Launched6th Gen Intel Xeon Scalable (Xeon 6 P-Cores)

Intel · Xeon 6700P Series

Intel Xeon 6776P

64 P‑cores, 128 threads, 336 MB L3, 88 PCIe 5.0 lanes and 8‑channel DDR5/MRDIMM for GPU‑dense AI servers.

AI Inference & Training Host NodesHPC ClustersIn‑Memory DatabasesVirtualized InfrastructureGPU‑Accelerated Analytics

Cores / Threads

64/ 128

Base / Boost

2.3/ 3.9 GHz

PCIe Lanes

88

L2 Cache

128MB

L3 Cache

336MB

TDP

350W

Socket

FCLGA4710

Verdict

8.7/ 10

87

Quick Verdict

A high‑core‑count, cache‑rich server CPU tailored for GPU‑accelerated AI and HPC platforms, offering excellent memory bandwidth and I/O, but with a 350 W TDP and premium pricing that makes sense primarily in dense multi‑GPU servers where its features are fully utilized.

Best for:AI Inference & Training Host NodesHPC ClustersIn‑Memory DatabasesVirtualized InfrastructureGPU‑Accelerated Analytics

Overview

Launch

2025

Status

Launched

Generation

6th Gen Intel Xeon Scalable (Xeon 6 P-Cores)

Market

2S Server / AI Host CPU

About this CPU

The Intel Xeon 6776P is a 64-core, 128-thread server processor based on the Granite Rapids-SP architecture, designed for dual-socket AI, HPC, and database servers that need high core counts, large cache, and wide PCIe 5.0 connectivity.

Intel’s Xeon 6776P sits in the 6700P P‑core lineup with 64 cores, 128 threads, a 2.3 GHz base and 3.9 GHz max turbo, plus a 336 MB L3 cache and 88 PCIe 5.

0 lanes. It supports 8‑channel DDR5 and MRDIMM memory up to 4 TB, and is designed for 2S servers hosting large GPU accelerators, in‑memory databases, and HPC workloads. Its 350 W TDP and enterprise feature set make it a strong fit for AI host nodes and virtualized infrastructure, though it is overkill for lightly‑threaded or purely throughput‑oriented tasks that could use smaller, cheaper Xeon 6 SKUs.

Specifications

ArchitectureGranite Rapids-SP (P-Cores, Redwood Cove)
Manufacturing ProcessIntel 3 (7nm-class)
Cores / Threads64 / 128
Base Clock2.3 GHz
Boost Clock3.9 GHz
L3 Cache336 MB
TDP350 W
SocketFCLGA4710
Memory TypeDDR5, MRDIMM
Memory SpeedDDR5-6400; MRDIMM-8800; max 8000 MT/s
Memory ChannelsOcta-Channel (8)
Max Memory4096 GB
PCIe Version / LanesPCIe 5.0 × 88
Integrated GraphicsNone
Octa-Channel88 PCIe Lanes
Target Audience
GamersStreamersContent CreatorsDevelopersWorkstation UsersOffice UsersStudents

Performance

Productivity
N/A

Strong multi‑threaded throughput for server workloads; specific application scores not available without official benchmarks.

Virtualization
N/A

Well‑suited to large‑scale virtualization thanks to high core count, memory capacity, and I/O, but numerical scores are not published for this SKU.

Gaming
N/A

Not targeted at gaming; single‑thread performance is reasonable for a server CPU, but gaming is not a design target and benchmark data is unavailable.

Efficiency
N/A

350 W TDP at 64 cores results in moderate performance‑per‑watt compared to lower‑core Xeon 6 SKUs; efficiency is acceptable when the cores and I/O are fully utilized.

GamingNot applicable
  • Server‑focused processor with no gaming‑oriented benchmarks.
  • Single‑thread boost up to 3.9 GHz is decent, but gaming is not a target use case.
  • Use desktop or workstation CPUs for gaming‑centric builds.
CreatorVery Good
Blender (CPU rendering)V‑Ray / ArnoldHandBrake / FFmpeg encodingDaVinci Resolve (CPU‑bound stages)After Effects (rendering)
AI / MLStrong
  • Optimized as a host CPU for GPU‑accelerated AI systems (e.g., NVIDIA DGX B300).
  • Supports Intel AMX, DL Boost, and AVX‑512 for CPU‑side AI inference.
  • Best leveraged orchestrating GPUs rather than as a standalone AI accelerator.
Industry Impact
Gaming
Negligible
Workstations
Moderate
Content Creation
Moderate
Virtualization
High

Architecture

Intel 3 (7nm-class)

Process Node

Granite Rapids-SP

Codename

64C / 128T

Core Config

336 MB

L3 Cache

350 W

TDP

Architecture Overview

Granite Rapids-SP is Intel’s 6th‑generation Xeon Scalable server platform, using a chiplet‑based design with performance cores (P‑cores) on an Intel 3 process. It emphasizes high core counts, large caches, and wide I/O for AI and HPC servers.

