Taipei – One of the biggest challenges facing companies and researchers working with artificial intelligence technologies has been the high costs associated with procuring specialized AI hardware. However, American semiconductor giant Intel believes it has found a solution with its newly unveiled third-generation AI accelerator Gaudi 3.
Speaking at the recently concluded Intel Tech Tour in Taiwan, Anil Nanduri, Vice President and General Manager of Intel's Data Center and AI business revealed that the new Gaudi 3 promises significantly lower costs when compared to the massively popular NVIDIA H100 AI accelerator. “We believe Gaudi 3 can run any AI model sizes that the NVIDIA chip can handle, but deliver much better value for the dollars spent,” noted Nanduri.
Intel claims Gaudi 3 offers 1.9 times more AI training performance per dollar and 2.3 times more inferencing performance compared to the H100 accelerator. Most importantly, the complete Gaudi 3 kit including 8 accelerators is priced at $125,000 compared to over $430,000 for a similar NVIDIA setup, representing a huge savings.
Another key differentiator highlighted is that Gaudi 3 is based on open frameworks like PyTorch rather than a proprietary network. It also supports the open RoCE network protocol making it an open platform well-suited for academic and research organizations. An Indian deeptech startup leveraging Gaudi 2 chips for its AI servers has already indicated plans to upgrade to the new third-gen accelerators.
Overall, Intel is positioning the Gaudi 3 as a cost-efficient alternative to NVIDIA's dominance in the AI hardware market. With four-times lower prices and comparable performance, the new chip promises to make advanced AI more accessible to enterprises, governments and startups worldwide. Only time will tell if it truly succeeds in giving NVIDIA serious competition.
