New MLPerf benchmarks put Intel in a close second place behind Nvidia, which is significantly better than expected.

Intel seems to be on the rise with its current AI accelerator, Habana Gaudi 2. After a difficult initial delivery in this market, Gaudi 2 is a smash hit according to the latest MLPerf benchmarks. That’s good news for Team Blue, which can grab a share of the lucrative AI accelerator market, including hallucinatory margins thanks to the current AI hype.

A looming chip capacity shortage at TSMC would also be particularly good news for Intel to get its foot in the door. AMD is also rumored to be focusing on AI accelerators, reportedly targeting the new high-end RDNA 4 GPUs in the pipeline. Every chip designer is turning to AI to grab a piece of the pie that Nvidia mostly eats today.

Better than Nvidia A100

The Intel Habana Gaudi 2 AI accelerator scores a top spot in the MLPerf benchmark over the Nvidia A100 chip, Nvidia’s previous generation GPU. The latest and most powerful Nvidia H100 remains the top chip, individually as a GPU and in combination with a powerful ARM CPU under the name Nvidia GH200.

The numbers in Intel’s presentation are impressive, but we are still missing important parameters such as consumption (TDP) and temperature information. On the other hand, today everyone is begging for hardware and an additional supplier of high-end chips is more important.

Gaudi 2 is the result of the acquisition of Habana Labs in 2019 for $2 billion. Until then, Intel was working feverishly on Nervana processors. Since the takeover, Gaudi has been the focus. The second version has three times more Tensor cores (24 pieces) and additional SRAM and HBM2E memory, i.e. two and three times more, respectively. Gaudi 2 has 24 100 Gigabit Ethernet ports, 14 more than its predecessor.