Quick Verdict
A high‑core‑count, I/O‑rich server CPU ideal for single‑socket AI and HPC systems, though its 350 W TDP and premium price demand careful platform and cooling design.
Overview
Launch
2025
Status
LaunchedGeneration
Intel Xeon 6 (Granite Rapids-SP P-core)
Market
Server / Data Center / AI / HPC
The Intel Xeon 6774P is a 64-core, 128-thread server processor based on the Granite Rapids-SP P-core architecture, targeting single-socket AI, HPC, and data‑center platforms with 8-channel DDR5/MRDIMM support and 136 PCIe 5.0 lanes.
Intel Xeon 6774P combines 64 P‑cores, 128 threads, 336 MB of L3 cache, and 136 PCIe 5.0 lanes in a 350 W envelope, making it a strong fit for GPU‑centric AI factories and memory‑intensive server workloads that benefit from a single NUMA domain. It supports DDR5‑6400 and MRDIMM‑8800, offering very high memory bandwidth, and integrates Intel AMX and other accelerators for AI and crypto.
Real‑world performance will depend heavily on platform tuning and cooling, but for single‑socket servers that need maximum I/O and memory bandwidth, it is a compelling option despite its high power draw and cost.
Specifications
Performance
Very strong multi‑threaded performance for server workloads, especially when leveraging AMX, QAT, and high memory bandwidth.
Excellent for consolidation of VMs and containers, though 350 W TDP requires robust cooling and power planning.
Not targeted at gaming; high latency and modest per‑core performance make it unattractive versus desktop or workstation CPUs for gaming use cases.
Performance per watt is reasonable at full load, but absolute power is high; efficiency advantages depend heavily on workload optimization.
- •Not designed for gaming; low single‑thread optimization vs desktop CPUs.
- •High latency mesh and server‑tuned memory timings hurt game responsiveness.
- •Only consider if server is also used for light gaming on the side.
- •Intel AMX provides dedicated INT8/BF16/FP16 matrix acceleration per core.
- •Well‑suited to CPU‑based inference for LLMs, vision transformers, and recommendation models.
- •Best when paired with GPUs for large‑scale training, but can handle moderate inference workloads alone.
Architecture
Intel 3 (~7 nm class)
Process Node
Granite Rapids-SP
Codename
64C / 128T
Core Config
336 MB
L3 Cache
350 W
TDP
Architecture Overview
Granite Rapids-SP is the P‑core only, chiplet‑based server architecture in Intel’s Xeon 6 family, using Intel 3 compute dies for cores and memory controllers and separate I/O dies for accelerators and PCIe.
CPU Design
64 P‑cores with Hyper‑Threading yield 128 threads; each core has 2 MB L2 cache and shares a 336 MB L3 cache across the mesh. Intel AMX units in each core accelerate matrix operations for AI workloads.
Memory Subsystem
Eight DDR5/MRDIMM memory controllers support DDR5‑6400 and MRDIMM‑8800, providing very high bandwidth and up to 4 TB capacity per socket, with ECC and total memory encryption support.
PCIe & I/O
136 PCIe 5.0 lanes in single‑socket R1S mode, derived from repurposing UPI links, making it attractive for GPU‑centric AI systems and high‑density NVMe storage without multi‑socket interconnect overhead.
Overclocking
The multiplier is locked; frequency is managed via Turbo Boost, Priority Core Turbo, and Intel Speed Select Technology rather than user overclocking.
- Higher core count (64 vs 60) and much larger L3 cache (336 MB vs ~112.5 MB).
- Transition from DDR4/DDR5 to DDR5‑only with MRDIMM support and higher speeds.
- Integrated AMX, QAT, DLB, DSA, IAA accelerators vs more limited offload in prior generations.
- Shift from Intel 7 to Intel 3 process for compute dies, improving density and performance per watt.
Key Highlights
- 64 P‑cores with AMX for strong AI and HPC performance.
- 136 PCIe 5.0 lanes in single‑socket R1S mode for GPU and NVMe expansion.
