Επεξεργαστής AMD EPYC 9645 2,3 GHz 256 MB L3 Socket SP5

The AMD EPYC 9645 is a server processor from the 5th Gen AMD EPYC 9005 series, operating at a base frequency of 2.3 GHz with boost up to 3.7 GHz and featuring a 256 MB L3 cache. Designed for data center, cloud and AI workloads, this Socket SP5 CPU offers up to 96 cores and supports dodeca-channel DDR5 memory with maximum memory bandwidth of 576 GB/s, delivering high density, performance and energy efficiency for demanding enterprise and cloud environments.

Advantages

The EPYC 9645 provides high integer and AI inference performance on CPU-only servers and can be used as a host CPU for GPU-accelerated systems to improve overall system throughput. With Zen 5 architecture and chiplet design it delivers strong per-core and per-socket performance while improving energy efficiency and total cost of ownership compared with previous and competing solutions. The processor’s large L3 cache, high core count and wide DDR5 memory interface enable increased virtual machine density and reduced rack and power footprint for equivalent workloads.

Suitable for

• Virtualization and cloud infrastructure (high vCPU density)
• CPU-based AI inference for models up to mid-size (for example language models ~13B parameters and similar workloads)
• GPU-host systems for accelerated AI and ML workloads
• Enterprise databases, analytics and high-throughput server applications

Key specifications

• Processor family: AMD EPYC 9005 (5th Gen)
• Model: EPYC 9645
• Base frequency: 2.3 GHz; Boost frequency: up to 3.7 GHz
• Cores: 96
• Cache: 256 MB L3
• Socket: SP5
• Memory: Dodeca-channel DDR5-SDRAM, up to 576 GB/s bandwidth
• On-board graphics: Not available

How to use

Install the EPYC 9645 into a compatible server motherboard with Socket SP5 and follow the server vendor’s installation and cooling guidelines. Configure the system firmware and BIOS settings to match memory and CPU topology (dodeca-channel memory, NUMA regions) and apply appropriate power and thermal limits for your chassis. For virtualization or cloud deployments, allocate vCPUs according to workload requirements; for GPU-accelerated systems, pair the processor with supported GPUs and validate driver and firmware compatibility.

Recommendations

For best results, use balanced DDR5 memory configurations across channels, ensure adequate cooling and power delivery, and keep system firmware and drivers up to date. When deploying AI inference or GPU-accelerated workloads, validate application scaling and choose EPYC 9645 configurations that match the desired GPU-to-CPU ratio and memory capacity.