The AMD Epyc 9375F is built on a 4nm process node and carries a Thermal Design Power of 320W, reflecting the thermal envelope expected for this class of server processor. It supports 64-bit computing and comes with PCIe 5.0 for high-bandwidth device connectivity, though it does not include integrated graphics, making a discrete or external graphics solution necessary for display output.
The AMD Epyc 9375F runs 32 cores at a base clock of 3.85GHz across all cores, supporting 64 threads in total, with a turbo clock speed of 4.8GHz for sustained peak frequency under load. The clock multiplier sits at 38.5, though the multiplier is locked and cannot be adjusted. Cache is distributed across three levels: 2560KB of L1, 32MB of L2 at 1MB per core, and a 256MB L3 cache allocated at 8MB per core — providing substantial on-chip storage to help reduce memory latency during compute-intensive tasks.
The AMD Epyc 9375F uses DDR5 memory across 12 channels, supporting speeds of up to 6000MHz and delivering a maximum memory bandwidth of 576GB/s. It can address up to 9000GB of total memory, making it well-suited for workloads that depend on large memory pools. ECC memory support is also included, providing hardware-level error correction to help maintain data integrity in server environments where reliability is essential.
The AMD Epyc 9375F supports multithreading, allowing each physical core to handle two threads simultaneously for more efficient utilization under parallel workloads. It also includes the NX bit, a hardware security feature that helps prevent certain classes of malicious code execution. On the instruction set side, the processor supports a broad range of extensions — including AVX and AVX2 for vectorized floating-point operations, FMA3, F16C, AES for hardware-accelerated encryption, MMX, SSE 4.1, and SSE 4.2 — covering a wide spectrum of compute and security-oriented workloads.
In PassMark testing, the AMD Epyc 9375F achieves a multi-threaded score of 95,768, reflecting its throughput across all cores and threads under parallel workloads. Its single-threaded PassMark result stands at 3,762, representing per-core performance in tasks that do not scale across multiple threads.