Cerebras Systems, backed by UAE-based G42 and in partnership with MBZUAI and India's C-DAC, is deploying an 8 exaFLOPS AI supercomputer in India. The system will use approximately 64 WSE-3 wafer-scale accelerators, each delivering 125 petaFLOPS of sparse 16-bit compute. Unlike Nvidia and AMD GPUs, Cerebras chips use on-chip

4m read time From go.theregister.com
Post cover image

Sort: