Surprisingly Useful AI Article Enhancements
AMD has released its EPYC 8005 series of server CPUs, a lineup targeting edge computing, telecommunications and cloud storage environments where power, space and thermal constraints are critical factors.
The new processors are built on AMD’s Zen 5 architecture and range from eight to 84 cores, operating within a 70 to 225 watt TDP envelope in a single-socket platform.
Performance Gains Over Previous Generation
The top-of-stack 84-core EPYC 8635P delivers 40 per cent higher integer performance compared to its 64-core predecessor, the EPYC 8534P, along with 9.5 per cent higher performance per watt, according to AMD’s published benchmark figures.
Against Intel’s competing offerings, AMD is claiming a wider gap. Within the same TDP class, the 84-core EPYC 8635P offers more than double the cores of the 40-core Intel Xeon 6716P-B while drawing 10 watts less power – 225W compared to 235W.
AMD’s own benchmark data puts the integer performance advantage at 91 per cent over the Xeon 6716P-B.
How It Stacks Up Against Competitors
AMD provided a three-way comparison of its EPYC 8635P against Intel’s Xeon 6 6776P-B and the Nvidia Grace CPU Superchip using SPECpower benchmarks.
The EPYC 8635P recorded 24,408 overall ssj_ops per watt, compared to 21,433 for Intel’s Xeon 6776P-B and 13,218 for the Nvidia Grace CPU Superchip.
The AMD chip operates at a maximum TDP of 225 watts, while the Intel part runs at 325 watts and the Nvidia Grace Superchip at 500 watts. AMD noted that the Nvidia figure reflects the full Grace Superchip module, which integrates two Grace CPU dies within a single package, making direct TDP comparisons less straightforward.
On a per-dollar basis, AMD claims a single-socket server built with the 84-core EPYC 8635P delivers 48 per cent better integer performance per CPU watt per CPU dollar than one built with a 72-core Intel Xeon 6776P-B.
Memory, I/O And Platform Details
The EPYC 8005 series supports six channels of DDR5-6400 ECC memory, with capacity up to 3 TB. For I/O, the chips provide 96 PCIe Gen 5 lanes and eight PCIe Gen 3 lanes.
By comparison, Intel’s Xeon 6776P-B offers eight channels of DDR5-6400 but only 32 PCIe Gen 5 lanes and 16 PCIe Gen 4 lanes.
The chips share the same x86 instruction set architecture and AVX-512 support found in AMD’s flagship EPYC 9005 data centre processors, meaning workloads can move between cloud and edge without code changes or recompilation.
AMD has also included enterprise-class reliability, availability and serviceability (RAS) features, along with wide thermal operating ranges designed for air-cooled systems and features enabling OEM certification of NEBS-compliant designs for harsh telco and outdoor edge environments.
Samsung Deploys Multi-Cell vRAN On A Single Server
Among the early adopters, Samsung tested its multi-cell vRAN software on a single server equipped with the 84-core EPYC 8635P.
The test achieved 54 cell networks, 9.5 Gb/s downlink throughput and 2.0 Gb/s uplink throughput on a single server.
Related: Best Business Laptops for work & school
Related: Best Gaming Laptops
Related: Best Portable Laptop
Michael Kim, Vice President and Head of vRAN Software R&D Group Networks Business at Samsung Electronics, pointed to the combination of AMD’s processing technology and Samsung’s vRAN software as a catalyst for network evolution.
“We are accelerating the industry’s transition to AI-ready, cloud-native networks,” Kim stated. “This powerful combination delivers the performance, flexibility and scalability operators need to evolve toward fully software-driven infrastructures.”
AMD has added Low-Density Parity Check (LDPC) optimisations to the EPYC 8005 series specifically for vRAN workloads. These target Layer 1 performance and are designed to reduce latency and accelerate forward-error-correction processing for 5G, freeing up compute resources for Layer 2 processing and enabling higher uplink throughput.
Retail Edge AI Without GPUs
The EPYC 8005 series is also being positioned for in-store AI deployments. WobotAI is using the processors to run intelligent video AI agents on in-store servers, transforming existing camera infrastructure into continuous monitoring and analytics systems.
Will Kelso, President of WobotAI, highlighted the ability to run AI workloads entirely at the edge.
“With AMD EPYC 8005 Server CPUs, our platform runs entirely at the edge – delivering high performance in a small footprint,” Kelso explained. “That means less reliance on GPUs and the cloud – just efficient, cost-effective TCO and scalable AI running directly where the data is generated.”
Cloud Storage Performance
For cloud storage applications, AMD is targeting software-defined storage platforms that need a balance of compute, memory and I/O without excessive power draw.
Benchmarks using the CephFS RADOS suite show single-socket servers with the EPYC 8635P delivering approximately 1.23 times the throughput of the previous-generation EPYC 8534P across read, write and random read operations.
The 96 lanes of PCIe Gen 5 are intended to support dense SSD configurations and network-attached storage, while the single-socket design allows operators to consolidate storage, virtualisation and security functions onto fewer nodes.
Positioning In The AMD Lineup
AMD has positioned the EPYC 8005 series as a complement to its EPYC 9005 data centre processors rather than a replacement. The EPYC 9005 remains the company’s flagship for mainstream data centre workloads, while the EPYC 8005 is aimed at deployments where power, thermal or space budgets require a different balance.
Target customers include system builders, dedicated hosting providers and businesses looking to scale data centre infrastructure efficiently while creating capacity for AI and future growth.
Further details are available on the AMD EPYC 8005 Server CPU page.
Last Updated on May 20, 2026 by Nick Ross
