HPE has given CIOs contemplating empty car parks while workers remain at home an alternative – drop in an AI datacenter in a box.
The containerized AI Mod POD was just part of an NVIDIA enriched product lineup the vendor dropped to coincide with the kick off of the GPU maker’s GTC conference.
The PODs will be targeted at AI or HPC, feature Direct Liquid Cooling, and support up to 1.5MW per module. The vendor claimed they will be 30% cheaper compared to brick and mortar devices.
More to the point, they will be much faster to deploy than either a greenfield AI datacentre, or a retrofitted site.
Trish Damkroger, SVP GM HPC Computing and AI at HPE, speaking ahead of GTC, said they can be up and running in months. As Jensen Huang said in his speech, AI datacenter deployments take two to three years. If you’re lucky.
“It's easy,” added Damkroger. “You just gotta level some space, and you can have a datacentre in your backyard up and running in months.” Or, she suggested, simply use the car parking space left over by employees continuing to work from home post Covid. Or presumably those who have lost their positions due to the encroachment of AI.
HPE also took the wraps up new AI servers, paralleling NVIDIA’s chip rollout. These include the NVIDIA GB300 NVL72 by HPE aimed at “service providers and cutting-edge enterprises” looking to deploy large AI clusters for trillion parameter models.
And it debuted a trio of ProLiant Compute servers, supporting NVIDIA platforms, including the HGX B300, GB200 Grace Blackwell NVL4 and NVIDIA RTX™ PRO 6000 Blackwell. All of which will come to market in the second half of the year.
HPE also got behind NVIDIA’s plans to revolutionize storage by embedding accelerators - ie, GPUs - and making data querying semantic, not a question of retrieval. NVIDIA said HPE would be infusing its AI Data Platform capabilities into HPE Private Cloud for AI, HPE Data Fabric, HPE Alletra Storage MP and HPE GreenLake for File Storage.
See also: NVIDIA datacentre chief to Europe: GPUs are coming, but have you got the power?
What HPE presumably won’t be selling is systems based on NVIDIA’s Grace Blackwell-powered DGX Station and DGX Spark designs.
Huang took the wraps of these in his GTC keynote, saying “This is what a PC should look like.”
But since the split that created HPE and client focused HP Inc, presumably it’s the latter that will be responsible for pushing Huang’s designs onto enterprise desktops.
