---
title: Intel Unveils D270 GPU for LLM Inference with 480GB Memory
url: https://www.elseif.net/intel-unveils-d270-gpu-for-llm-inference-with-480gb-memory
published: 2026-10-16T14:03:34+00:00
language: en
section: Models
source: https://www.blognone.com/node/151897
organizations: Intel
publisher: elseif
---

# Intel Unveils D270 GPU for LLM Inference with 480GB Memory

Intel has introduced the D-series GPU, specifically the D270 model, designed primarily for running large language model inference workloads. The chip, which carries the code name Crescent Island, is built to handle large parameter models on a single card or in multi-card configurations. The primary selling point of the D270 is its high memory capacity, which allows it to support models with up to 300 billion parameters on a single unit.

When connected in a set of eight cards, the system can handle models with over one trillion parameters. The card features 480GB of LPDDR5X memory, providing a memory bandwidth of 1.5 terabytes per second. It delivers 1.3 petaflops of performance in FP8 floating-point operations. The hardware is developed on the Xe3P architecture and includes 32 cores along with 256 XMX matrix processing units. The card uses a standard PCIe 5.0 interface and is designed for air cooling.

It has a power consumption of 350 watts and is intended for use in existing traditional data centers. Intel has removed traditional 3D graphics processing hardware and ray tracing units from this design. This decision was made to maximize the area dedicated to AI processing, as the card is not intended for graphics tasks or gaming.