---
title: "Intel launches GPU D270 for AI inference with up to 480GB of memory on one card"
url: https://anawords.com/story/intel-gpu-d270-crescent-island-ai-inference-card
published: 2026-10-13T11:12:02+05:30
updated: 2026-10-13T11:12:02+05:30
section: Chips & Compute
author: Mayank Sahu
checked: sources checked by AI, approved by editor
version: 1
---

# Intel launches GPU D270 for AI inference with up to 480GB of memory on one card

The air-cooled 350W PCIe card, earlier codenamed Crescent Island, uses LPDDR5X instead of HBM. Intel says sampling has begun; its own card ships with 160GB.

**In short:** Intel introduced the GPU D270 at the OCP Global Summit on 12 October 2026. The AI inference card, earlier codenamed Crescent Island, offers up to 480GB of LPDDR5X memory in a 350W, air-cooled PCIe card. Intel says customer sampling is under way.

## What launched

Intel introduced its GPU D-Series of AI inference accelerators, earlier codenamed Crescent Island, at the Open Compute Project Global Summit on 12 October 2026.[^2][^3] The first card is the GPU D270.[^2][^3]

## Key specs

- **Memory:** up to 480GB of LPDDR5X per card, without HBM.[^1][^2][^3] StorageReview reports that Intel’s own branded card has 160GB, with 480GB versions coming from partners.[^2]
- **Compute:** 32 Xe3P cores and 256 XMX matrix engines.[^1][^3] StorageReview reports 1.3 petaflops of FP8 compute and 1.5TB/s of memory bandwidth.[^2]
- **Power and form:** 350W, air-cooled, standard PCIe 5.0 card that fits existing servers.[^1][^3]

Intel says one card can hold models of up to about 300 billion parameters, and an eight-card server can go past one trillion, according to StorageReview.[^2]

## Why the memory matters

Running large models needs a lot of memory. LPDDR5X is slower than the HBM used on top AI chips but cheaper and easier to get, which matters during the current memory shortage.[^2] StorageReview notes that Intel’s current Gaudi 3 card has 128GB of HBM2e.[^2]

## Software

Intel’s page lists support for vLLM and PyTorch.[^1] Phoronix reports Linux driver support is progressing upstream.[^3]

## Availability

Intel says customer sampling is under way with server makers and cloud providers.[^2] Intel’s product page gives no shipping date or price.[^1]

## Same event

Nvidia contributed its Vera Rubin rack designs at the same summit, which we [cover separately](/story/nvidia-vera-rubin-rack-designs-ocp-mgx-1-2).

## Fact ledger

- [Reported] Intel introduced the GPU D270 at OCP Global Summit on 12 October 2026. (source 2,3)
- [Confirmed] Up to 480GB of LPDDR5X memory per card, 350W, air-cooled PCIe 5.0. (source 1,2,3)
- [Company claim] 1.3 petaflops FP8 and 1.5TB/s bandwidth. (source 2)
- [Unconfirmed] Customer sampling is under way. (source 2)
- [Not known] Price and shipping date.

## Sources

1. [Intel: GPU D-Series product page](https://www.intel.com/content/www/us/en/products/details/discrete-gpus/data-center-gpu/d-series.html) — official
2. [StorageReview: Intel GPU D-Series packs 480GB LPDDR5X](https://www.storagereview.com/news/intel-gpu-d-series-480gb-lpddr5x-1-3-pflops-fp8-350w-pcie-card)
3. [Phoronix: Intel announces the GPU D Series for Crescent Island](https://www.phoronix.com/news/Intel-GPU-D-Series)