Intel Shares First Benchmarks of Crescent Island D270 GPU
Intel has released initial benchmarks for its next-generation Crescent Island D270 GPU, highlighting a flagship accelerator designed around the Xe3P architecture. The D270 achieves up to 5.9x higher FP64 throughput and competitive AI benchmark performance compared to Nvidia's high-end hardware. This development positions Intel as a serious alternative to Nvidia in the high-performance computing and AI inference markets. The promising throughput and lower latency could offer developers more viable hardware options outside of the dominant Nvidia ecosystem. In specific AI benchmarks, the D270 GPU delivered 29% lower latency and 1.12x throughput compared to the Nvidia RTX PRO 6000 Blackwell. The chip features 32 Xe cores, 256 XMX engines, native FP4 to FP64 support with full-rate double-precision compute, and draws 350W.
## BACKGROUND
The Intel Crescent Island GPU is built using a 3 nm production process and features scalable LPDDR5X memory ranging from 160GB up to 480GB. FP64 refers to 64-bit floating-point precision, which is critical for heavy scientific simulations and high-performance computing (HPC) workloads.