LaunchedXeon 6 (Granite Rapids-SP)

Intel · Xeon 6700P Series

Intel Xeon 6728P

24 P-cores with AMX and 8-channel DDR5 for AI, databases, and virtualized infrastructure.

AI Inference and Small LLM HostingVirtualization and VDIIn-Memory DatabasesEnterprise ApplicationsHigh-End Workstations

Cores / Threads

24/ 48

Base / Boost

2.7/ 4.1 GHz

PCIe Lanes

88

L3 Cache

144MB

TDP

210W

Socket

FCLGA4710

Verdict

8.6/ 10

86

Quick Verdict

A balanced Granite Rapids-SP SKU with strong per-core performance, large cache, and serious AI acceleration, best suited for memory-intensive and AI-augmented server workloads rather than cost-sensitive or purely throughput-oriented deployments.

Best for:AI Inference and Small LLM HostingVirtualization and VDIIn-Memory DatabasesEnterprise ApplicationsHigh-End Workstations

Overview

Launch

2025

Status

Launched

Generation

Xeon 6 (Granite Rapids-SP)

Market

2S/4S/8S Server and High-End Workstation

About this CPU

The Intel Xeon 6728P is a 24-core, 48-thread server and workstation processor based on the Granite Rapids-SP platform, offering 144 MB of L3 cache, 8-channel DDR5-6400 memory, and 88 PCIe 5.0 lanes in a 210 W TDP envelope.

Intel’s Xeon 6728P sits in the middle of the Granite Rapids-SP stack with 24 P-cores, 48 threads, a 2.7 GHz base and 4.1 GHz max turbo, and 144 MB of shared L3 cache.

It supports 8-channel DDR5-6400 with up to 4 TB of memory per socket, 88 PCIe 5.0 lanes, and Intel’s latest AMX and on-die accelerators. That combination makes it a strong fit for virtualized infrastructure, in-memory databases, and AI inference workloads that benefit from high memory bandwidth and matrix math acceleration, though its 210 W TDP and premium pricing push it toward enterprise and professional use rather than budget homelab builds.

Specifications

ArchitectureGranite Rapids-SP (Redwood Cove P-cores)
Manufacturing ProcessIntel 3 compute tiles + Intel 7 I/O tiles (commonly marketed as ~3 nm class)
Cores / Threads24 / 48
Base Clock2.7 GHz
Boost Clock4.1 GHz
L3 Cache144 MB
TDP210 W
SocketFCLGA4710
Memory TypeDDR5
Memory Speed6400 MT/s
Memory ChannelsOcta-Channel (8)
Max Memory4096 GB
PCIe Version / Lanes5.0 × 88
Integrated GraphicsNone
Octa-Channel88 PCIe Lanes
Target Audience
GamersStreamersContent CreatorsDevelopersWorkstation UsersOffice UsersStudents

Performance

Productivity
90Excellent

Strong multi-threaded and memory-bandwidth-heavy productivity performance; benefits from cache, DDR5 channels, and AMX where applicable.

Virtualization
93Excellent

Excellent for high-VM-count and memory-intensive virtualization thanks to 8-channel DDR5 and large cache.

Gaming
60Below Average

Not designed for gaming; can game in a pinch but lacks iGPU and is far from cost-effective for pure gaming builds.

Efficiency
70Good

Competitive performance-per-watt within Granite Rapids, but 210 W TDP is still substantial and requires robust cooling.

