HPC Cloud Updates WE 24 Nov 2024
Updates to AWS, Azure & GCP in the last week relevant for HPC practitioners
AWS
Need to comply with NIST SP 800-223 for your HPC workload? AWS have got a reference architecture for you
A few small things from AWS that contribute to an easier life in managing your AWS compute for HPC. Its funny the don’t link these releases together and sell it a bit more…
Faster, more accurate EC2 scaling? Yes please
Couple it with CPU performance based attribute selection
make better use of your capacity reservations too
and see the lineage of your AMIs
Individually those are small but good improvements. Coming together in the same week, during SC24 no less, and no one span them into a HPC flavoured press release? Maybe they’re all too busy writing AI content re:Invent
c7i-flex and m7i-flex now available in Malaysia
and r8g in Stockholm
and c6a and r6a in Hyderabad
Next gen FSx for Lustre now in California. That should give the AI bros something to be happy about.
Azure
I think Azure were the only cloud provide with a major announcement at SC24 this week. Introducing the HBv5
Also released this week and quite possibly just as interesting for a lot of HTC and single node HPC workloads, the Da/ Ea/ Fa v6 family
Can’t believe this hasn’t had more fanfare around it. I’ve been waiting for this for a while! Azure Spot Placement Score. If you need spot capacity this will help you figure out how much you might be able to get hold before you request it
Azure Compute Fleet will now support multi region. Nice.
and attribute based instance selection
If you’re still using Scale Sets, you can now emit custom metrics to determine the order in which machines are upgraded
and also Zonal expansion
Azure seem to be taking a different path to AWS and Google in the Redis/Valkey split with their new managed Redis offering
Google Cloud
Google had pre-empted SC24 with news last week of all the HPC related updates they had released this year but announced the arrive of NVIDIA GB200s to Google Cloud