Intel made major waves at Computex 2026 in Taipei by launching its next-generation Xeon 6+ processors, expanding its 800 Series Ethernet portfolio, and detailing its upcoming Crescent Island data center GPU. The new silicon lineup is engineered specifically to tackle data center density, power limitations, and the heavy orchestration demands of modern, agentic AI workloads.
Intel Xeon 6+: Performance Density for Agentic AI
Built on the advanced Intel 18A process node, the Xeon 6+ processors are designed for cloud-native environments where throughput per core, latency predictability, and rack power limits are critical.
-
Massive Core Counts: Features up to 288 Efficient-cores (E-cores) on a single chip.
-
Generational Leaps: Delivers up to a 2.5x performance boost compared to the previous generation, alongside a 45% improvement in performance-per-thread-per-watt.
-
Infrastructure Consolidation: Supports up to a 9:1 server consolidation ratio, drastically lowering data center physical footprints.
-
Next-Gen I/O: Outfitted with 12-channel DDR5 memory support, 96 lanes of PCIe Gen 5, Compute Express Link (CXL) support, and silicon-level security features.
Intel confirmed the processors are already undergoing active testing within telecom network infrastructures and are being prepared for immediate data center deployment.
Intel Ethernet E835: High-Efficiency 200GbE Networking
To eliminate networking bottlenecks as distributed AI workloads scale, Intel introduced the Ethernet E835 controllers and adapters. These solutions target enterprise cloud setups and high-throughput AI fabrics.
-
Massive Throughput: Supports data rates ranging from 10GbE up to 200GbE across versatile port layouts (including 2x25GbE, 4x25GbE, 1x200GbE, and 2x200GbE).
-
Leading Power Efficiency: Intel claims the E835-CQDA2 adapter achieves up to 1.9x higher performance-per-watt than Nvidia’s ConnectX-6 Dx and up to 1.4x over Broadcom’s BCM957508.
-
Enterprise Adoption: Major server vendors—including Cisco, Dell, HPE, Lenovo, and Supermicro—are already actively integrating the E835 architecture into their upcoming hardware blueprints.
Intel Crescent Island GPU: Tailored for Edge and AI Workloads
Intel also lifted the curtain on Crescent Island, its next-generation data center GPU purpose-built to execute large, token-heavy AI workloads at a lower total cost of ownership (TCO).
-
Xe3P Architecture: Built on Intel’s newest graphics architecture, focusing heavily on performance-per-watt gains while retaining backward software compatibility with the existing Xe ecosystem.
-
High-Capacity Memory: Packs up to 480GB of LPDDR5X memory, giving it the massive bandwidth and capacity required to keep large language models (LLMs) cached locally.
-
Standard Form Factor: Designed around a 350W air-cooled PCIe card footprint, making it easy to drop into standard, pre-existing data center server racks without requiring complex liquid-cooling overhauls.
