Intel has finally unveiled what its "Crescent Island" D-Series GPUs will look like, introducing a dedicated AI accelerator in a server form factor after previous projects stalled. The newly named D-Series GPU features a total of 32 Xe3P cores, 256 XMX engines, 32 MB of unified L2 cache, and up to 480 GB of LPDDR5X memory. Intel rates it for up to 1.3 Peta FLOPS of FP8 performance, with support for formats ranging from FP4 to FP64.
A media engine with four encoders and four decoders is also included, which could assist with multimodal workloads where video generation and understanding is a must. The card comes with 160 GB memory configuration out of the box, which is exclusive to Intel-branded "Crescent Island" graphics cards, while the company allows its ODMs to equip the card with up to 480 GB of LPDDR5X memory for maximum VRAM capacity. What is particularly interesting about the Xe3P is its pipeline.
The XMX engines have been redesigned from the four-deep systolic design on the regular Xe3 to 16-deep processing on the Xe3P. Additional details include FP4 data format support alongside FP8, as well as large register files of 1 MB per core and L1 cache of 512 KB per core. In total, there are 256 Vector engines and 256 of the new XMX engines, each divided into groups of eight per Xe3P core.
Additionally, in the Xe3P, there are no graphics processing elements to output imagery, as the silicon is designed purely for AI inference workloads that require long context and agentic tasks. The accelerator is designed as a 350 W, air-cooled PCIe 5.0 x16 card, allowing it to fit into conventional server infrastructure without the need for dedicated liquid cooling. Intel is also leveraging its common Xe software foundation and support for established AI frameworks to facilitate deployment.
Customer sampling is underway, but pricing, general availability, and independent performance results have yet to be disclosed. Ultimately, the cost per generated token and software maturity will determine whether the D-Series can compete with established AI accelerator platforms. Intel envisions a single card powering models up to 300 billion parameters in size, with a 8x GPU cluster running trillion-parameter models without issues.
Source: TechPowerUp News · Summarized by HeadlinesBriefing