While many companies are turning to NVIDIA’s solutions to set up and run work tasks linked to artificial intelligence, IBM Cloud is turning to Intel’s offering. The Gaudi 3 accelerators are advertised as being affordable yet offering respectable performance. In benchmark tests, the chips rival NVIDIA’s H100!
Gaudi 3: it rivals the H100 and H200, and it’s cheaper!
Clearly, IBM’s main motivation in choosing Intel accelerators is money. When you put it like that, it sounds very penny-pinching, but it has to be said that Intel’s solution seems to offer a very interesting performance/price ratio. Indeed, benchmarks have been carried out outside the company’s internal channels. These, on models such as Granite-3.1-8B, Mixtral-8x7B or Llama-3.1-405B, showed that the Intel chip rivalled NVIDIA’s H200 and H100.
Between all these chips, the blue solution is far more advantageous in terms of cost. Yes, an H100, depending on the configuration (SXM, PCIe, etc.) is estimated to cost between $30000 and $40000. In contrast, an Intel solution is valued at around $16,000. The price difference is massive!
What’s more, this is good news for Intel, since IBM Cloud is, so to speak, the first CSP( CloudServices Provider ). It has to be said that Intel’s start in artificial intelligence has been slow, as the brand has been slow to invest in the sector. While NVIDIA has been the first to invest in cutting-edge, very expensive hardware, the blue chip may have a card to play by focusing on solutions that are accessible in terms of cost! In fact, Intel’s communication seems to be heading in this direction:
” This collaboration underscores our shared commitment to making AI more accessible and cost-effective for enterprises worldwide
– Saurabh Kulkarni, Vice President at Intel.
