Intel on Tuesday introduced one of many world’s first giant deployments of its Gaudi2 accelerators for synthetic intelligence (AI) workloads. The supercomputer — inbuilt collaboration with Dell — shall be utilized by Stability AI for its generative AI purposes.
Intel’s new supercomputer is predicated on the corporate’s newest Intel Xeon Scalable processors and runs as many as 4,000 Intel Habana Gaudi2 AI {hardware} accelerators. Assuming that every machine employs eight Habana Gaudi2 processors in OAM form-factor (which is typical for OAM machines), the machine options as many as 500 nodes. Provided that Intel’s Gaudi2 affords formidable efficiency that is similar to that of Nvidia’s H100 in choose purposes, the supercomputer must be fairly highly effective. We’re most likely taking a look at 7–8 FP16 ExaFLOPS of AI efficiency.
Intel and Dell didn’t disclose whether or not they’ll proceed to crew up for AI supercomputers going ahead. In the meantime, for the chip designer, it is essential that one of many world’s largest server suppliers has PowerEdge machines with its Xeon Scalable CPUs and Habana Gaudi2 accelerators developed particularly for AI workloads. These machines are designed to deal with each large-scale coaching and base-level inferencing.
In addition to Dell, Alibaba Cloud mentioned that the 4th Technology Xeon can obtain a 3x inference pace in its model-serving platform DashScope program when used with Superior Matrix Extensions (Intel AMX) accelerators and varied software program optimizations.
Intel’s processors energy quite a few supercomputers and even Nvidia’s DGX H100 methods for AI and high-performance computing (HPC) purposes, however the firm’s personal AI accelerators and compute GPUs should not as fortunate and there are only a few Gaudi/Gaudi2 or Ponte Vecchio-based AI deployments up to now. With Stability AI working with Gaudi2, it is going to be attention-grabbing to see what different firms additionally select to present it a shot.