Intel Xeon 6900P Series (Granite Rapids-AP) Processor

The Intel Xeon 6900P Series, code-named Granite Rapids-AP, represents the apex of Intel's server processor engineering, integrating up to 128 high-performance P-cores (Performance-cores) on a single socket built on Intel's Intel 3 process node. The architecture employs a multi-tile design combining CPU tiles with a base tile, enabling unprecedented core counts while maintaining cache coherency and memory bandwidth. Each processor supports up to 504 MB of total Intel Smart Cache (L2 + L3), ensuring that latency-sensitive AI inference pipelines and numerical computing kernels have rapid access to working data sets without bottlenecking on memory subsystem latency.

A defining capability of the Xeon 6900P is its 12-channel DDR5 memory architecture, the widest memory interface ever deployed in a mainstream x86 server processor, supporting DDR5-6400 speeds and the revolutionary MRDIMM (Multiplexed Rank DIMM) standard at up to 8800 MT/s per channel. This yields aggregate memory bandwidth exceeding 1 TB/s per socket, critical for feeding the compute engines of large language model inference, seismic analysis, and computational fluid dynamics simulations. The platform supports up to 6 TB of DDR5 per socket, accommodating the largest in-memory databases and analytics workloads without external memory expansion.

Built into each Xeon 6900P are 64 enhanced Intel Advanced Matrix Extensions (AMX) units with BF16 and INT8 support, delivering multiple teraFLOPS of AI compute natively in the CPU without requiring a dedicated accelerator for many inference tasks. The processor integrates up to 136 PCIe 5.0 lanes per socket, providing ample bandwidth for high-speed NVMe storage arrays, multi-port 400GbE networking, and GPU co-processing cards. The platform's support for Compute Express Link (CXL) 2.0 enables memory pooling and expansion architectures that future-proof large-scale data center deployments requiring flexible memory tiering.

Ideal for

• Large-scale AI inference serving for transformer-based language and vision models requiring sustained high-throughput token generation without discrete GPU dependencies
• High-performance computing clusters running computational fluid dynamics, weather modeling, and finite element analysis simulations that demand maximum floating-point throughput per socket
• In-memory relational and NoSQL databases such as SAP HANA and Redis Enterprise operating on datasets exceeding 3 TB per node with microsecond query latency requirements
• Real-time financial risk analytics and Monte Carlo simulation engines processing millions of market scenarios per second across dense core counts
• Memory-intensive genomics and bioinformatics pipelines performing genome assembly, variant calling, and protein structure analysis on large reference datasets
• Hyperconverged infrastructure nodes combining high-core-count compute with CXL memory expansion for dynamic workload consolidation across virtualized enterprise environments

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.