Pure Storage has announced a full-stack AI-validated design: GenAI Pod. Furthermore, its FlashBlade//S500 server is now certified for Nvidia’s DGX SuperPOD.
The company cites that companies struggle with the complexity of AI deployment. Storage is crucial here, as proprietary data must be linked to AI models and the required hardware. One challenging component is bringing all the data together, but Pure hopes to point to where the data should end up with its new solution. A Pure Storage GenAI Pod is a validated design that runs vector databases and foundation models. Pure has already partnered with Cisco, Meta, and Red Hat, among others, to deliver this full AI stack.
Portworx and Nvidia
After the initial deployment, Pure aims to achieve trouble-free daily use of the GenAI Pod. That leans primarily on the Kubernetes storage platform from Portworx, acquired by Pure in 2020, and Nvidia. From the latter comes NIM, a set of microservices for generative AI. Finally, the open-source vector database Milvus serves as the basis for fine-tuning foundation models.
NIM microservices leverage Nvidia’s AI Enterprise platform, which Nvidia describes as the “operating system” of generative AI. It brings together cluster management, foundation models, and app building for the fastest possible deployment of AI.
Those looking to get started with AI the fastest at the enterprise level can turn to Pure’s FlashBlade//S500, which is currently certified for Nvidia’s SuperPOD. This certification was already expected, but now the performance of this server is guaranteed.
These servers are equipped with up to 3,000 TB of DFM storage. They are explicitly intended for “extreme performance,” including GenAI training. So now, that designation takes on some extra meaning as Nvidia’s highest-end hardware works with Pure’s storage offerings. The secret behind these SuperPODs is their high scalability, as they are linkable actually to turn them into an AI supercomputer.
Also read: AI hyperscaler CoreWeave partners with Pure Storage