Intel Unveils GPU D270: 480GB LPDDR5X Memory Enables Single-Card 300B Model Inference

1 reports archived1 independent sourcesupdated 10/13/2026, 09:04:37

Synthesis & Latest Updates

1 Sources Cross-Validated

Intel officially named its data center inference GPUs as the Intel GPU D-series at the OCP Summit, launching with the D270 model based on the Xe3P architecture. Featuring up to 480GB of LPDDR5X memory and 1.5TB/s FP8 throughput, the card targets long-context and large-model inference bottlenecks, supporting 300B parameter models on a single card.

LATEST/Intel officially named its data center inference GPUs as the Intel GPU D-series at the OCP Summit, launching with the D270 model based on the Xe3P architecture. Featuring up to 480GB of LPDDR5X memory and 1.5TB/s FP8 throughput, the card targets long-context and large-model inference bottlenecks, supporting 300B parameter models on a single card.

HEAT TRENDHourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

TIMELINECoverage timeline

Total 1 reports · Latest first

IT之家 · 智能时代T2·Oct 13, 09:0468 pts
Intel Unveils GPU D270: 480GB LPDDR5X Memory Enables Single-Card 300B Model Inference
- The Intel GPU D270 uses the Xe3P architecture with up to 480GB of LPDDR5X memory and 1.5TB/s bandwidth.
- It delivers 1.5TB/s FP8 compute power within a 350W TDP, featuring standard PCIe AIC form factor and air cooling support.
- Optimized for LLMs and long contexts, it supports 300B models on a single card and 1T models in an 8-card system.
View Evidence Original Source