Intel Announces 350W D270 GPU With Massive Financial Computing Gains
Intel announced its 350W data center GPU D270, codenamed Crescent Island and built on the Xe 3P architecture, delivering up to 5.9x higher financial computing throughput than NVIDIA's 600W RTX PRO 6000 Blackwell server GPU. The accelerator features 32 Xe cores, 256 XMX engines, and native support for precision levels ranging from FP4 to full-rate FP64 double precision. This launch highlights Intel's competitive push into high-performance computing (HPC) and financial modeling by combining energy efficiency with massive memory bandwidth. By providing a 350W air-cooled solution that outperforms a higher-wattage competitor, Intel offers data centers a powerful alternative for mixed AI and numerical simulation workflows. The D270 card comes equipped with 160 GB of LPDDR5X memory providing 1.5 TB/s of bandwidth, which is scalable up to 480 GB in partner designs to keep massive datasets resident on the GPU. It also integrates into a decoupled architecture alongside SambaNova SN50 RDU units and Intel Xeon 6 processors to accelerate agentic AI workflows.
## BACKGROUND
High-performance computing (HPC) often requires double-precision (FP64) calculations to prevent rounding errors in scientific and financial simulations. Modern data center architectures are increasingly shifting towards heterogeneous systems that pair traditional numerical solvers with AI inference engines to handle complex agentic workloads.