Quick Verdict
The Xeon 6972P is a purpose-built data-center processor that trades single-thread speed and power envelope for massive parallelism and memory bandwidth, making it a strong fit for bandwidth-heavy HPC and AI workloads, particularly in dual-socket deployments where MRDIMMs can be fully utilized.
Overview
Launch
2024
Status
LaunchedGeneration
Xeon 6 (6th Gen Xeon Scalable)
Market
Data Center
A 96-core server processor from Intel’s Xeon 6900P series (Granite Rapids-AP) designed for dual-socket HPC, AI, and cloud platforms with 12 DDR5/MRDIMM channels, 96 PCIe 5.0 lanes, and Intel AMX for AI acceleration.
The Intel Xeon 6972P packs 96 Redwood Cove P-cores with Hyper-Threading for 192 threads, 480 MB of L3 cache, and 12 memory channels supporting DDR5-6400 and MRDIMM-8800. It operates at a 2.4 GHz base and up to 3.
9 GHz boost, with a 500 W TDP and dual-socket scalability via six 24 GT/s UPI links. The design integrates accelerators like Intel AMX, QAT, DLB, DSA, and IAA, plus security features including TDX, TME-MK, and SGX. Official guidance positions the 6900-series for high-performance, high-bandwidth cloud, HPC, and AI workloads.
Specifications
Performance
Geared toward server workloads (compiles, simulations, large-scale analytics) rather than desktop productivity.
96 cores and 192 threads enable dense virtualization; performance scales with software and storage configuration.
Not designed for gaming; this is a server-class processor without integrated graphics and with a power envelope inappropriate for desktop gaming systems.
At 500 W TDP, efficiency depends heavily on workload utilization and whether MRDIMMs are used to offset memory bottlenecks.
- •Server platform; not intended for gaming use.
- •No integrated graphics and requires server platform and cooling.
- •Intel AMX accelerates INT8/BF16 inference and some training workloads.
- •Large memory bandwidth with MRDIMMs benefits large model serving.
- •DLB and DSA can help with data movement and scheduling overhead.
Architecture
Intel 3
Process Node
Granite Rapids-AP
Codename
96C / 192T
Core Config
480 MB
L3 Cache
500 W
TDP
Architecture Overview
Granite Rapids-AP is Intel’s performance-core (P-core) variant in the Xeon 6 family, optimized for HPC, AI, and cloud workloads. It uses Redwood Cove P-cores with Hyper-Threading and a disaggregated multi-chiplet design connected via EMIB, supporting high core counts and 12-channel memory. Intel 3 manufacturing and up to 480 MB of shared L3 cache provide both compute density and large on-die data locality. The platform integrates dedicated accelerators for AI (AMX), crypto and compression (QAT), data movement (DSA), load balancing (DLB), and analytics (IAA).
CPU Design
96 P-cores (Redwood Cove) with Hyper-Threading (2-way SMT), base 2.4 GHz and up to 3.9 GHz single-core turbo. Cache hierarchy includes per-core L2 and a 480 MB shared L3. Intel Speed Select Technology - Performance Profile (SST-PP) allows a lower-power mode at 400 W with a 2.0 GHz base.
Memory Subsystem
12 memory channels with support for both DDR5-6400 RDIMMs and MRDIMM-8800, delivering significantly higher bandwidth than typical 8-channel server platforms. ECC is supported, and maximum memory size is up to 3 TB per socket (implementation dependent).
PCIe & I/O
96 PCIe 5.0 lanes directly from the CPU, suitable for multiple NVMe SSDs, DPUs/SmartNICs, and accelerators without chipset bottlenecks.
Overclocking
The 6972P does not have an unlocked multiplier; it operates within Intel’s specified TDP and turbo windows. Enthusiast-style ‘overclocking’ is not a supported use case for this server part.
