Overview
- Intel has started putting its D-Series flagship, the D270 (formerly Crescent Island), into customer hands for testing as the company moves from previews to practical evaluation.
- The D270 is built on Intel’s Xe3P architecture with 32 Xe3P cores, 256 XMX engines, about 1.3 PFLOPs FP8 peak compute, a 32 MB unified L2 cache, and a 350W thermal design power.
- Instead of HBM, Intel chose LPDDR5X memory with a reference 160 GB configuration expandable to 480 GB and an estimated peak bandwidth near 1.5 TB/s, trading peak throughput for much higher capacity and lower cost.
- The card is an air-cooled PCIe Gen5 form factor meant to fit standard servers and to be paired with Intel Xeon CPUs and OEM systems from partners such as ASUS, Dell, and HPE to ease deployment and lower integration cost.
- Customer feedback from the sampling round will determine whether enterprises adopt the D270 as a cost-effective inference and agentic-workload option or view it as niche because some AI workloads need the higher bandwidth HBM provides.