In a bid to take the lead in the ongoing AI hardware race, Intel has thrown its gauntlet against Nvidia. During Computex, CEO Pat Gelsinger unveiled pricing for Intel’s next-gen Gaudi 2 and Gaudi 3 AI accelerator chips, and the numbers look exciting.
Pricing for products like these is typically kept hidden from the public, but Intel has now offered some official figures. The flagship Gaudi 3 accelerator will cost around $15,000 per unit when bought individually, which is 50 percent cheaper than Nvidia’s H100 data center GPU.
The less powerful Gaudi 2 also challenges Nvidia’s pricing drastically. A complete 8-chip Gaudi 2 accelerator kit will be priced at $65,000 to system vendors. Intel claims that just one-third the price of similar setups from Nvidia and other companies. For the Gaudi 3, that same 8-accelerator kit configuration costs $125,000. Intel claims it’s two-thirds cheaper than alternative solutions at that high-end performance spectrum.
Also Read: Asus Introduces Thor-Inspired Power Station Mjolnir
Performance And Software Make Intel Gaudi 3 Worthy
Speaking of the Gaudi 3 pricing, Nvidia’s newly launched Blackwell B100 GPU costs around $30,000 per unit. On the other hand, the Blackwell CPU+GPU combo, the B200, comes at a price of $70,000.
But pricing is only one part of the scenario. Performance and the software ecosystem are equally important. Intel suggests the Gaudi 3 keeps pace with or even outshines Nvidia’s H100 across a variety of important AI training and interference workloads.
Benchmarks shown by Intel show the Gaudi 3 delivering up to 40 percent faster training times compared to the H100 in large 8,192-chip clusters. Even a smaller 64-chip Gaudi 3 setup offers 15 percent higher throughput compared to H100 on the popular LLaMA 2 language model, as per the company. For AI interference, the company claims a 2x speed advantage over the H100 on models such as LLaMA and Mistral.
Also Read: Computex 2024: Asus Announces ROG Zephyrus G16 Gaming Laptop
Nvidia Still A Big Competitor To Beat
While the Gaudi chips use open standards like Ethernet or simple deployment, they lack optimizations for Nvidia’s CUDA platform which most AI software depends on as of now. Convincing companies to refactor their code for Gaudi could be challenging.
To boost adoption, the company says it has around 10 major server vendors -- including new Gaudi 3 partners such as Asus, Foxconn, Gigabyte, Inventec, Quanta, and Wistron. Known names such as Dell, HPE, Lenovo, and Supermicro are also on board.
Still, Nvidia is the biggest name when it comes to the data center world. In the final quarter of 2023, the company claimed a 73 percent share of the data center processor market, and that number has continued to grow, taking away the stakes of both Intel and AMD. The consumer GPU market also isn’t a different story, with Nvidia grabbing an 88 percent share.