Quick Verdict
The Xeon 6706P-B brings Granite Rapids P-cores to a BGA footprint, with 40 cores, 160 MB of L3 cache, and on-die accelerators (AMX, QAT, DSA, DLB) that shine in telecom, security, and edge AI. Its 235 W TDP and 4-channel DDR5-6400 deliver strong throughput, though the BGA package locks platform choice and 48 PCIe lanes are fewer than many OEM-socket SKUs.
Overview
Launch
2025
Status
LaunchedGeneration
6th Gen Xeon
Market
Server (networking/edge/embedded)
The Intel Xeon 6706P-B is a 40-core server processor built on the Intel 3 process with 160 MB of cache, designed for telecom and edge workloads that demand high per-core performance, AI acceleration, and rich I/O.
The Xeon 6706P-B offers 40 P-cores running at 2.5 GHz base and up to 3.5 GHz turbo, with 160 MB of cache and 4 channels of DDR5-6400 for up to 1.
13 TB of ECC memory. It integrates accelerators for AI (AMX, DL Boost), crypto and compression (QAT), data streaming (DSA), and load balancing (DLB), plus security features like TDX and SGX. The BGA4368 package and 1S scalability make it ideal for appliance and embedded designs rather than field-upgradeable servers.
Specifications
Performance
- •Not designed for gaming; server platforms typically lack high refresh graphics support and optimizations expected in gaming PCs.
- •Supports AMX and Intel DL Boost (AVX-512 VNNI), enabling competitive AI inference on CPU for recommendation, vision, and LLM small-batch workloads; official MLPerf results show Xeon 6 P-cores achieving notable uplift over prior generation.
Architecture
Intel 3
Process Node
Granite Rapids-D
Codename
40C / 80T
Core Config
160 MB
L3 Cache
235 W
TDP
Architecture Overview
Granite Rapids-D is the server- and edge-focused variant of the 6th Gen Xeon family, using Intel 3 with up to two compute tiles and separate I/O dies that host DDR5 controllers, PCIe 5/4 lanes, and accelerators like QAT, DSA, and DLB. This modular design lets Intel scale core counts and I/O independently across products.
CPU Design
The 6706P-B packs 40 performance-only cores (P-cores) without hyper-threading alternatives, each core supporting AVX-512 and AMX. Hyper-Threading doubles threads to 80, improving throughput for concurrent workloads.
Memory Subsystem
Four memory channels support DDR5 up to 6400 MT/s with ECC, enabling up to 1.13 TB of capacity. This is a channel reduction versus high-end socketed SKUs, but matches typical 1S appliance needs.
PCIe & I/O
48 total lanes—32 of PCIe Gen 5 and 16 of Gen 4—are routed from the I/O dies. This allows multiple NVMe drives, 100G NICs, or accelerators without switches, albeit fewer lanes than flagship socketed platforms.
Overclocking
The SKU ships in a BGA package without an exposed multiplier; overclocking is not supported, and power/frequency management is handled by platform firmware.
- Move from 10nm to Intel 3 process improves efficiency and clocks
- Adoption of AMX and updated accelerators significantly boosts AI and crypto performance
- PCIe Gen 5 support doubles bandwidth for NVMe and accelerators compared to Gen 4
- Increased L3 cache size up to 160 MB improves data locality versus prior D-series
Key Highlights
- 40 P-cores with 80 threads and high per-core performance for server workloads
- 160 MB of L3 cache improves throughput for memory-bound tasks
- Integrated accelerators (AMX, QAT, DSA, DLB) offload AI, crypto/compression, and networking
- PCIe 5.0 + PCIe 4.0 for modern NVMe, NICs, and accelerators
- Strong security features including TDX, SGX, and Total Memory Encryption
- Quad-channel DDR5-6400 with ECC for reliable, high-bandwidth memory
- BGA package prevents field upgrades and limits platform flexibility
- 48 PCIe lanes are fewer than high-end socketed Xeon platforms
- No integrated graphics; dedicated GPU required if display output is needed
- 235 W TDP requires robust thermal solution in dense appliance designs
- Supports only single-socket configurations
History
Intel first disclosed the split between P-core and E-core Xeon 6 lines in 2022, and by early 2025 it completed the rollout with Granite Rapids-D for edge/network and Sierra Forest for throughput workloads. The 6706P-B arrived in this wave as a BGA-focused 40 P-core SKU targeting appliances and telecom gear, bringing Intel 3, AMX, and vRAN Boost to fixed designs. Intel’s public launch materials and ARK confirm Q1’25 availability, with specific guidance like 235 W TDP and 48 PCIe lanes.
By integrating accelerators that used to be separate add-ins, Intel aimed to reduce board space and power in edge nodes, aligning with operator needs for denser, more efficient infrastructure.
Improvements over Previous Generation
- Move from 10nm to Intel 3 process improves efficiency and clocks
- Adoption of AMX and updated accelerators significantly boosts AI and crypto performance
- PCIe Gen 5 support doubles bandwidth for NVMe and accelerators compared to Gen 4
- Increased L3 cache size up to 160 MB improves data locality versus prior D-series
Alternatives & Competitors
Should You Buy It?
