Qualcomm Launches AI250 & AI200 With Big Reminiscence Footprint For AI Knowledge Middle Workloads

What that boils all the way down to for the AI200 is help for as much as 768GB of LPDDR per card to allow distinctive scale and adaptability for AI inference. Qualcomm’s AI200 rack additionally integrates a hexagonal NPU and, as a complete, is is direct liquid cooled (DLC).

In the meantime, the AI250 affords the identical 768GB reminiscence capability per card whereas additionally introducing what Qualcomm says is an progressive reminiscence structure based mostly on near-memory computing for a generational leap in effectivity and efficiency for AI inference workloads. In accordance with Qualcomm, it delivers greater than 10x increased efficient reminiscence bandwidth, and at a a lot decrease energy consumption.

“With Qualcomm AI200 and AI250, we’re redefining what’s doable for rack-scale AI
inference. These progressive new AI infrastructure options empower clients to deploy
generative AI at unprecedented TCO, whereas sustaining the flexibleness and safety fashionable knowledge
facilities demand,” stated Durga Malladi, SVP & GM, Know-how Planning, Edge Options & Knowledge
Middle, Qualcomm Applied sciences, Inc.

“Our wealthy software program stack and open ecosystem help
make it simpler than ever for builders and enterprises to combine, handle, and scale already
educated AI fashions on our optimized AI inference options. With seamless compatibility for main AI frameworks and one-click mannequin deployment, Qualcomm AI200 and AI250 are
designed for frictionless adoption and fast innovation,” Malladi added.