At Computex 2026, Intel announced its strategy for the Data Center Group (DGC), outlining plans to scale operations to meet the evolving demands of modern workloads. The company highlighted how the surge in artificial intelligence has reshaped data center requirements, emphasizing the need to boost performance‑per‑watt, enhance per‑core efficiency, increase core density per rack, and expand memory bandwidth.
Team Blue acknowledged that today’s data centers must pair accelerators—particularly GPUs—with traditional x86 CPUs for both training and inference tasks. In essence, foundational data centers are transforming into AI training hubs. Over the next five years, Intel projects that AI workloads will account for 50% of all data center traffic, with the bulk of that activity focused on inference.
While we previously covered Intel’s CPU advancements, this piece zeroes in on Crescent Lake GPUs, Team Blue’s latest high‑performance graphics card designed to accelerate AI inference workloads.
Crescent Island is built on Intel’s Arc Xe 3P architecture, which also underpins the current Panther Lake integrated GPUs. As Intel’s flagship card to date, it offers up to 480 GB of VRAM—a substantial leap from the 160 GB buffer it shipped with when first revealed last year.
Unlike many high‑end professional GPUs that rely on HBM for power efficiency, Intel’s new GPU utilizes LPDDR5X memory, marking a strategic departure aimed at balancing performance with energy consumption.
The new Intel workstation GPUs are equipped with an advanced air-cooling system capable of handling a 350‑watt TDP, ensuring robust performance under heavy loads. According to Intel, these cards are engineered to tackle next‑generation AI workloads, supporting an extensive spectrum of data types and micro‑scaling formats—from native FP4/MXFP4 to FP64 and beyond—making them versatile for a wide range of professional applications.
News Source: Neowin
No comments