- 8‑channel DDR5/MRDIMM with up to 8800 MT/s speed and 4 TB capacity.
- Large 336 MB L3 cache and 128 MB L2 cache reduce memory bottlenecks.
- Rich set of integrated accelerators (QAT, DLB, DSA, IAA) and RAS features.
- Well‑suited to single‑NUMA‑domain designs, reducing software complexity.
- High 350 W TDP requires robust cooling and power delivery.
- Premium price point (Intel RCP ~$7,571) limits use to high‑end deployments.
- Locked multiplier and server‑oriented turbo behavior limit enthusiast tuning.
- No integrated graphics; not suitable for headless workstation or desktop use.
- Platform and motherboard ecosystem is still maturing compared to older Xeon generations.
History
The Intel Xeon 6774P emerged from Intel’s effort to regain leadership in the server CPU market with the Xeon 6 family, codenamed Granite Rapids. After AMD EPYC gained significant share with high core counts and advanced I/O, Intel responded with a chiplet‑based P‑core architecture that splits compute and I/O dies, enabling up to 128 P‑cores in the Xeon 6900P series and scaled‑down 64‑core parts like the 6774P in the 6700P line. Launched in early 2025, the 6700P series targets single‑socket and small‑scale multi‑socket servers with a focus on AI, HPC, and data‑center workloads, introducing features such as the R1S single‑socket mode with 136 PCIe 5.
0 lanes and Priority Core Turbo to improve performance for latency‑sensitive tasks. The 6774P sits near the top of the 6700P stack, offering 64 P‑cores, 336 MB of L3 cache, and 8‑channel DDR5/MRDIMM support, making it a key building block for GPU‑centric AI factories and modern HPC clusters.
Improvements over Previous Generation
- Higher core count (64 vs 60) and much larger L3 cache (336 MB vs ~112.5 MB).
- Transition from DDR4/DDR5 to DDR5‑only with MRDIMM support and higher speeds.
- Integrated AMX, QAT, DLB, DSA, IAA accelerators vs more limited offload in prior generations.
- Shift from Intel 7 to Intel 3 process for compute dies, improving density and performance per watt.
Alternatives & Competitors
Should You Buy It?
Recommended for the right buyer
Single‑socket AI factories, HPC servers, and in‑memory database appliances that can leverage 136 PCIe 5.0 lanes and 8‑channel DDR5/MRDIMM bandwidth.
Avoid if…
- Building a general‑purpose office or home server where 350 W TDP is overkill.
- Prioritizing maximum core density above all else (AMD EPYC 9005 offers higher thread counts).
- Needing multi‑socket scale‑up rather than single‑socket scale‑out.
- Budget‑sensitive builds where platform cost and cooling are hard to justify.
Use Cases
Interesting Facts
Granite Rapids-SP is Intel’s first Xeon 6 P‑core‑only server family, complementing the E‑core‑only Sierra Forest line.
The 136 PCIe 5.0 lane R1S mode repurposes UPI links to turn the CPU into a single‑socket I/O powerhouse for GPU‑centric AI systems.
Xeon 6 compute dies use the Intel 3 process, while I/O dies are on a more mature node, similar to chiplet strategies from AMD.
Priority Core Turbo on 6774P allows up to 8 cores to reach 4.6 GHz, a uncommonly high single‑core boost for a 64‑core server CPU.
Intel AMX on Granite Rapids-SP supports INT8, BF16 and FP16, bridging the gap between classic AVX‑512 and dedicated AI accelerators.
HPE and other OEMs position 6774P‑based servers as building blocks for NVIDIA AI factories and similar GPU‑heavy reference architectures.
The 336 MB L3 cache is shared across all 64 cores, giving a much larger cache per core than earlier Xeon Scalable generations.
Xeon 6700P SKUs can be partitioned into high‑ and low‑priority cores via Intel Speed Select Technology, trading per‑core performance for power efficiency.
Red Hat reports significant HPC and AI speedups on Xeon 6 vs 5th Gen Xeon, especially when using AMX and newer RHEL kernels.
