Intel Targets High-Performance Computing Beyond AI Inference, Broadening GPU D270's Applications
- Input
- 2026-11-17 13:56:00
- Updated
- 2026-11-17 13:56:00

[Financial News] Intel is setting its sights on the high-performance computing (HPC) market with a graphics processing unit (GPU) specialized in artificial intelligence (AI) inference. As more industries, including scientific research and finance, combine AI inference with simulations, the company aims to broaden the range of GPU applications by leveraging high-precision computing and large memory capacity.
Expanding GPU applications into finance and science
Intel said on the 17th that it would expand the use of the AI-inference-optimized Intel GPU D270, code-named Crescent Island, into high-performance computing.In the high-performance computing market, there is a growing trend to combine data analysis and AI inference with conventional numerical modeling and simulation. The emergence of agentic AI, in particular, is changing research methods, with AI analyzing intermediate simulation results and autonomously adjusting the next steps in a computation.
Based on the Xe 3P architecture, the Intel GPU D270 is equipped with 32 Xe cores and 256 XMX engines. It supports everything from the low-precision computing required for AI inference to high-precision (FP64) computing needed to ensure mathematical accuracy in scientific and financial applications.
It is equipped with 160 GB (gigabytes) of LPDDR5X memory and provides memory bandwidth of 1.5 TB (terabytes) per second. Depending on the partner's design, memory capacity can be expanded to up to 480 GB. With a 350 W (watt) power design and air cooling, it can also be installed in existing enterprise servers.
In its own performance tests using pre-production chips, Intel recorded throughput of up to 5.9 times that of NVIDIA Corporation's RTX PRO 6000 Blackwell Server Edition when valuing financial options. The comparison GPU has a 600 W power design, higher than the D270's.
Intel believes the D270 can also improve computing efficiency through its large memory capacity in seismic modeling, which is used in the oil and gas industry.
Meeting demand for AI and high-performance computing with an open ecosystem
Intel also plans to improve AI inference efficiency by combining its GPUs with AI chips from other companies.The Intel GPU D270 handles the prefill stage, in which AI processes input information, while SambaNova Systems' SN50 AI accelerator handles the decode stage, in which it generates responses. Intel Xeon processors manage the overall workflow and data communications. The key is to reduce bottlenecks by using different chips for different computational tasks.
Intel is also putting effort into building an open software ecosystem. This approach helps developers use existing code without being locked into a proprietary programming environment from a particular company.
Intel said it used the open programming model SYCL to port an existing seismic imaging application to the D270 environment within a month.
Intel plans to demonstrate a disaggregated AI inference architecture combining seismic modeling technology with SambaNova Systems' AI accelerator at Supercomputing 2026 (SC26), which is being held in Chicago, United States.
[email protected] Yoon Hong-jip Reporter