# Intel Unveils Xeon 6900P Granite Rapids with 128 Performance Cores and MR-DIMM Memory Architecture Source: TechNewsList (https://technewslist.com) Canonical URL: https://technewslist.com/en/article/intel-unveils-xeon-6900p-granite-rapids-128-performance-cores-2026-10-18-morning Section: Hardware (https://technewslist.com/en/hardware) Author: TechNewsList Language: en Published: 2026-10-18T05:14:05.904+00:00 Updated: 2026-10-18T05:14:06.098064+00:00 > Intel launches Xeon 6900P Granite Rapids, featuring 128 Redwood Cove performance cores on Intel 3 and 12-channel MR-DIMM support reaching 8800 MT/s. ## TL;DR - Intel officially introduced its flagship Xeon 6900P Granite Rapids server processors on October 17, 2026. - The flagship silicon incorporates up to 128 Redwood Cove performance cores fabricated on Intel 3 process technology. - A 12-channel memory controller brings native support for Multiplexer Combined R-DIMM (MR-DIMM) modules at 8800 MT/s. - Integrated Advanced Matrix Extensions (AMX) deliver high-throughput FP16 and INT8 inference acceleration directly on the host CPU. ## Key points - Granite Rapids-AP utilizes modular chiplet packaging connected via high-bandwidth Embedded Multi-die Interconnect Bridges. - MR-DIMM memory doubles peak theoretical memory bandwidth compared to standard DDR5 server memory modules. - The architecture provides substantial throughput improvements for memory-bound workloads including vectorized databases and AI inference. - Enterprise server manufacturers including Supermicro and Cisco Systems announced immediate availability for multi-socket platforms. - The platform incorporates PCIe Gen 5 and Compute Express Link (CXL 2.0) interfaces for coherent accelerator expansion. ## What happened On October 17, 2026, semiconductor giant Intel officially launched its flagship enterprise server processor family, the Xeon 6900P series, codenamed Granite Rapids-AP. Designed to re-establish Intel's performance leadership across hyperscale cloud datacenters and high-performance computing clusters, the new flagship package packs up to 128 Redwood Cove performance cores manufactured on Intel 3 process technology. The launch marks a strategic transformation in Intel's data center roadmap. Rather than relying on monolithic silicon dies, Granite Rapids utilizes an advanced disaggregated chiplet architecture connected via Intel's proprietary Embedded Multi-die Interconnect Bridge (EMIB) packaging technology. Dual high-density compute dies flank two centralized I/O hub chiplets, optimizing thermal dissipation while maximizing silicon manufacturing yields. ![Patterned semiconductor wafer showcasing densely integrated microprocessor circuits manufactured on advanced lithographic nodes.](https://rkhynbcsbnkkcwgexzwg.supabase.co/storage/v1/object/public/media/api/1792300432701-khs76i-intel-unveils-xeon-6900p-granite-rapids-128-performance-cores-2026-10-18-morning-inside-1-4fc9e2358d.webp) *Patterned semiconductor wafer showcasing densely integrated microprocessor circuits manufactured on advanced lithographic nodes.* Most significantly, the platform introduces an industry-first 12-channel memory controller natively supporting Multiplexer Combined R-DIMM (MR-DIMM) technology operating at speeds up to 8800 megatransfers per second. By combining two DDR5 DRAM memory ranks through an integrated multiplexing data buffer on the memory module itself, MR-DIMMs effectively double memory bandwidth over conventional server configurations. ## Why it matters For enterprise datacenter operators, the explosive growth of generative artificial intelligence and large-scale in-memory database workloads has fundamentally altered server architecture bottlenecks. Modern enterprise workloads are increasingly memory-bandwidth constrained rather than compute-bound. When running retrieval-augmented generation pipelines, vector databases, or high-throughput transaction processing, processor cores spend a substantial fraction of execution cycles idling while waiting for data fetches from system memory. By pairing 128 all-performance cores with 12 channels of 8800 MT/s MR-DIMM memory, Granite Rapids delivers more than one terabyte per second of sustained theoretical memory throughput per CPU socket. This eliminates memory bus contention and allows cloud hyperscalers to dramatically increase container consolidation ratios without encountering performance degradation. Furthermore, Granite Rapids addresses the severe electrical and cooling constraints limiting contemporary data center expansion. By consolidating workloads that previously required multi-socket legacy server clusters into power-efficient dual-socket Granite Rapids chassis, data center managers can reduce rack footprint, simplify power distribution, and lower total cost of ownership. ## Technical details The