Despite being a server CPU, some early adopters consider Xeon 6774P for high‑end workstations with PCIe/IO requirements that exceed desktop platforms.
People Also Ask
Is the Intel Xeon 6774P good for AI workloads?
Yes. It includes Intel AMX for INT8/BF16/FP16 matrix acceleration and is designed specifically for AI inference and training, especially in GPU‑centric single‑socket servers.
How many PCIe lanes does the Intel Xeon 6774P have?
It provides up to 136 PCIe 5.0 lanes in single‑socket R1S mode, which is among the highest PCIe lane counts from a single Xeon CPU.
Does the Intel Xeon 6774P support DDR5?
Yes. It supports DDR5‑6400 and MRDIMM‑8800 memory with eight channels, enabling very high memory bandwidth and capacities up to 4 TB.
What socket does the Intel Xeon 6774P use?
It uses the FCLGA4710 (LGA4710) socket, also referred to as Socket E2 in Intel’s Xeon 6 documentation.
Is the Intel Xeon 6774P unlocked for overclocking?
No. The multiplier is locked; frequency control is via Turbo Boost, Priority Core Turbo, and Intel Speed Select Technology rather than manual overclocking.
What is the TDP of the Intel Xeon 6774P?
Intel specifies a 350 W TDP for the Xeon 6774P, requiring robust cooling and power delivery in the server platform.
Can the Intel Xeon 6774P be used in multi‑socket systems?
It is primarily intended for single‑socket platforms; multi‑socket configurations typically use Xeon 6900P series SKUs with UPI links instead of the R1S PCIe lane configuration.
Does the Intel Xeon 6774P have integrated graphics?
No. It has no integrated GPU and is designed for headless server and appliance use cases.
What is Priority Core Turbo on the Intel Xeon 6774P?
Priority Core Turbo allows up to 8 cores to operate at a higher boost frequency (up to 4.6 GHz on this SKU), improving performance for latency‑sensitive or lightly‑threaded workloads.
Is the Intel Xeon 6774P better than AMD EPYC 9654 for AI?
It depends. Xeon 6774P offers higher per‑socket PCIe lane count and AMX acceleration, while EPYC 9654 provides more cores and threads; the best choice depends on whether your AI workload favors I/O and matrix math or raw thread count.
Frequently Asked Questions
What generation is the Intel Xeon 6774P?
It is part of the Intel Xeon 6 family, specifically the Granite Rapids-SP P‑core generation (6700P series).
What process node is the Intel Xeon 6774P built on?
Its compute dies use the Intel 3 process (~7 nm class), while I/O dies use a more mature node for cost and yield optimization.
How much L3 cache does the Intel Xeon 6774P have?
It has 336 MB of shared L3 cache across all 64 cores.
What memory speeds does the Intel Xeon 6774P support?
It supports DDR5‑6400 and MRDIMM‑8800 memory speeds, with a maximum of 8000 MT/s as specified by Intel.
What is the maximum memory capacity supported by the Intel Xeon 6774P?
It supports up to 4 TB of memory per socket, depending on DIMM type and density.
Does the Intel Xeon 6774P support ECC memory?
Yes, ECC memory is supported and is standard for server platforms using this CPU.
What accelerators are integrated into the Intel Xeon 6774P?
It includes Intel QAT, DLB, DSA, IAA, and AMX, plus support for Intel TDX, SGX, and total memory encryption for security.
Is the Intel Xeon 6774P suitable for gaming?
No. It is a server CPU optimized for AI, HPC, and I/O‑heavy workloads; gaming performance is poor compared to desktop or workstation CPUs.
What is the recommended customer price for the Intel Xeon 6774P?
Intel’s recommended customer price is approximately $7,571 for the tray CPU, though actual pricing varies by reseller and volume.
What operating systems support the Intel Xeon 6774P?
It is supported on major server OSs including Windows Server, major Linux distributions (RHEL, Ubuntu, SUSE), and VMware ESXi, with optimized kernels and drivers available for Xeon 6 features.