GamingPoor
  • No integrated graphics; discrete GPU required.
  • High platform cost makes it unattractive for gaming versus consumer CPUs.
  • Adequate for casual gaming but not a target use case.
CreatorGood
Blender (CPU mode)V-Ray / Arnold renderingAdobe Premiere Pro / DaVinci Resolve (with GPU)Autodesk Maya / 3ds MaxSimulation and CAE
AI / MLVery Good
  • Intel AMX accelerates INT8 and BF16 matrix operations for deep learning inference.
  • Good fit for CPU-based LLM inference and small-to-medium model serving.
  • AI performance per core significantly better than pre-AMX Xeon generations.
Industry Impact
Gaming
Low
Workstations
High
Content Creation
Moderate
Virtualization
High

Architecture

Intel 3 compute tiles + Intel 7 I/O tiles (commonly marketed as ~3 nm class)

Process Node

Granite Rapids-SP

Codename

24C / 48T

Core Config

144 MB

L3 Cache

210 W

TDP

Architecture Overview

Granite Rapids-SP is Intel’s 6th-generation Xeon Scalable server architecture, built on a tile-based design with Redwood Cove P-cores on Intel 3 compute tiles and Intel 7 I/O tiles, connected via embedded interconnects.

CPU Design

The Xeon 6728P uses one HCC compute tile with 24 Redwood Cove P-cores and 48 threads via Intel Hyper-Threading, along with 144 MB of shared L3 cache and integrated accelerators.

Memory Subsystem

An 8-channel DDR5 memory controller supports DDR5-6400 with ECC and up to 4 TB per socket, delivering high bandwidth for memory-bound workloads and CXL 2.0 memory expansion.

PCIe & I/O

88 PCIe 5.0 lanes from the CPU provide substantial I/O bandwidth for GPUs, NVMe, and networking, with flexibility for multi-socket topologies.

Overclocking

The multiplier is locked; Xeon 6728P is not designed for enthusiast overclocking and relies on Intel Turbo Boost and Speed Select Technology for performance tuning.

Generation Comparison
Intel Xeon Platinum 8260 (Cascade Lake-SP, 24C/48T, ~36 MB L3, DDR4-2933, PCIe 3.0)Intel Xeon 6728P
  • Move from DDR4-2933 to DDR5-6400 with 8 channels vs 6 channels.
  • L3 cache increased from ~36 MB to 144 MB.
  • PCIe 3.0 to PCIe 5.0 with more lanes.
  • Addition of AMX, QAT, DLB, DSA, IAA accelerators.
  • Significant IPC and efficiency gains from Redwood Cove cores and Intel 3 process.

Key Highlights

24 Redwood Cove P-Cores
Higher single-thread performance and IPC compared to prior Xeon generations, with AVX-512 and AMX support for AI and HPC workloads.
144 MB L3 Cache
Large shared cache improves performance for database, analytics, and latency-sensitive server workloads.
8-Channel DDR5-6400
Substantially more memory bandwidth than mainstream desktop platforms, ideal for memory-bound and virtualized applications.
AMX and On-Die Accelerators
Intel AMX accelerates INT8/BF16 matrix operations for AI inference; QAT, DLB, DSA, and IAA offload crypto, data streaming, and analytics workloads.
88 PCIe 5.0 Lanes
Plenty of high-bandwidth I/O for GPUs, NVMe, and SmartNICs in server and workstation configurations.
Strengths
  • 24 high-performance Redwood Cove P-cores with strong IPC.
  • 144 MB L3 cache benefits latency-sensitive workloads.
  • 8-channel DDR5-6400 with up to 4 TB per socket.
  • 88 PCIe 5.0 lanes for GPUs, NVMe, and networking.
  • AMX and on-die accelerators (QAT, DLB, DSA, IAA) for AI and data processing.
  • Supports 2S/4S/8S configurations for scalable enterprise deployments.
Weaknesses
  • High platform cost and 210 W TDP require robust cooling and power delivery.
  • Locked multiplier; no overclocking headroom.
  • No integrated graphics; must pair with discrete GPU or BMC.
  • Memory and motherboard ecosystem are more expensive than consumer platforms.
  • Less core-count density than higher-end Granite Rapids or EPYC 9004 SKUs.

