Intel dubs Crescent Island D-Series GPU, launching sampling for 350W air-cooled servers
The new accelerator delivers 1.3 PFLOPS FP8 and 480GB memory for enterprise servers.
Key points
- Intel officially named the Crescent Island chip the GPU D-Series at OCP 2026.
- The device targets air-cooled rack servers with a 350W thermal design power.
- Customer sampling is underway, though no broad launch date has been confirmed.
Intel announced at the Open Compute Project 2026 event that its Crescent Island AI inference accelerator will be marketed as the GPU D-Series family. According to Tom's Hardware, the company revealed theoretical peak performance figures and final memory specifications for the chip. This disclosure follows a series of gradual updates Intel has provided throughout 2026 regarding the hardware's capabilities and intended deployment environments.
Root cause
The GPU D-Series is built on Intel's Xe3P graphics architecture, which functions primarily as a massively parallel computing platform. Tom's Hardware notes that the chip lacks most or all graphics-specific functional units, distinguishing it from traditional discrete GPUs. This architectural choice aligns with the product's focus on enterprise AI inference workloads rather than general-purpose graphics rendering or consumer gaming applications.
Attack path
The hardware is designed for deployment in traditional air-cooled rack servers, scaling up to eight GPUs per system. Intel states that this configuration can support inference on models exceeding one trillion parameters. The chip is positioned for large language models, long-context inference, and concurrent services, aiming to serve multi-tenant AI, robotics, and other latency-sensitive tasks without requiring liquid cooling infrastructure.
Affected versions
Intel is releasing the GPU D-Series in an air-cooled PCIe form factor with a 350W thermal design power. Each unit includes up to 480GB of LPDDR5X memory, delivering 1.5TB/s of bandwidth. The system achieves 1.3 PFLOPS of FP8 compute throughput. This performance envelope differs significantly from liquid-cooled competitors, such as Nvidia's Rubin GPUs, which target higher throughput but require different infrastructure support.
Mitigation
Intel is leveraging its common Xe software foundation to support these accelerators, a stack already established for AI across its integrated and discrete GPU products. The company is collaborating with a broad ecosystem of OEMs, ODMs, neocloud providers, and independent software vendors. Customer sampling of the GPU D-Series products is currently in progress, according to Tom's Hardware, although specific details regarding a general market launch remain undisclosed.
What to do and how to stay safe: Intel
- Review current server cooling capacity to ensure compatibility with 350W air-cooled PCIe accelerators.
- Evaluate existing AI inference workloads for suitability with the Xe3P architecture and software stack.
- Monitor Intel’s ecosystem partners for updated software drivers and firmware during the sampling phase.
- Assess memory bandwidth requirements of large language models against the 1.5TB/s LPDDR5X specification.
Step-by-step guide: Prevent Prompt Injection Attacks: Practical Defences for AI Systems
General security guidance from the TechSec Observer newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is the official name for Intel's Crescent Island chip?
Intel has officially named the Crescent Island AI accelerator the GPU D-Series.
Does the GPU D-Series require liquid cooling?
No, the GPU D-Series is designed for air-cooled rack servers with a 350W TDP.
What is the memory bandwidth of the GPU D-Series?
The chip delivers 1.5TB/s of memory bandwidth via up to 480GB of LPDDR5X.
