OpenAI’s Jalapeno chip beats Nvidia in efficiency tests: Here's how

In a significant move, OpenAI announced that its Jalapeno inference chip developed with Broadcom surpassed Nvidia’s GB300 on standard AI workload power and energy consumption.

OpenAI's published number of chips built was meant to reduce alliance with Nvidia, but the company confirmed that Jalapeno developed with Broadcom surpasses Nvidia’s GB300.

Jalapeno was not tested against Vera Rubin, but Nvidia’s next-generation deployment has just started shipping, and it cannot train models at all. Jalapeno reached the results at 700 watt and electricity is the energy consumption in any data center.

According to Chip chief Richard, high cost-efficient scalability maximizes output while minimizing expenses. He further said Jalapeno offers the “best of both worlds with lower latency as AI typically have to make a trade-off between the two.”

OpenAI plans to deploy Jalapeno in small volumes by the end of this year. Nonetheless, OpenAI does not expect to revamp its semiconductor lineup with Jalapeno adding that its entire overall compute strategy still relies on strong partners like Nvidia; however, the company will continue developing the second and third generations of the new chip.