- Higher core count (96 vs 64 in the prior high-end)
- More memory channels and support for MRDIMMs
- Increased PCIe lane count and updated interconnects
- Intel 3 process and updated Redwood Cove P-core microarchitecture
- New accelerators and expanded I/O (CXL 2.0 support)
Key Highlights
- 96 cores and 192 threads for high parallelism.
- 12-channel DDR5 and MRDIMM support for exceptional memory bandwidth.
- 96 PCIe 5.0 lanes for dense NVMe, accelerator, and NIC connectivity.
- Integrated AI accelerators (AMX), plus QAT, DLB, DSA, IAA for specialized tasks.
- Dual-socket scalability with UPI 2.0 for large NUMA domains.
- Strong enterprise security features (TDX, TME-MK, SGX, TXT, Boot Guard).
- High 500 W TDP requires robust server cooling and power infrastructure.
- Moderate base clock (2.4 GHz) is lower than many desktop/workstation parts.
- No integrated graphics; not suitable for non-server use cases.
- MRDIMMs may increase system cost and power compared to DDR5 RDIMMs.
- Platform lock-in to LGA7529-based 6900P infrastructure.
History
The Xeon 6972P belongs to Intel’s Granite Rapids-AP generation, introduced on September 24, 2024, as part of the Xeon 6 family rebranding. Granite Rapids splits Intel’s server lineup into performance-core (P-core) and efficiency-core (E-core) tracks, with the 6900P series targeting HPC, AI, and high-throughput cloud workloads. The 6972P’s 96 cores, 12-channel memory, and 96 PCIe 5.
0 lanes reflect Intel’s push to increase bandwidth alongside compute density, especially for data-intensive applications. Early adopters include OEM platforms such as HPE ProLiant Compute XD. Intel positions Granite Rapids as a response to growing demand for memory-bound and AI workloads, pairing Intel 3 with Redwood Cove cores and a disaggregated, multi-chiplet design to scale core counts and I/O.
In comparative demonstrations, Intel highlighted 96-core Granite Rapids-AP SKUs like the 6972P against prior-gen Xeon and AMD EPYC competitors, emphasizing memory bandwidth and AI inferencing performance. The 6972P’s feature set—MRDIMM support, Intel AMX, QAT/DLB/DSA/IAA accelerators, and robust security—reflects how server CPU design has evolved to integrate domain-specific accelerators alongside general-purpose cores, a trend that shapes the competitive landscape against AMD’s Zen 4/5-based EPYC parts.
Improvements over Previous Generation
- Higher core count (96 vs 64 in the prior high-end)
- More memory channels and support for MRDIMMs
- Increased PCIe lane count and updated interconnects
- Intel 3 process and updated Redwood Cove P-core microarchitecture
- New accelerators and expanded I/O (CXL 2.0 support)
Alternatives & Competitors
Should You Buy It?
Recommended for the right buyer
New dual-socket HPC or AI cluster deployments where high memory bandwidth and PCIe 5.0 I/O are critical; organizations already standardizing on Intel Xeon 6 server platforms.
Avoid if…
- You need a workstation or desktop CPU.
- You want to minimize power and cooling overhead per node.
- You prefer a single-socket platform with lower-cost CPUs.
Use Cases
Interesting Facts
Granite Rapids-AP launched on September 24, 2024, marking the 6th generation of Xeon Scalable (Xeon 6).
The 6972P pairs Intel’s Redwood Cove P-cores with a disaggregated design using EMIB links to connect compute and I/O tiles.
Intel’s product brief highlights 6900-series processors as a new class of server platform with higher TDPs, memory channels, and I/O lanes tailored to HPC and AI.
Intel lists both DDR5-6400 and MRDIMM-8800 support, with MRDIMMs designed to deliver more than 37% additional memory bandwidth.
The CPU includes multiple instance counts of accelerators: 4x QAT, 4x DLB, 4x DSA, and 4x IAA by default, reflecting Intel’s data-flow focus.
Security features go beyond encryption: Intel TDX provides hardware-based confidential computing and TME-MK supports multi-key total memory encryption.
