Intel Unveils D270 Crescent Island GPU at SC26

The new Xe3P-based hardware outperforms NVIDIA's Blackwell Server Edition in financial and AI workloads while using significantly less power.

Intel has officially unveiled its latest high-performance hardware at SC26, showcasing the D270 GPU, codenamed "Crescent Island." Built on the new Xe3P architecture, the chip features 32 Xe3P cores and marks a significant push into the specialized compute market. The hardware is designed for high-density environments, utilizing an LPDDR5X memory subsystem that supports up to 480 GB of memory. Intel is offering the D270 in several configurations, including standard air-cooled 350W parts, liquid-cooled variants, and dual-GPU options for more demanding deployments.

Efficiency Gains in Financial Services and AI

The core of Intel's presentation focused on how the D270 stacks up against the competition, specifically the NVIDIA RTX PRO 6000 Blackwell Server Edition. Despite the D270 operating at a 350W thermal design power—significantly lower than the 600W required by the NVIDIA card—Intel reported substantial performance leads in targeted sectors. In four Financial Services workloads, the Crescent Island GPU secured an average 4.5x performance lead, peaking at a 5.9x advantage over the Blackwell-based rival.

Beyond financial modeling, Intel showcased the D270's versatility through a Seismic modeling demo utilizing FP32 compute. The GPU also demonstrated its strengths in artificial intelligence benchmarks, where it offered 29% lower latency and 1.12x the throughput of the RTX PRO 6000 Blackwell. These gains are supported by a flexible compute engine that provides native FP64 double-precision capabilities alongside support for lower precision FP4-FP8 operations, positioning the D270 as a highly efficient alternative for enterprise-scale AI and scientific research.