CPU Design

The Xeon 6776P uses 64 Redwood Cove P‑cores arranged in compute tiles, each core supporting Hyper‑Threading for 128 threads. Cache hierarchy includes 2 MB L2 per core and a shared, tile‑level L3 totaling 336 MB, with some L3 capacity allocated at the tile level rather than strictly per core.

Memory Subsystem

An 8‑channel integrated memory controller supports DDR5 and MRDIMM DIMMs, with official speeds up to DDR5‑6400 and MRDIMM‑8800, and a maximum supported speed of 8000 MT/s. Up to 4 TB of memory is supported in a 2‑socket configuration, providing high bandwidth for AI and HPC datasets.

PCIe & I/O

Up to 88 PCIe 5.0 lanes are exposed from the CPU, with additional platform‑level lanes from the PCH. In 1‑socket Granite Rapids‑SP platforms, some UPI links can be repurposed as PCIe 5.0 lanes, enabling high‑density GPU and NVMe connectivity.

Overclocking

The multiplier is locked; overclocking is not supported. Intel instead provides Priority Core Turbo (PCT) and Speed Select Technology (SST) to prioritize up to 8 cores at higher frequencies (up to 4.6 GHz) for latency‑sensitive workloads.

Generation Comparison
Intel Xeon Platinum 8380 (3rd Gen Xeon Scalable, Ice Lake‑SP)Intel Xeon 6776P
  • Roughly double the L3 cache (336 MB vs 60 MB) and significantly more L2 cache per core.
  • Move from PCIe 4.0 to PCIe 5.0 and from 6‑channel to 8‑channel DDR5/MRDIMM memory.
  • Higher core count (64 vs 40) and higher per‑socket performance in multi‑threaded workloads.
  • Integration of accelerators (QAT, DLB, DSA, IAA, AMX) by default in Xeon 6700P.

Key Highlights

64 P‑Cores with Hyper‑Threading
64 performance cores and 128 threads provide massive parallelism for AI, HPC, and virtualized workloads.
336 MB L3 Cache
Large shared last‑level cache reduces memory access latency for data‑intensive applications.
8‑Channel DDR5/MRDIMM up to 4 TB
High memory bandwidth and capacity, especially with MRDIMMs, feed bandwidth‑hungry AI and HPC kernels.
88 PCIe 5.0 Lanes
Plenty of high‑speed I/O for multiple GPUs, NICs, and NVMe storage in dual‑socket servers.
Priority Core Turbo (PCT)
Allows up to 8 priority cores to run at higher frequency (up to 4.6 GHz) to boost latency‑sensitive tasks.
Strengths
  • 64 cores and 128 threads for highly parallel workloads
  • 336 MB L3 cache reduces memory bottlenecks
  • 8‑channel DDR5/MRDIMM with up to 4 TB memory capacity
  • 88 PCIe 5.0 lanes for GPUs, NICs, and NVMe
  • Built‑in accelerators (QAT, DLB, DSA, IAA, AMX) for AI, networking, and analytics
  • Priority Core Turbo to boost critical threads
Weaknesses
  • High 350 W TDP requires robust cooling and power delivery
  • Premium pricing typical of high‑core‑count Xeon SKUs
  • Locked multiplier; no overclocking headroom
  • Overkill for lightly‑threaded or small‑scale workloads
  • No integrated graphics; relies on discrete or BMC graphics

History

Launch Date
2025
Status
Launched
Generation
6th Gen Intel Xeon Scalable (Xeon 6 P-Cores)
Market
2S Server / AI Host CPU
The Story

Intel introduced the Xeon 6 family in 2024, starting with E‑core‑only Sierra Forest parts at Computex in June 2024, focused on throughput and density. The P‑core‑based Granite Rapids‑SP lineup, including the Xeon 6700P and 6500P series, was launched later, on February 24, 2025, as the performance‑oriented branch for AI and HPC. The Xeon 6776P is one of the 64‑core, 350 W SKUs in this 6700P P‑core family, designed specifically to serve as a host CPU in GPU‑accelerated AI systems such as NVIDIA’s DGX B300.

It reflects Intel’s strategy to combine high core counts, large caches, and integrated accelerators to compete with AMD’s EPYC 9004 and 9005 series in AI and data center markets.

Improvements over Previous Generation

  • Roughly double the L3 cache (336 MB vs 60 MB) and significantly more L2 cache per core.
  • Move from PCIe 4.0 to PCIe 5.0 and from 6‑channel to 8‑channel DDR5/MRDIMM memory.
  • Higher core count (64 vs 40) and higher per‑socket performance in multi‑threaded workloads.
  • Integration of accelerators (QAT, DLB, DSA, IAA, AMX) by default in Xeon 6700P.

Alternatives & Competitors

Intel Xeon 6774P
Same core count and cache with higher base clock (2.5 GHz), better if you need slightly higher frequency at similar TDP.
Intel Xeon 6736P
36‑core, 205 W alternative with lower cost and power when you don’t need 64 cores.
AMD EPYC 9534
64‑core, 280 W competitor with 12 memory channels and 128 PCIe 5.0 lanes, offering different memory/I/O trade‑offs.
AMD EPYC 9575F
Higher‑frequency Zen 5 64‑core CPU at 400 W, aimed at GPU‑heavy AI servers where clock speed matters.
Intel Xeon 6781P
80‑core SKU with more performance headroom for extremely parallel workloads, at similar platform cost.
Direct Competitors
AMD EPYC 9534AMD EPYC 9575FIntel Xeon 6774PIntel Xeon 6781PAMD EPYC 9654

Should You Buy It?

