dgx-bionic-r450+cuda11.0-repo
sudo apt update
sudo apt full-upgrade

(Upgrading to 5.0 may not be advisable, as that does not yet support KVM virtualization, but ymmv)

Please let us know in the AIDA DGX-2 Chat <https://zulip.aida.medtech4health.se/#narrow/stream/13-dgx2> or contact aida-compute <mailto:aida-compute at medtech4health.se> if you have any questions or input!

Cheers!
/Joel Hedlund
AIDA DGX-2 Service owner
AIDA Data director

-------------- next part --------------
En HTML-bilaga skiljdes ut...
URL: <http://lists.liu.se/pipermail/aida-compute-users/attachments/20201110/2d311a20/attachment.html>