Cerebras unveils CS-4 system built from three WSE-3 Turbo wafers, claiming up to 30x GPU speed

Cerebras/WSE-3 / CS-3

Cerebras announced the CS-4, a rack-scale system built from three of its newly released WSE-3 Turbo wafer-scale chips. The company says it runs AI inference up to 30 times faster than GPU solutions, a comparison it ran itself, and that first CS-4 shipments begin this quarter.

• Each CS-4 system is built from three WSE-3 Turbo wafer-scale chips.
• The 30x figure compares tokens per second per user on GPT-OSS-120B with identical prompts; the release gives 4,400+ tokens per second per user for the CS-4.
• The release says the CS-4 is up to twice as fast as the CS-3 and up to 10x better per watt; no independent benchmark is cited.
• First shipments are said to begin this quarter; the release gives no unit count, pricing or customers.