--- title: "Intel Unveils Dual-Chip AI Accelerator Based on D270: Two GPUs with 32 Xe3P Cores and Copper Water Block" description: "Intel has showcased a dual-chip AI accelerator based on Crescent Island D270 components: featuring two GPUs with 32 Xe3P cores, PLX interconnect, and liquid cooling with a copper water block." date: 2026-10-17T11:38:18.475Z lang: en url: https://xab.info/en/posts/intel-unveils-dual-chip-ai-accelerator-crescent-island tags: [intel, ai-accelerator, crescent-island, gpu, liquid-cooling] author: "Alexander Buran" publisher: "XAB.info" --- # Intel Unveils Dual-Chip AI Accelerator Based on D270: Two GPUs with 32 Xe3P Cores and Copper Water Block ![Engineer holding the Intel Crescent Island dual-chip AI accelerator with a copper water block](https://xab.info/media/2026/10/17/intel-dvuhchipovyi-ii-uskoritel-crescent-island/intel-dvuhchipovyi-ii-uskoritel-crescent-island-1.webp) ## Key Points - Intel introduced a dual-chip AI accelerator using the same components as the Crescent Island D270. - Each of the two GPUs contains 32 Xe3P cores, connected via a PLX interface, with liquid cooling featuring a copper water block. - Intel plans to sell accelerators separately and in systems: up to 8 cards per device and 80 in server racks; retail availability of the dual-chip card has not been announced. Immediately after presenting the D270 AI accelerator from the Crescent Island lineup, Intel has revealed another configuration using the same components—a dual-chip adapter demonstrated just days ago. The new product is built on the same graphics accelerators as the base D270, but combines two such chips into a single package, significantly scaling up computational power for artificial intelligence tasks. The showcased prototype drew attention primarily due to the unusual design of its cooling system and a distinctive copper water block bearing the Intel logo. ### Dual-Chip Configuration Using the Same Components According to the disclosed details, the dual-chip adapter relies on two GPUs identical to those powering the D270. Each of these accelerators contains 32 cores based on the Xe3P architecture. The chips are interconnected via a PLX interface, establishing communication between the two compute blocks inside a single device. Thus, the company effectively doubles the core count within a single slot while maintaining the same underlying hardware base as the already announced model. ### Unusual Cooling: Copper Water Block and Liquid Cooling System Cooling for the dual-chip solution is handled by a liquid cooling system. Published images reveal a massive copper water block engraved with "intel" covering the main surface of the board, along with a characteristic yellow quick-disconnect fitting for attaching the cooling loop. This approach is typical for high-load computing platforms where passive or air cooling is insufficient, allowing the temperatures of two powerful GPUs to be kept within operational limits under sustained AI workloads. ### Form Factor and Open Questions Regarding Mass Production At the same time, it remains unclear whether Intel itself will release such cards as a standalone product for the end market. The showcased sample with an external water block and specific mounting hardware suggests it is targeted more at system integrators and server/workstation configurations than retail sales. The company has not yet disclosed details on whether the dual-chip version will become commercially available to the general public and in what form. ### Intel's Strategy: From Individual Cards to Server Racks Meanwhile, broader delivery strategies have come to light: Intel plans to sell its AI accelerators not only individually but also as part of complete turnkey systems. Such configurations are expected to combine up to eight cards in a single device, scaling up to 80 accelerators in full server racks. This highlights the company's intention to cover the entire spectrum of deployment scenarios—from compact workstations to large-scale data center platforms for training and running AI models. ### What This Means for the AI Accelerator Market The demonstration of the dual-chip configuration based on Crescent Island shows that Intel is developing its lineup not as a singular product, but as a family of scalable solutions sharing the Xe3P architecture. The ability to scale from a single chip to multi-card and multi-node system configurations enables the company to compete in the data center AI segment, where performance per accelerator is only part of the equation, with cluster-building capability being equally critical. A final verdict on the commercial success of the lineup will have to wait until Intel discloses pricing, delivery timelines, and mass-production terms. ## Fact-Check & Sources - [Это не что-то из стимпанка, а двухчиповый ИИ-ускоритель Intel. Компания раскрыла новые детали о поколении Crescent Island](https://www.ixbt.com/news/2026/10/16/eto-ne-cto-to-iz-stimpanka-a-dvuxcipovyi-ii-uskoritel-intel-kompaniia-raskryla-novye-detali-o-pokolenii-crescent-island.html) - Primary source: two GPUs with 32 Xe3P cores each, PLX interconnect, liquid cooling with a copper water block; sales plans for standalone and system integration up to 8 and 80 cards; uncertainty regarding retail release. ## FAQ ### What is the Intel dual-chip AI accelerator recently showcased? It is an adapter from the Crescent Island lineup built on two GPUs identical to the D270 components, where each chip contains 32 Xe3P cores connected via a PLX interface. ### How is this device cooled? Through a liquid cooling system featuring a massive copper water block with the Intel logo and a quick-disconnect fitting, as visible in published photographs. ### Will Intel sell these cards separately? Intel plans to sell accelerators both individually and as part of systems (up to 8 cards per device and up to 80 in server racks), though the company has not announced a retail release specifically for the dual-chip card.