Processore AMD EPYC 9655P 2,6 GHz 384 MB L3 socket SP5

5th Gen AMD EPYC processors are purpose-built to accelerate data center, cloud, and AI workloads. The AMD EPYC 9005 series delivers high enterprise computing performance with configurations up to 192 cores and chiplet architecture designed for high memory bandwidth and capacity.

Advantages

AMD EPYC 9005 processors offer strong integer and AI inference performance, high energy efficiency, and reduced total cost of ownership. They enable higher per-server density, lower rack count and power consumption, and options optimized as host CPUs for GPU-accelerated systems to improve overall system ROI.

Key specifications

• Architecture: Zen 5 / Zen 5c cores
• Maximum cores: up to 192 cores per processor
• Target workloads: data center, cloud, AI inference, and GPU-hosted systems
• AI inference: optimized performance for language models (≈13B parameters and below) and other CPU-efficient AI tasks
• GPU hosting: configurations available tuned as host CPUs for multi‑GPU servers
• Density: higher vCPU support compared with comparable 144-core alternatives
• Efficiency: designed for improved energy efficiency and lower TCO

How to use

Deploy AMD EPYC 9005 processors in rack or blade servers according to your platform vendor guidelines. For CPU-only AI inference tasks such as language models up to ~13 billion parameters, select high-core configurations. For GPU-accelerated workloads, choose EPYC 9005 SKUs optimized as host CPUs and combine them with validated GPU configurations to maximize throughput and GPU utilization. Follow standard data center best practices for cooling, power delivery, BIOS/firmware updates, and vendor-provided tuning guides.

Recommendations

For AI inference at moderate model sizes or for dense virtualized cloud deployments, evaluate AMD EPYC 9005 options to reduce rack footprint and power cost; for GPU-heavy training or large model inference, pair EPYC 9005 host-optimized SKUs with validated multi-GPU servers and review vendor benchmark results for your specific workload.