Intel gave the first architectural details of its Gaudi 3 third-generation AI processor at Vision 2024 this week in Phoenix, Arizona. Gaudi 3 is made up of two identical silicon dies, each with a central region of 48 megabytes of cache memory, joined by a high-bandwidth connection, surrounded by four engines for matrix multiplication and 32 programmable units called tensor processor cores. It produces double the AI compute of Gaudi 2 using 8-bit floating-point infrastructure. It also provides a fourfold boost for computations using the BFloat 16 number format.

Intel gave the first architectural details of its Gaudi 3 third-generation AI processor at Vision 2024 this week in Phoenix, Arizona. Gaudi 3 is made up of two identical silicon dies, each with a central region of 48 megabytes of cache memory, joined by a high-bandwidth connection, surrounded by four engines for matrix multiplication and 32 programmable units called tensor processor cores. It produces double the AI compute of Gaudi 2 using 8-bit floating-point infrastructure. It also provides a fourfold boost for computations using the BFloat 16 number format.