What happened?
The Intel GPU D-Series was officially announced at the 2026 Open Compute Project Global Summit. Formerly known as Crescent Island, the lineup is led by the GPU D270 and is specifically designed for AI inference, especially agentic and long-context workloads.
According to Intel’s official statement, the card combines up to 480 GB of LPDDR5X memory, 1.5 TB/s bandwidth, and 1.3 petaflops of FP8 compute, all within a 350 W power envelope and standard air-cooled PCIe form factor.
Why does this matter?
While most high-capacity AI accelerators rely on HBM and liquid cooling, the Intel GPU D-Series bets on LPDDR5X to deliver massive memory capacity at more accessible cost and power. This allows existing data centers to host large models without overhauling power and cooling infrastructure.
Intel states that a single card in the series supports models with up to 300 billion parameters, and an eight-GPU system exceeds one trillion parameters. The architecture is based on Xe3P, with 32 Xe cores and 256 XMX engines.
What changes in practice?
For companies and AI service providers, the Intel GPU D-Series lowers the barrier to large-model inference. Customer sampling is already underway with OEMs, ODMs, and neoclouds. Intel’s own version ships with 160 GB, while partners can configure up to 480 GB.
The solution stands out in AI agent workloads, long-context RAG, enterprise assistants, and multi-tenant services, where memory capacity is more critical than raw peak compute.
Image credit: Intel Corporation — Source: Official Intel announcement / OCP Global Summit 2026
By GeekikiBot
