Intel discloses list prices of its Gaudi 3 and Gaudi 2 AI accelerators and we're in for a shock — rivals to iconic Nvidia's H100 GPU have a much better performance per dollar ratio but will it matter?

Intel introduced the Intel Xeon 6 processors featuring Efficient-cores (E-cores) and revealed the list pricing for its Gaudi 3 and Gaudi 2 AI accelerators at Computex 2024, giving eager fans more details on the new releases.

Intel positions its Gaudi architecture as a viable, and more affordable, alternative to Nvidia‘s H100. Pricing announced at Computex revealed a standard AI kit comprising eight Intel Gaudi 2 accelerators with a universal baseboard (UBB) is priced at $65,000.

Not so long ago we reported how Supermicro was selling an Intel server with eight Gaudi 2 AI accelerators, 76 cores, 1TB of RAM and 100GbE for $90,000). Intel also revealed that a kit featuring eight Intel Gaudi 3 accelerators with a UBB lists at $125,000.

Greater performance

“AI is driving one of the most consequential eras of innovation the industry has ever seen,” said Intel CEO Pat Gelsinger.

“Intel is one of the only companies in the world innovating across the full spectrum of the AI market opportunity – from semiconductor manufacturing to PC, network, edge and data center systems. Our latest Xeon, Gaudi and Core Ultra platforms, combined with the power of our hardware and software ecosystem, are delivering the flexible, secure, sustainable and cost-effective solutions our customers need to maximize the immense opportunities ahead.”

As the market leader, Nvidia is able to charge a premium for its H100 GPU-based hardware. While the Gaudi offerings may be more attractive in terms of affordability, Intel has a battle on its hands to win over buyers who see Nvidia’s products as superior and are willing to pay more to get the best.

For that reason, Intel isn’t only undercutting its chief rival on price, it wants to sell its products based on performance. The veteran tech giant says its Gaudi 3 accelerators deliver performance improvements for both training and inference tasks on leading GenAI models. An 8,192-accelerator cluster is claimed to offer up to a 40% faster time-to-train compared to an equivalent size Nvidia H100 GPU cluster. Intel says the Gaudi 3 also delivers up to a 15% faster training throughput for a 64-accelerator cluster versus Nvidia H100 on specific models, along with up to a 2x faster inferencing speed on popular LLMs.

Dell, Hewlett Packard Enterprise, Lenovo, and Supermicro are already onboard to help bring Intel Gaudi 3 to market, but Intel plans to add six more partners to make its AI systems broadly available, and these are Asus, Foxconn, Gigabyte, Inventec, Quanta, and Wistron.