The 6972P supports two performance profiles via Intel SST-PP: a 500 W profile at 2.4 GHz and a 400 W profile at 2.0 GHz.
With six 24 GT/s UPI links, dual-socket configurations provide substantial inter-socket bandwidth for scaling large workloads.
Granite Rapids’ P-core architecture (Redwood Cove) inherits and refines prior Golden/Raptor Cove ideas with larger L2 caches and higher clocks.
Intel’s press materials position Xeon 6 as a unified family with common software stacks across E-core and P-core models, simplifying deployment across cloud and edge.
People Also Ask
What socket does the Intel Xeon 6972P use?
It uses the FCLGA7529 socket, part of the Intel Xeon 6900P server platform.
Does the Xeon 6972P support DDR5?
Yes, it supports DDR5-6400 RDIMMs and also MRDIMM-8800 for higher bandwidth.
How many PCIe lanes does the Xeon 6972P have?
Up to 96 PCIe 5.0 lanes from the CPU, ideal for multiple NVMe SSDs, NICs, and accelerators.
What is the TDP of the Intel Xeon 6972P?
500 W TDP, with a 400 W SST-PP profile also supported at 2.0 GHz base.
Can the Xeon 6972P be used in a workstation?
Technically possible on compatible LGA7529 workstation boards, but it is primarily designed for server platforms with appropriate cooling and power.
Does the Xeon 6972P have integrated graphics?
No, it lacks integrated graphics and is intended to pair with discrete server GPUs or iGPUs if needed.
What is Intel AMX on the Xeon 6972P?
Intel Advanced Matrix Extensions accelerate matrix operations for AI inference (INT8/BF16) and certain training workloads.
Is the Xeon 6972P multiplier unlocked?
No, the 6972P is not multiplier-unlocked; it operates within Intel’s specified frequencies and turbo behavior.
What is MRDIMM and why does it matter for the 6972P?
MRDIMM (Multiplexed Rank DIMM) can increase effective memory bandwidth versus standard RDIMMs, benefiting bandwidth-heavy HPC and AI workloads.
How many sockets can the Xeon 6972P support?
Up to two sockets per node, with six 24 GT/s UPI links for inter-socket communication.
Frequently Asked Questions
Is the Xeon 6972P suitable for gaming?
No, it is a server processor with high TDP and no integrated graphics; it is not designed or recommended for gaming.
What is the recommended customer price for the Xeon 6972P?
Intel’s listed RCP is $11,446, though actual pricing may vary by region and volume.
Does the Xeon 6972P support CXL?
The Xeon 6900P platform supports CXL 2.0; CXL devices can be used if the server platform implements them.
What security features does the Xeon 6972P include?
It supports Intel TDX, TME-MK, SGX, TXT, Boot Guard, Control-Flow Enforcement Technology, and extensive RAS capabilities.
What are the ordering codes for the Xeon 6972P?
Intel lists at least two tray ordering codes: PK8072006022900 and PK8072006129700.
Can I underclock the Xeon 6972P to save power?
Intel Speed Select Technology - Performance Profile includes a 400 W mode at 2.0 GHz base; further power policies are typically managed via BIOS/management firmware.
What workloads benefit most from the 6972P’s MRDIMM support?
Memory-bandwidth-bound HPC simulations, AI training and inference, and large analytics workloads tend to benefit most.
Does the Xeon 6972P support AVX-512?
Yes, it supports AVX-512 extensions, including AVX-512 FMA units relevant to HPC and AI.
Is the Xeon 6972P compatible with previous-generation Xeon servers?
No, it requires the LGA7529-based 6900P platform; it is not backward compatible with older Xeon sockets.
What is the ‘High Priority Cores’ feature on the 6972P?
Intel ARK shows 36 high-priority cores at 2.6 GHz and 60 low-priority cores at 2.2 GHz via Speed Select Technology, enabling differentiated QoS in multi-tenant environments.