Intel announced its new Gaudi 3 AI accelerator earlier this year, rocking up to 128GB of HBM2e memory with up to 3.7TB/sec of memory bandwidth... but how does it stack up against NVIDIA's dominant Hopper H100 80GB AI GPU?
Well, that comes down to cost -- Intel's new Gaudi 3 AI accelerator costs just $16,000 per AI accelerator with 128GB of HBM2e memory, compared to $30,000+ for NVIDIA's Hopper H100 AI GPU with 80GB of faster HBM3 memory. Intel is charging around $65,000 per 8 x Gaudi 2 accelerators on a baseboard, and $128,000 per 8 x Gaudi 3 AI accelerators on a baseboard.
This means that a single Gaudi 2 AI accelerator costs around $8125, while Gaudi 3 costs $15,650, when purchased in bulk, with the baseboards, of course.
This comes directly from Tirias Research analyst Jim McGregor, who posted the numbers on X earlier this month. Remember: this is buying the Gaudi 3 AI accelerators in bulk, with the baseboard. Intel needs to get as much of that luscious AI chip market, as NVIDIA is absolutely dominating with its H100 80GB AI GPU, even at a much higher $30,000 per chip.
NVIDIA CEO Jensen Huang said that its next-gen Blackwell B200 AI GPUs will cost between $30,000 and $40,000 each, and will absolutely blow Intel's latest Gaudi 3, and its own H100 AI GPUs out of the water. Once there's enough Blackwell B200 AI GPUs on the market, we should expect Hopper H100 AI GPUs to drop in price below their average of around $30,000 a pop.
Intel discounting its Gaudi 3 AI accelerators to gain market share isn't surprising, as it is the complete underdog in the AI market, no matter how Intel puts it, that's the truth. Intel isn't just fighting NVIDIA either, that's in the pure hardware battle, but there's custom, in-house designed AI processors used in cloud service providers like Amazon Web Services (AWS) and Google. These custom chips don't have a markup, and it makes it even harder for Intel to compete with.
- Read more: Intel announces Gaudi 3 AI accelerator: 128GB HBM2e at up to 3.7TB/sec, up to 900W power
- Read more: NVIDIA CEO: our next-gen Blackwell GPU will cost $30,000 to $40,000 each
Getting into specifics of the new Gaudi 3 AI accelerator: Intel has 32 Tensor Cores on each chip, for a total of 64 Tensor Cores. Each chip features 48MB of SRAM, for a total of 96MB of SRAM per full package. The SRAM on the Intel Gaudi 3 AI accelerator features 12.8TB/sec of bandwidth, supported by the HBM memory on the Gaudi 3. The 128GB of HBM2e memory features up to 3.7TB/sec of memory bandwidth.
Intel's new Gaudi 3 AI accelerator in its PCIe form factor design (HL-388) uses the newer PCIe 5.0 interface with full PCIe 5.0 x16 lanes. In its PCIe 5.0 form, the new Intel Gaudi 3 AI accelerator has a TDP of 450W and 600W, a power level we don't see much in this form factor. Intel also has an OAM version (HL-328/325L/335) with a TDP of between 450W and 900W for server air cooling, while the water-cooled Gaudi 3 in OAM form features up to 900W TDP.