Recommended for the right buyer

Dual‑socket AI or HPC servers with multiple high‑end GPUs where you need 64 cores, large cache, and maximum PCIe 5.0 lanes for I/O density.

Avoid if…

  • Building a small 1S server with moderate I/O needs
  • Prioritizing performance per watt over absolute performance
  • Running lightly‑threaded enterprise apps that don’t exploit 64 cores
  • Seeking a gaming or desktop workstation CPU

Use Cases

AI Inference & Training Host Nodes
Excellent
HPC Clusters (CFD, CAE, Weather)
Excellent
In‑Memory Databases & Analytics
Excellent
Virtualization & VDI Back‑Ends
Very Good
General‑Purpose Enterprise Servers
Good

Interesting Facts

The Xeon 6776P is used as the host CPU in NVIDIA’s DGX B300 AI systems, highlighting its role in GPU‑centric AI servers.

Granite Rapids‑SP can repurpose UPI links as additional PCIe 5.0 lanes in single‑socket configurations, boosting I/O density.

ServeTheHome reports that Granite Rapids‑SP SKUs with 336 MB L3 cache provide about 4.2 MB L3 and 2 MB L2 per core, closing the cache gap with AMD EPYC.

Intel positions Xeon 6700P P‑core CPUs as offering double the memory bandwidth and AI acceleration in every core compared to 5th‑Gen Xeon Scalable.

The 6776P is a 64‑core, 350 W part in a lineup that also includes 36‑core and 86‑core SKUs, letting enterprises right‑size core count and power.

People Also Ask

What is the Intel Xeon 6776P used for?

It is designed as a host CPU for dual‑socket AI and HPC servers, especially those using multiple GPUs, such as NVIDIA DGX B300 systems.

How many PCIe lanes does the Xeon 6776P have?

Intel ARK lists 88 PCIe 5.0 lanes from the CPU in a dual‑socket configuration.

Does the Xeon 6776P have integrated graphics?

No; it relies on discrete GPUs or BMC graphics for display output.

What memory does the Xeon 6776P support?

It supports DDR5 and MRDIMM DIMMs, with official speeds up to DDR5‑6400 and MRDIMM‑8800, and a max speed of 8000 MT/s.

Is the Xeon 6776P good for gaming?

It can run games at a basic level, but it is not optimized for gaming; desktop or workstation CPUs are better choices.

What is the TDP of the Xeon 6776P?

The thermal design power is 350 W.

Can you overclock the Xeon 6776P?

No, the multiplier is locked; Intel provides Priority Core Turbo instead for selective frequency boosts.

What is the difference between Xeon 6776P and 6774P?

Both are 64‑core, 336 MB L3, 350 W CPUs, but the 6774P has a higher base clock (2.5 GHz vs 2.3 GHz).

Which platform does the Xeon 6776P use?

It uses the Intel Xeon 6 Granite Rapids‑SP platform with the LGA4710 socket.

How much L3 cache does the Xeon 6776P have?

336 MB of shared L3 cache.

Frequently Asked Questions

Does the Xeon 6776P support ECC memory?

Yes, ECC is supported; Intel ARK explicitly lists ECC memory support for this SKU.

What is Priority Core Turbo on the 6776P?

PCT allows up to 8 cores to run at a higher turbo frequency (up to 4.6 GHz on this SKU) to improve performance for latency‑sensitive threads while others run at base frequency.

What accelerators are integrated into the Xeon 6776P?

Intel lists QuickAssist Technology (QAT), Dynamic Load Balancer (DLB), Data Streaming Accelerator (DSA), In‑Memory Analytics Accelerator (IAA), and Advanced Matrix Extensions (AMX) as integrated accelerators.

Is the Xeon 6776P a unlocked CPU?

No, the multiplier is locked; overclocking is not supported.

Can I run the Xeon 6776P in a single‑socket server?

Yes, but some UPI links will be repurposed as PCIe 5.0 lanes in 1S configurations, changing the lane distribution compared to dual‑socket use.

What is the maximum memory capacity with Xeon 6776P?

Up to 4 TB of memory is supported in a 2‑socket system, depending on DIMM type and configuration.

What instruction extensions does the 6776P support?

Intel lists AMX, SSE4.2, AVX, AVX2, and AVX‑512, along with two AVX‑512 FMA units.

Does the Xeon 6776P support CXL?

Granite Rapids‑SP as a family supports CXL 2.0; Intel ARK does not list CXL explicitly for this SKU, but the platform supports it.

Is the Xeon 6776P suitable for virtualization?

Yes, with high core count, large memory, and extensive I/O, it is well‑suited for virtualized infrastructure and VDI back‑ends.

What cooling is recommended for the Xeon 6776P?

High‑end air or liquid cooling solutions designed for 350 W+ server CPUs are recommended, following Intel’s thermal guidelines.