Improve the Performance, Efficiency, and ROI of GPU Computing with Hammerspace Tier 0

Servers equipped with GPUs are expensive but vital for modern computing, particularly projects involving AI. Due to huge demand, however, GPU systems can be difficult to obtain, even when budget is available. And then there is the challenge of keeping them fed. GPUs require significant power, which is in short supply in most data centers. They also require lots of data, and to maximize ROI they need it to be supplied as fast as possible to minimize idle time. This often means organizations must purchase new high-performance storage and networking, which consumes additional funding and power, two resources that are chronically in short supply.

With all these constraints, organizations are rightly focused on making the most of the GPUs they have, which means keeping them busy doing productive work as much as possible. Yet there is one resource in every GPU server that is frequently overlooked, unused, or at best, underutilized. That resource is storage, specifically NVMe SSDs housed locally within the GPU server itself.

Beginning with version 5.1, Hammerspace unlocks this GPU-local NVMe storage to create a new Tier 0 of ultra- high performance, persistent, shared storage that can be used to accelerate compute jobs and reduce time spent on overhead tasks such as checkpointing. This efficiency has a snowball effect, not only increasing performance but lowering costs (both acquisition and operational) and accelerating time to value. Spend less, yet get more work done? That’s right. Continue reading to learn how.