ព័ត៌មានសំខាន់
Intel has officially introduced its Crescent Island D-Series GPU, a dedicated AI accelerator designed for server environments. This hardware marks a shift for the company following the stalling of previous initiatives.
The D-Series utilizes Xe3P cores, featuring 256 XMX engines and 32 MB of unified L2 cache. The architecture supports memory capacities up to 480 GB of LPDDR5X, with a base configuration of 160 GB for Intel-branded units. Performance is rated at 1.3 PetaFLOPS for FP8, with additional support for FP4 through FP64 formats.
The Xe3P silicon is optimized strictly for AI inference, omitting traditional graphics output elements. It features a redesigned 16-deep systolic pipeline for XMX engines, alongside 1 MB of register files and 512 KB of L1 cache per core, specifically targeting agentic tasks and long-context workloads.
Designed as a 350 W air-cooled PCIe 5.0 x16 card, the accelerator is intended for standard server infrastructure. While customer sampling has commenced, Intel has not yet announced pricing or general availability. Future market success will depend on software maturity and the cost per generated token.