Recommended for the right buyer
Fixed-form-factor appliances, edge servers, and telecom infrastructure where 40 cores with built-in accelerators and BGA mounting are required by design.
Avoid if…
- You need socketed upgradability or field replaceable CPUs
- Your workloads benefit from many PCIe Gen 5 lanes beyond 48
- You prefer higher per-core frequency or higher core counts within the same power envelope
Use Cases
Interesting Facts
Marked as Products formerly Granite Rapids-D on Intel ARK, confirming the edge/network pedigree of this die variant.
Unlike most socketed Xeons, the 6706P-B is sold only in BGA4368, targeting embedded and appliance OEMs.
Intel ARK lists its Recommended Customer Price at $3,584 (guidance) for a 40-core P-core SKU.
It includes Intel vRAN Boost, making it tailored for baseband and virtual RAN workloads.
The I/O die hosts DSA, QAT, and DLB accelerators, reflecting its design to offload data center workloads.
Intel 3 node marks the company’s first use of EUV in high-volume server production, improving density and efficiency.
It supports Intel Trust Domain Extensions (TDX) for confidential computing, alongside legacy SGX.
Memory configuration maxes at 1.13 TB, reflecting four DDR5 channels at 6400 MT/s with current DIMM capacities.
Turbo Boost 2.0 is specified for the 6706P-B, with a max turbo of 3.5 GHz across the 40 cores.
Designed for Communications Commercial Temperature use, indicating support for higher ambient temperature environments typical of telecom gear.
People Also Ask
What is the core count of the Intel Xeon 6706P-B?
The Xeon 6706P-B has 40 performance cores and 80 threads with Hyper-Threading enabled.
Does the Intel Xeon 6706P-B support DDR5 memory?
Yes, it supports four channels of DDR5 up to 6400 MT/s with ECC for up to 1.13 TB of memory.
How many PCIe lanes does the Intel Xeon 6706P-B provide?
It provides 48 total PCIe lanes: 32 lanes of PCIe Gen 5 and 16 lanes of PCIe Gen 4.
Is the Intel Xeon 6706P-B socketed or BGA?
The 6706P-B is a BGA4368 (soldered) package designed for embedded and appliance platforms.
What is the TDP of the Intel Xeon 6706P-B?
Intel lists the TDP as 235 W for the Xeon 6706P-B.
Does the Intel Xeon 6706P-B have integrated graphics?
No, the Xeon 6706P-B does not have integrated graphics; a discrete GPU is required for display output.
What process node is the Intel Xeon 6706P-B built on?
It is manufactured using Intel’s Intel 3 process, as listed on Intel ARK.
What accelerators are integrated into the Intel Xeon 6706P-B?
The 6706P-B includes Intel AMX, QAT, DSA, and DLB, plus Intel vRAN Boost and DL Boost.
What is the base and boost clock of the Intel Xeon 6706P-B?
Base frequency is 2.5 GHz and max turbo frequency is 3.5 GHz according to Intel ARK.
Is the Intel Xeon 6706P-B unlocked for overclocking?
No, it is a BGA part with a locked multiplier; overclocking is not supported.
Frequently Asked Questions
What socket does the Intel Xeon 6706P-B use?
It uses the FCBGA4368 package and is not designed for socketed, field-replaceable installation.
Can the Xeon 6706P-B be used in multi-socket servers?
Intel ARK lists Scalability as 1S Only, so it is designed for single-socket platforms.
Does the Intel Xeon 6706P-B support Optane persistent memory?
No, Intel ARK indicates Intel Optane Persistent Memory is not supported for this SKU.
What security features does the Intel Xeon 6706P-B include?
It supports Intel TDX, SGX (with SPS), Total Memory Encryption (multi-key), Boot Guard, and other RAS features.
What is the L3 cache size of the Intel Xeon 6706P-B?
The processor has 160 MB of L3 cache according to Intel’s specification page.
Is the Intel Xeon 6706P-B suitable for AI workloads?
Yes, with AMX and DL Boost it is suited for CPU-based AI inference; Intel cites ~1.9x AI performance gains vs prior generation in MLPerf tests.
What is the maximum memory capacity of the Intel Xeon 6706P-B?
Intel ARK lists Max Memory Size as 1.13 TB, dependent on memory type and DIMM capacities.
Does the Intel Xeon 6706P-B support CXL?
The ARK page does not explicitly list CXL support; confirmation would require the platform or chipset documentation.
What is Intel vRAN Boost on the Xeon 6706P-B?
vRAN Boost is an integrated acceleration feature for virtualized radio access network (vRAN) workloads, offloading baseband processing.
What operating temperatures are supported by the Intel Xeon 6706P-B?
Intel lists Use Conditions as Communications Commercial Temp, indicating support for telecom-grade thermal environments.