HPC Cloud Updates WE 28 Jul 2024

Updates to AWS, Azure & GCP in the last week relevant for HPC practitioners

HPC Cloud Updates WE 28 Jul 2024

AWS

Want to play with Llama 3.1 (including 405B) on AWS? It’s is now on Bedrock.

Meta Llama 3.1 405B now generally available in Amazon Bedrock - AWS
Discover more about what’s new at AWS with Meta Llama 3.1 405B now generally available in Amazon Bedrock
Meta Llama 3.1 generative AI models now available in Amazon Bedrock - AWS
Discover more about what’s new at AWS with Meta Llama 3.1 generative AI models now available in Amazon Bedrock

Looking to deploy your HPC cache in Jakarta? d3en instances now available

Amazon EC2 D3en instances are now available in Asia Pacific (Jakarta) region - AWS
Discover more about what’s new at AWS with Amazon EC2 D3en instances are now available in Asia Pacific (Jakarta) region

Azure

Really wish they’d sort out Azure Updates. Pretty terrible it’s been semi functional for this long. 

Public preview of sixth gen Intel Emerald Rapids based VMs is now available! Esv6 and Dsv6 variants are both available in US East and US West.

Public Preview: 6th generation Intel-based VMs – Dv6/Ev6
Public Preview: 6th generation Intel-based VMs – Dv6/Ev6

Need to upgrade your disk to something faster and still doing clickops? This might help then:

Public Preview: Convert to Azure Premium SSD v2 disks
Public Preview: Convert to Azure Premium SSD v2 disks

GCP

Want to let your Llamas roam free somewhere other than in AWS’ clouds? Google has you covered!

Llama 3.1 on Vertex AI | Google Cloud Blog
Meta’s Llama 3.1 family of models, including a new 405B model, is now available on Vertex AI. Try Llama 3.1 on Google Cloud today.

If high performance file systems are your thing (and lets face it, too many HPC applications still rely on a basic filesystem) then this might make for interesting reading:

UC Berkeley migrates to Filestore for JupyterHub | Google Cloud Blog
By migrating to Filestore, UC Berkeley’s data science team has enjoyed a new level of stability and performance for its for its JupyterHub deployment.