History

Launch Date
2025
Status
Launched
Generation
Xeon 6 (Granite Rapids-SP)
Market
2S/4S/8S Server and High-End Workstation
The Story

The Xeon 6728P launched in February 2025 as part of Intel’s Xeon 6700P Granite Rapids-SP family, filling the mid-range of Intel’s new server platform. It arrived after the high-core-count Xeon 6900P Granite Rapids-AP parts and alongside the Xeon 6500P series, targeting customers who needed strong per-core performance and 8-channel DDR5 but not the maximum core counts or the highest TDPs. Built on Redwood Cove P-cores and Intel 3 compute tiles, the 6728P replaced older Cascade Lake and Ice Lake 24-core Xeons in many server designs, offering a substantial jump in memory bandwidth, cache, and AI capabilities through AMX and on-die accelerators.

Over time, it has become a common choice for enterprise servers and AI-focused workstations where 24 high-frequency cores, large cache, and 8-channel memory provide a good balance between cost and performance.

Improvements over Previous Generation

  • Move from DDR4-2933 to DDR5-6400 with 8 channels vs 6 channels.
  • L3 cache increased from ~36 MB to 144 MB.
  • PCIe 3.0 to PCIe 5.0 with more lanes.
  • Addition of AMX, QAT, DLB, DSA, IAA accelerators.
  • Significant IPC and efficiency gains from Redwood Cove cores and Intel 3 process.

Alternatives & Competitors

Intel Xeon 6521P
Same 24C/48T Granite Rapids-SP family but lower TDP and price if you don’t need the full 210 W performance profile.
AMD EPYC 9224
24-core Zen 4 server CPU with 12-channel DDR5 and competitive performance; better if you prioritize core count or prefer AMD’s platform.
Intel Xeon 6731P
Higher 32C/64T count within the same Granite Rapids-SP platform if you need more threads and can afford the higher TDP.
AMD EPYC 9124
Lower-cost 16-core Zen 4 server CPU if your workload doesn’t require 24 cores and you want to reduce platform cost.
Intel Xeon Platinum 8260 (used)
Older 24-core Cascade Lake-SP part available on the used market at lower cost if you don’t need DDR5, PCIe 5.0, or AMX.
Direct Competitors
AMD EPYC 9224Intel Xeon 6521PIntel Xeon 6731PAMD EPYC 9124Intel Xeon Platinum 8380

Should You Buy It?

Recommended for the right buyer

Building or upgrading a 2S/4S server or workstation for AI inference, in-memory databases, or virtualization where 8-channel DDR5 and AMX are valuable.

Avoid if…

  • Your workload is modest and fits cheaper Xeon E or older Xeon Scalable CPUs.
  • You prioritize maximum cores per socket and can afford Xeon 6900P or EPYC 9004.
  • You are building a pure gaming or light-use desktop.

Use Cases

AI Inference / Small LLM Hosting
Excellent
Virtualization and VDI
Excellent
In-Memory Databases (e.g., SAP HANA)
Excellent
Enterprise Application Servers
Very Good
High-End Workstation (CAD/CAE/Rendering)
Good

Interesting Facts

Granite Rapids-SP uses a chiplet design with Intel 3 compute tiles and Intel 7 I/O tiles, a major shift from monolithic Xeon dies.

Redwood Cove P-cores first appeared in Meteor Lake client CPUs and were adapted for server use in Granite Rapids.

SPEC CPU2017 results published by Cisco and others show multi-socket Xeon 6728P systems delivering over 1100 SPECrate2017_int_base on integer workloads.

The 6728P supports Intel Speed Select Technology with configurable TDP profiles, allowing a 185 W mode for denser deployments.

Versus lists the 6728P as having a 3 nm process technology, reflecting Intel 3’s ~3 nm class marketing rather than a literal node size.

The 6728P can address up to 4 TB of DDR5 memory per socket when using 64 GB DIMMs.

Intel’s Xeon 6 launch emphasized up to 2x higher AI inference performance vs prior Xeon generations, heavily leveraging AMX.

The 6700P/6500P series is the first Xeon 6 family to support 8-socket configurations with the LGA4710 socket.

Granite Rapids-SP includes CXL 2.0 support for memory expansion and cache-coherent accelerators.

