AMD EPYC 9825 Server Processor – 144 Cores / 288 Threads, 2.2 GHz, 384 MB L3 Cache, Socket SP5, OEM

Processor AMD EPYC 9825 operates at a base frequency of 2.2 GHz with up to 3.7 GHz boost and features a large 384 MB L3 cache. Part of the 5th Gen AMD EPYC family and compatible with Socket SP5, this server-class CPU provides up to 144 cores and supports dodeca-channel DDR5 memory with a maximum theoretical memory bandwidth of 576 GB/s, making it suitable for dense virtualized, cloud and AI workloads that require high core counts, memory capacity and energy-efficient performance.

Advantages

The AMD EPYC 9825 offers high core density (144 cores) and a substantial L3 cache, delivering strong integer and multithreaded performance for server workloads. Support for Socket SP5 and dodeca-channel DDR5 memory enables high memory bandwidth and capacity. The processor is designed to provide improved per-server performance and energy efficiency compared with previous generations, helping reduce rack count, power consumption and total cost of ownership for dense deployments.

Suitable for

• Virtualization and cloud infrastructure (large numbers of vCPUs)
• CPU-only AI inference for medium-sized language and image models
• High-performance compute and enterprise server applications
• GPU-host systems where a high-core-count host CPU is required

Key specifications

• Processor family: AMD EPYC (5th Gen)
• Model: 9825
• Base frequency: 2.2 GHz; Boost frequency: up to 3.7 GHz
• Cores: 144
• L3 cache: 384 MB
• Socket: SP5
• Memory: dodeca-channel DDR5-SDRAM; max theoretical bandwidth 576 GB/s
• On-board graphics: Not available

How to use

Install the AMD EPYC 9825 into a compatible Socket SP5 motherboard following the server vendor's installation guide and thermal specifications. Use appropriate cooling and power delivery for high-core-count operation. Configure memory in matched DDR5 modules to take advantage of the dodeca-channel topology and populate channels according to the motherboard documentation. For virtualization or AI inference workloads, pair the CPU with sufficient RAM and, when needed, GPUs with a host CPU optimized for multi-GPU configurations.