Quick Verdict
A strong 16-core Xeon 6 SKU for customers who value high per-thread clocks, large DDR5 bandwidth, and extensive I/O over maximum core density, with excellent security and acceleration features for modern data centers.
Overview
Launch
2025
Status
LaunchedGeneration
6th Gen Intel Xeon Scalable (Xeon 6 Granite Rapids-SP)
Market
1S/2S Server & Data Center
The Intel Xeon 6725P is a 16-core, 32-thread server processor based on the Granite Rapids-SP (Redwood Cove P-core) architecture, built on Intel’s Intel 3 process and targeting high-frequency, memory-intensive workloads in virtualized, database, and AI inference environments.
With 16 Redwood Cove P-cores, 32 threads, 3.7 GHz base and 4.8 GHz max turbo, 192 MB of L3 cache, and 8-channel DDR5-6400, the Xeon 6725P is tailored for virtualized infrastructure, in-memory databases, and AI inference where memory bandwidth and I/O matter as much as raw core count.
It supports up to 4 TB of RAM per socket, 88 PCIe 5.0 lanes, and Intel’s latest data center accelerators (AMX, QAT, DLB, DSA, IAA), making it a balanced choice for modern enterprise and cloud deployments.
Specifications
Performance
No standardized benchmark scores available; expected to perform very well in server-side productivity and database workloads thanks to 16 high-frequency cores and 8-channel DDR5-6400.
No official scores; strong virtualization fit due to 16 cores/32 threads, large memory capacity, and Intel VT-x/VT-d/EPT support.
Not applicable; this is a server CPU without integrated graphics and is not targeted or benchmarked for gaming workloads.
235 W TDP for 16 cores is typical for performance-optimized Xeon 6 SKUs; efficiency is competitive within its class but not a low-power part.
- •Server processor without integrated graphics
- •Not validated or marketed for gaming
- •Gaming benchmarks are not meaningful for this segment
- •AMX instructions accelerate matrix workloads for inference
- •Suited to CPU-based AI inferencing and data preprocessing
- •Not a replacement for dedicated GPUs or accelerators for large models
Architecture
Intel 3
Process Node
Granite Rapids-SP
Codename
16C / 32T
Core Config
192 MB
L3 Cache
235 W
TDP
Architecture Overview
The Xeon 6725P belongs to Intel’s Granite Rapids-SP family, which uses a chiplet design with multiple compute dies built on the Intel 3 process and separate I/O dies. The compute dies house Redwood Cove P-cores and DDR5 memory controllers, while the I/O dies handle PCIe, UPI, and accelerators, connected via EMIB and a mesh interconnect that behaves like a monolithic die to software.
CPU Design
Each Redwood Cove P-core has a 2 MB private L2 cache and shares a slice of the distributed L3 cache; with 16 cores active, the 6725P exposes 192 MB of L3 and 32 MB of aggregate L2. Hyper-Threading provides two threads per core, and the mesh interconnect scales well across the chiplet topology.
Memory Subsystem
Eight DDR5-6400 memory channels are spread across the compute dies, delivering high aggregate bandwidth and supporting up to 4 TB of RAM per socket. Memory controllers are placed at the edges of the compute dies to optimize latencies, and the platform supports sub-NUMA clustering for locality-aware software.
PCIe & I/O
Up to 88 PCIe 5.0 lanes originate from the I/O dies, enabling multi-GPU, high-NVMe, and smart NIC configurations. The lane count is sufficient for dual-socket systems with substantial accelerator and storage bandwidth needs.
Overclocking
The multiplier is locked; Xeon 6 server SKUs are not designed for enthusiast overclocking. Instead, Intel offers Speed Select Technology (SST) to fine-tune core counts and turbo budgets within the TDP envelope.
- Higher max turbo (4.8 vs 4.3 GHz) for better single-thread performance
- Much larger L3 cache (192 MB vs 72 MB) for memory-bound workloads
- Same platform features (DDR5-6400, 88 PCIe 5.0 lanes, Intel 3) with more cache
Key Highlights
- 16 high-frequency P-cores with up to 4.8 GHz turbo
- 192 MB L3 cache and 8-channel DDR5-6400 for memory-intensive workloads
- 88 PCIe 5.0 lanes for flexible I/O and accelerator configurations
- Intel 3 process and chiplet architecture for scalable performance
- Strong security and acceleration: AMX, QAT, DLB, SGX, TDX, MK-TME
- Higher TDP (235 W) than lower-core Granite Rapids-SP SKUs
- No integrated graphics (typical for server CPUs)
- Locked multiplier limits overclocking headroom
- Premium price segment typical of Xeon 6 performance-core parts
- Core count modest vs some competing EPYC 9005 SKUs at similar price
History
The Xeon 6725P emerged as part of Intel’s Granite Rapids-SP rollout, which marked the shift from the mixed P+E core approach of earlier Xeon Scalable generations to a dedicated P-core lineup for performance-optimized servers. Granite Rapids-SP introduced the Redwood Cove core architecture with 2 MB L2 caches and a chiplet strategy using Intel 3 compute dies and separate I/O dies, connected via EMIB. The 6725P sits in the mid-range of the 6700P stack, offering 16 cores with a large 192 MB L3 cache and high DDR5-6400 bandwidth, designed to replace earlier Xeon Gold/Platinum SKUs in virtualization and database roles.
