Intel® GPU D-series
Intel® GPU D-series (Intel® GPU D270) is an inference-optimized data center GPU built for agentic AI workloads. It pairs up to 480 GB of LPDDR5X memory per card with 32 Xᵉ3P cores and 256 XMX matrix engines in a standard, air-cooled PCIe 5.0 form factor, deliberately omitting high bandwidth memory (HBM).
Key facts
What is Intel® GPU D-series?
Scale inference capacity — larger contexts, more concurrent sessions and more tokens — within the power, racks and standard PCIe servers you already run, without HBM.
Who is Intel® GPU D-series best for?
Data center operators, AI infrastructure architects, enterprises and cloud providers running large-scale LLM inference and agentic AI workloads
What are its main limitations?
- No pricing information was published on the site (the pricing page returned an error page)
- No HBM — the design targets memory-bound inference rather than HBM-class workloads
- No published benchmarks or quantified performance figures on the page
Limitations are based on product materials or editorial synthesis; confirm on the official site.
When was this information verified?
October 20, 2026
Sources
- Official homepageofficial homepage · Oct 20, 2026
Key features
- ✓Up to 480 GB LPDDR5X memory per card for memory-bound inference
- ✓32 Xᵉ3P cores and 256 XMX Xᵉ3P matrix engines to accelerate prefill and prompt processing
- ✓350 W air-cooled, passive cooling design — no liquid-cooling retrofit required
- ✓Standard PCIe 5.0 form factor and interconnect using existing servers, with no proprietary networking
- ✓Die area reallocated from graphics and ray-tracing pipelines to Xᵉ3P cores and matrix compute
- ✓Precision flexibility from native low-precision inference up to FP64 scientific computing
- ✓Zero-day support for leading open models, served on vLLM and the open PyTorch ecosystem
- ✓Designed to keep models and long context resident without high bandwidth memory (HBM)
Use cases
- →Running agentic AI fleets with 100K-token contexts and dozens of concurrent sessions
- →Whole-context reasoning with more of the model and long context kept resident in memory
- →Efficient token serving to maximize tokens per rack and per watt
- →Scaling inference capacity within existing power and rack budgets
- →FP64 scientific computing alongside AI inference on the same platform
Pros
- +Very large 480 GB LPDDR5X memory pool per card for memory-bound inference
- +Standard PCIe 5.0 air-cooled design fits existing servers with no bespoke networking
- +350 W passive cooling avoids liquid-cooling retrofit between evaluation and deployment
- +Open by design with vLLM and PyTorch ecosystem support and zero-day support for leading open models
Cons
- −No pricing information was published on the site (the pricing page returned an error page)
- −No HBM — the design targets memory-bound inference rather than HBM-class workloads
- −No published benchmarks or quantified performance figures on the page
Positioning
- Core value: Scale inference capacity — larger contexts, more concurrent sessions and more tokens — within the power, racks and standard PCIe servers you already run, without HBM.
- Ideal for: Data center operators, AI infrastructure architects, enterprises and cloud providers running large-scale LLM inference and agentic AI workloads
- Product type: Data center GPU card (standard PCIe 5.0 form factor, 350 W air-cooled / passively cooled)
Frequently asked questions
What is Intel® GPU D-series?
Intel® GPU D-series is a line of inference-optimized data center GPUs; the featured model, Intel® GPU D270, is built for agentic AI workloads with long contexts and many concurrent sessions.
How much memory does the Intel® GPU D270 have?
Up to 480 GB of LPDDR5X memory per card. The design uses LPDDR5X rather than high bandwidth memory (HBM).
What are the power and cooling requirements?
The card is rated at 350 W with air cooling and passive cooling, so no liquid-cooling retrofit is required.
What form factor and interconnect does it use?
It uses a standard PCIe 5.0 form factor and interconnect, deployed in existing servers with no proprietary networking or bespoke systems.
Which software frameworks are supported?
Intel states zero-day support for leading open models served on vLLM and the open PyTorch ecosystem.
How much does Intel® GPU D-series cost?
Pricing is not published on the Intel website; the pricing page was unavailable, so no price or plan details could be confirmed.
You may also like
Curated suggestions based on similar categories.
AI Manga Translator
AI Manga Translator is a web tool and Chrome extension that translates raw manga, comics and scanned comic images into multiple languages while keeping the structure of the original image intact. It uses AI translation engines such as GPT-4o, DeepL, Papago and Google Translate, and offers a free tier of 3 translations per user per week.
AI Manga Translator
AI Manga Translator is a web-based tool that translates raw manga and comic images into multiple languages in one click while preserving the structure and style of the original artwork. It also offers a Chrome extension for reading translated manga directly on manga sites.
AutoChart.eu
Autochart.ai is an AI-powered clinical documentation platform built by clinicians for clinicians. It captures patient encounters to generate notes, referrals, emails, and other documentation, and also offers products for patient intake/voice workflows and practitioner training. The platform supports any EMR and emphasizes healthcare privacy compliance.
MP3 to Transcript
ToText is a browser-based transcription tool that converts MP3, MP4, M4A, MOV, WAV, WebM and other audio/video files into editable, time-aligned text. It adds speaker labels, timestamps, summaries, meeting notes, chapters, action items and translations, with exports to TXT, DOCX, PDF, SRT, VTT, CSV, JSON, ASS and XLSX.
ReWords AI
ReWords AI is a browser-based AI image text editor that detects and replaces baked-in text in PNG, JPG, and WebP images while making best-effort matches to the original font, color, and background. It also supports removing, translating, extracting, and blurring text without needing source design files.
Step 5 Preview
Step 5 Preview is StepFun's flagship model for agentic work, built on a sparse Mixture-of-Experts architecture with 600B total parameters (27B active per token), a 1M-token context window, and vision input. It targets software engineering, long-horizon agentic tasks, professional knowledge work, and finance, and is available through StepFun's products and API, with open weights planned for October 15.