Air-cooled with 480 gigabytes: Intel details its D-Series data centre GPU
Intel has released technical details of its GPU D-Series: up to 480 GB of memory at 350 watts with air cooling. It also unveiled Xeon racks with more than 46,000 processor cores.
Symbolic image · Photo: PiDatacenters / Wikimedia Commons, CC BY-SA 4.0 (cropped and resized)
Intel on Monday released new technical details of its data centre graphics processor line, previously code-named Crescent Island and now called the GPU D-Series. The chips are based on the Xe3P architecture and come with up to 480 gigabytes of LPDDR5X memory and 1.5 terabytes per second of memory bandwidth. Intel puts compute performance at 1.3 petaflops in the FP8 data format within a 350-watt power envelope.
A single card is designed to run AI models with up to 300 billion parameters, and an eight-GPU system models with more than one trillion parameters. The cards use the standard PCIe form factor and are air-cooled, which should ease deployment in existing data centres without liquid cooling. Intel's own card has 160 gigabytes of memory, while partners can build versions with up to 480 gigabytes. Customer sampling is under way.
Intel also presented rack designs with Xeon 6+ processors that scale beyond 46,000 processor cores per server rack, with total rack power of 133.45 kilowatts. Many of the designs follow the Open Compute Project's open Open Rack v3 standard.