Meta focuses on AI with homemade processors

Meta announced “an ambitious plan” to create “the next generation of ours AI infrastructure“, investing in datacenter advanced and developing home made processorsas well as a supercomputer together with NVIDIA.

Meta: processors, datacenters and supercomputers for AI

The first chip announced by Meta for artificial intelligence is called MTIA (Meta Training and Inference Accelerator). This is an in-house designed accelerator for inference. “MTIA offers more computing power and efficiency than CPUs and is tailored to our internal workloads. By installing both MTIA chips and GPUs, we will provide better performance, lower latency and higher efficiency for each workload.

Meta has also created another chip called MSVP (Meta Scalable Video Processor), designed to facilitate the transcoding of videos. MSVP was designed to handle the large amount of video content on Facebook and other platforms (over 4 billion videos are viewed on FB every day), but in the future MSVP will also serve to bring AI, VR and AR content into apps of Meta.

In addition to the chips, Meta also presented a new one AI-optimized datacenter design. And he announced the second phase of his 16,000 GPU supercomputer for AI research. These projects will allow Meta to develop larger and more sophisticated AI models. And to deploy them efficiently at scale.

Il Research SuperCluster (RSC) AI Supercomputer aims to offer almost 5 exaflops of power via 2000 NVIDIA DGX A100 systems for a total of 16,000 NVIDIA A100 Tensor Core accelerators connected via NVIDIA Quantum InfiniBand to 16 Tb/s. Monstrous computing power – all dedicated to artificial intelligence.

If the metaverse remains a dream, for Meta, AI is becoming a very definite future.