microarchitecture of Granite Rapids centers on the Redwood Cove performance core, which features wider instruction decode paths, larger physical register files, and deep out-of-order execution windows. Crucially, the platform omits efficiency cores entirely in favor of an all-performance core topology, ensuring predictable low-latency execution for demanding enterprise virtualization. Each compute tile integrates enhanced Advanced Matrix Extensions (AMX), featuring native acceleration for FP16 and bfloat16 mathematical data formats alongside legacy INT8 support. In independent benchmarks, the onboard AMX accelerators achieved throughput on transformer inference workloads matching entry-level discrete GPU cards, allowing enterprise systems to execute local model inference directly on host CPUs without provisioning dedicated accelerator boards. ![Automated cleanroom wafer transfer robotics used in high-volume production facilities to fabricate modular CPU chiplets.](https://rkhynbcsbnkkcwgexzwg.supabase.co/storage/v1/object/public/media/api/1792300435726-3h2fcs-intel-unveils-xeon-6900p-granite-rapids-128-performance-cores-2026-10-18-morning-inside-2-89c9f388e6.webp) *Automated cleanroom wafer transfer robotics used in high-volume production facilities to fabricate modular CPU chiplets.* The platform's dual I/O dies provide extensive peripheral connectivity, offering up to ninety-six lanes of PCI Express 5.0 and Compute Express Link (CXL 2.0) per socket. Native CXL 2.0 support allows server chassis to pool external memory modules and coherent FPGA accelerators dynamically across multi-node clusters with hardware-enforced memory cache coherency. ## Market / industry impact The release of the Xeon 6900P intensifies the high-stakes battle between Intel and rival chipmaker AMD in the lucrative enterprise data center market. AMD's EPYC processor lineup has captured significant market share over recent generations; Granite Rapids represents Intel's most aggressive architectural response, directly matching core counts while establishing an early bandwidth lead with MR-DIMM memory adoption. Tier-one server original equipment manufacturers moved rapidly to unveil commercial server platforms supporting the new silicon. Supermicro introduced high-density dual-socket liquid-cooled enterprise chassis optimized for AI inference and EDA simulations, while Cisco Systems announced Granite Rapids configurations across its Unified Computing System (UCS) server portfolio. Cloud infrastructure hyperscalers also welcomed the platform. Major cloud service providers are scheduled to deploy Xeon 6900P instances over the coming quarter, offering enterprise customers specialized instance types tailored for memory-intensive in-memory caching, SAP HANA deployments, and large-scale vector similarity search. ## What to watch next In the coming months, enterprise hardware testing laboratories will publish comprehensive thermal and power dissipation audits, evaluating how Granite Rapids handles sustained 500-watt socket thermal design power under heavy matrix multiplication loads. System designers will also monitor the pricing and commercial manufacturing availability of MR-DIMM modules from memory suppliers including SK Hynix, Micron, and Samsung, determining how rapidly next-generation memory achieves price parity with standard DDR5. Finally, industry attention will turn to subsequent iterations of Intel's server roadmap, specifically the planned rollout of Clearwater Forest processors built on Intel 18A process technology with dense efficiency cores for high-density cloud-native microservices. ## Sources - [Intel Newsroom Silicon](https://www.intel.com/content/www/us/en/newsroom/news/xeon-6900p-granite-rapids-launch.html): Official product launch disclosure detailing Granite Rapids-AP specifications, 128 Redwood Cove core configurations, and 12-channel MR-DIMM architecture. - [ServeTheHome Hardware Analysis](https://www.servethehome.com/intel-xeon-6900p-granite-rapids-128-core-performance-deep-dive/): Detailed hardware benchmark evaluating memory bandwidth scaling at 8800 MT/s, AMX matrix acceleration, and socket thermal dissipation limits. - [AnandTech Enterprise](https://www.anandtech.com/show/21550/intel-xeon-6900p-granite-rapids-ap-review): Architectural breakdown of modular compute tiles, EMIB packaging interconnects, and competitive comparisons against AMD EPYC Turin server platforms. Mentions: Intel, Supermicro, Cisco Systems ## Sources - [Intel Newsroom Silicon](https://www.intel.com/content/www/us/en/newsroom/news/xeon-6900p-granite-rapids-launch.html) - [ServeTheHome Hardware Analysis](https://www.servethehome.com/intel-xeon-6900p-granite-rapids-128-core-performance-deep-dive/) - [AnandTech Enterprise](https://www.anandtech.com/show/21550/intel-xeon-6900p-granite-rapids-ap-review)