Cerebras unveils WSE-3 with 4 trillion transistors

Cerebras/WSE-3 / CS-3

Cerebras unveiled the WSE-3, a single chip the size of a whole silicon wafer with 4 trillion transistors, 900,000 AI cores and 44 GB on-chip memory, made on TSMC 5 nm. It powers the CS-3 system at 125 petaflops, twice its predecessor at the same power and price, with clusters up to 2,048 systems.

• WSE-3: 4 trillion transistors, 900,000 AI cores, 44 GB on-chip SRAM on TSMC 5 nm; 125 petaflops peak performance.
• CS-3 supercomputer doubles WSE-2 performance at same power and price; clusterable to 2,048 systems for 256 exaFLOPs aggregate.
• Full 2,048-system CS-3 cluster can train Llama 70B from scratch in a single day; supports up to 24 trillion-parameter models.