The AMD Epyc 9565 is built on a 4nm semiconductor process and carries a Thermal Design Power rating of 400W, reflecting the sustained power demands of its multi-core design. It supports the PCIe 5.0 interface standard, enabling high-bandwidth connectivity for compatible expansion hardware. The processor is fully 64-bit capable and does not include integrated graphics, positioning it as a dedicated compute solution without any on-die display output.
The Epyc 9565 operates across 72 cores at a base clock of 3.15 GHz, delivering a combined CPU speed of 72 x 3.15 GHz, with a turbo clock of 4.3 GHz available under boosted conditions. It exposes 144 threads in total and uses a fixed clock multiplier of 31.5, with no unlocked multiplier option. The cache hierarchy spans 5760 KB of L1, 72 MB of L2 at 1 MB per core, and a substantial 384 MB of L3 cache — amounting to approximately 5.33 MB of L3 per core — providing considerable on-die storage for frequently accessed data across its many cores.
The Epyc 9565 uses DDR5 memory and supports speeds of up to 6000 MHz across 12 memory channels, enabling a maximum memory bandwidth of 576 GB/s. It can address up to 9000 GB of total system memory and includes support for ECC, allowing the processor to detect and correct certain classes of memory errors — a standard requirement in server and enterprise environments.
The Epyc 9565 supports multithreading, allowing each physical core to handle multiple threads simultaneously. It includes the NX bit for hardware-level memory protection against certain types of malicious code execution. The processor also covers a broad range of instruction sets — MMX, F16C, FMA3, AES, AVX, AVX2, SSE 4.1, and SSE 4.2 — enabling support for workloads that rely on vectorized math, hardware-accelerated encryption, and floating-point operations.
In PassMark testing, the Epyc 9565 achieves a multi-threaded score of 135221, reflecting the cumulative throughput of its 72 cores under parallel workloads. Its single-threaded PassMark result stands at 3696, representing the per-core performance available for tasks that do not scale across multiple threads.