It reflects Intel’s focus on maintaining strong per-thread performance and memory bandwidth while scaling core counts and I/O capabilities to compete with high-core-count Arm and AMD alternatives in modern data centers.
Improvements over Previous Generation
- Higher max turbo (4.8 vs 4.3 GHz) for better single-thread performance
- Much larger L3 cache (192 MB vs 72 MB) for memory-bound workloads
- Same platform features (DDR5-6400, 88 PCIe 5.0 lanes, Intel 3) with more cache
Alternatives & Competitors
Should You Buy It?
Recommended for the right buyer
Upgrading or building 1S/2S servers where you need strong per-thread performance, large DDR5 bandwidth, and 88 PCIe 5.0 lanes more than sheer core count.
Avoid if…
- You need maximum core density (consider 32+ core EPYC or Xeon 6730P instead)
- Your workload is heavily GPU-accelerated and CPU is not the bottleneck
- You prioritize lowest possible power per core
Use Cases
Interesting Facts
Granite Rapids-SP is Intel’s first Xeon Scalable family to use the Intel 3 process for compute dies, with I/O dies on a more mature node.
Each Redwood Cove P-core in Granite Rapids has 2 MB of dedicated L2 cache, a big increase over the 1.25 MB L2 of earlier Golden Cove designs.
Xeon 6725P supports 2S configurations with four UPI links at 24 GT/s, enabling coherent inter-socket communication.
Intel includes AMX, QAT, DLB, DSA, and IAA accelerators by default, making these SKUs particularly attractive for crypto, analytics, and load-balanced workloads.
The 192 MB L3 cache on the 16-core 6725P is the same cache size per socket as some higher-core Granite Rapids parts, just with fewer cores enabled.
People Also Ask
Is Intel Xeon 6725P good for virtualization?
Yes. With 16 cores/32 threads, 8-channel DDR5, and VT-x/VT-d support, it is well suited for hosting multiple VMs, especially when each VM benefits from high per-thread performance and memory bandwidth.
How much memory does the Xeon 6725P support?
Up to 4 TB of DDR5-6400 memory across 8 channels, depending on DIMM capacity and population rules.
How many PCIe lanes does the Xeon 6725P have?
Up to 88 PCIe 5.0 lanes from the CPU, suitable for multiple GPUs, NVMe drives, and smart NICs.
What is the difference between Xeon 6725P and 6724P?
The 6725P has a higher max turbo (4.8 vs 4.3 GHz) and much larger L3 cache (192 MB vs 72 MB), while both are 16-core Granite Rapids-SP parts with the same platform features.
Does Xeon 6725P have integrated graphics?
No. It is a server processor without integrated graphics; a discrete GPU or ASPEED/BMC is used for display if needed.
Can you overclock Xeon 6725P?
No. The multiplier is locked; Xeon 6 server SKUs are not designed for overclocking. You can use Speed Select Technology to adjust power and turbo profiles instead.
What process node is Xeon 6725P built on?
It is manufactured on Intel’s Intel 3 process (formerly ~5 nm EUV class), as listed on Intel’s regional ARK pages for this SKU.
Is Xeon 6725P suitable for gaming?
It can run games, but it is not optimized or benchmarked for gaming; there are cheaper and more appropriate desktop CPUs for that use case.
What accelerators are included in Xeon 6725P?
Intel QAT, DLB, DSA, and IAA accelerators are included by default, along with AMX for AI and DL Boost for vector instructions.
What security features does Xeon 6725P support?
It supports SGX, TDX, total memory encryption (MK-TME), CET, and Intel Boot Guard, providing a strong foundation for confidential computing and secure boot.
Frequently Asked Questions
What socket does the Intel Xeon 6725P use?
It uses the FCLGA4710 socket, part of the LGA4710 platform for Granite Rapids-SP processors.
How many UPI links does the Xeon 6725P support?
It supports up to 4 UPI links at 24 GT/s for coherent inter-socket communication in dual-socket systems.
What is the default TDP of the Xeon 6725P?
The default and maximum TDP is 235 W, as specified by Intel.
Does the Xeon 6725P support ECC memory?
Yes, it supports DDR5 ECC memory, which is standard for Xeon server platforms.
What is the max turbo frequency of the Xeon 6725P?
The max turbo frequency is 4.8 GHz, with an all-core turbo of 3.9 GHz under heavy multi-threaded loads.
Can the Xeon 6725P be used in a workstation?
It is primarily a server CPU; for workstations, Intel offers Xeon 600 Granite Rapids-WS SKUs with similar cores but optimized for single-socket workstation platforms.
What is the L3 cache size of the Xeon 6725P?
It has 192 MB of shared L3 cache across the 16 cores.
Does the Xeon 6725P support AVX-512 and AMX?
Yes, it supports AVX-512 and Intel AMX instructions, which accelerate HPC and AI workloads.
Is the Xeon 6725P unlocked for overclocking?
No, the multiplier is locked; it is not designed for overclocking.
What generation is the Xeon 6725P?
It is part of the 6th generation Intel Xeon Scalable family, codenamed Granite Rapids-SP.