The 6728P’s 144 MB cache gives 6 MB per core, more than many competing 24-core server CPUs.

People Also Ask

Is Intel Xeon 6728P good for AI workloads?

Yes. It includes Intel AMX for INT8/BF16 matrix acceleration and 8-channel DDR5 for high memory bandwidth, making it well suited for CPU-based AI inference and small LLM hosting.

How much memory does the Xeon 6728P support?

Up to 4 TB of DDR5-6400 memory per socket across 8 channels, depending on DIMM capacity and population rules.

What socket does the Xeon 6728P use?

It uses the FCLGA4710 socket (LGA4710), part of Intel’s Birch Stream platform for Granite Rapids-SP processors.

Does the Xeon 6728P have integrated graphics?

No. Like other Xeon 6700P/6500P processors, it has no integrated GPU and requires a discrete graphics card or BMC for display output.

Can you overclock the Xeon 6728P?

No. The multiplier is locked; performance tuning relies on Intel Turbo Boost and Speed Select Technology profiles rather than overclocking.

What process node is the Xeon 6728P built on?

Its compute tiles use Intel’s Intel 3 process (roughly 3 nm class), with I/O tiles on Intel 7, though some third parties simplify this to “3 nm” or “5 nm” in marketing.

How many PCIe lanes does the Xeon 6728P have?

It provides 88 PCIe 5.0 lanes from the CPU, enabling high-bandwidth connectivity for GPUs, NVMe, and networking cards.

Is Xeon 6728P better than EPYC 9224?

It depends on workload. Xeon 6728P offers AMX and higher DDR5 speed per channel, while EPYC 9224 offers 12-channel DDR5 and more cores at similar TDP; for many server workloads they are competitive.

What is the TDP of Intel Xeon 6728P?

The default TDP is 210 W. Intel Speed Select Technology also defines a 185 W configuration for denser server deployments.

Which servers use the Xeon 6728P?

Multiple vendors including HPE, Cisco, Fujitsu, and Lenovo offer servers based on Xeon 6728P, such as HPE ProLiant DL380 Gen12 and Cisco UCS X410 M8.

Frequently Asked Questions

What generation is the Intel Xeon 6728P?

It belongs to Intel’s Xeon 6 (Granite Rapids-SP) generation, which is the 6th generation of Xeon Scalable server processors.

How many cores and threads does the Xeon 6728P have?

It has 24 cores and supports Intel Hyper-Threading for 48 threads.

What is the max turbo frequency of the Xeon 6728P?

The maximum single-core turbo frequency is 4.1 GHz, with an all-core turbo of 3.9 GHz under ideal conditions.

What memory speed does the Xeon 6728P support?

It supports DDR5-6400 (6400 MT/s) with ECC across 8 memory channels.

Does the Xeon 6728P support CXL?

Yes, Granite Rapids-SP supports CXL 2.0 Type 3 devices for memory expansion and coherent accelerators.

Is the Xeon 6728P unlocked for overclocking?

No. The processor has a locked multiplier and is not designed for overclocking; it relies on Intel Turbo Boost and Speed Select Technology instead.

What accelerators are included in the Xeon 6728P?

It includes Intel QuickAssist Technology (QAT), Dynamic Load Balancer (DLB), Data Streaming Accelerator (DSA), and In-Memory Analytics Accelerator (IAA), plus Intel AMX for AI workloads.

What is the maximum memory capacity supported by Xeon 6728P?

Up to 4 TB of DDR5 memory per socket when using 64 GB DIMMs, subject to population and vendor support.

Can Xeon 6728P be used in workstations?

Yes. It is used in high-end workstations for AI development, rendering, and simulation, though most buyers will prefer Xeon W-2400/3400 or Granite Rapids-WS for single-socket workstation use.

Is Xeon 6728P suitable for homelab use?

Only for very high-end homelabs. Its cost, TDP, and platform complexity make it overkill for most homelab builds; cheaper Xeon E or older Xeon Scalable CPUs are usually more practical.