Intel reveals the first details of its Ponte Vecchio Xe-HPC graphics accelerator, intended for data center operations.
Intel Xe-HPC ‘Ponte Vecchio’ with up to 128 Xe cores
Intel’s ‘Ponte Vecchio’ is available in the latest A0 version, which achieves a unique precision performance of 45 TFLOPs. This graphics accelerator also offers support for ray tracing. In total, the chip has up to 128 Xe cores and 128 dedicated ray-tracing units.
This is the first data center accelerator from Intel to use the new Xe HPC architecture. The chips will include computing, Rambo, HBM, and EMIB modules, for a total of 47 modules with 100 billion transistors.
The Xe-HPC section “Xe-Core” consists of a block with 8 vector and 8 matrix engines. The difference with the Xe-HPG architecture is that Ponte Vecchio will have fewer engines, but they will run on higher buses (512 bits and 4096 bits, respectively, versus 256b and 1024b).
Ponte Vecchio’s main block is called Xe-HPC Slice, which combines 16 Xe cores. Within the same block are 128 ray-tracing units. Each Xe core is connected to an RT unit, although these cores are not intended for gaming, but for other tasks.
With these specifications, Intel’s graphics accelerator will be much more powerful than the green company’s Nvidia A100. We’re talking about a performance of 45 TFLOPs versus 19.5 TFLOPs.
The GPU is expected to make its official debut in 2022.