TurboVS
AMD Epyc 9845 specifications and in-depth review

AMD Epyc 9845

Manufacturer: AMD

The AMD Epyc 9845 is a high-core-count server processor designed for demanding enterprise workloads. With 160 cores running at a base frequency of 2.1 GHz and a turbo speed of 3.7 GHz, it offers substantial parallel processing capacity across its 320 threads. Built on a 3 nm semiconductor process, it carries a thermal design power rating of 390W, reflecting its scale as a data center-class component.

On the memory side, the Epyc 9845 supports DDR5 across 12 channels, reaching speeds up to 6000 MHz with a maximum capacity of 9000 GB and a peak bandwidth of 576 GB/s. ECC memory support is included, along with a sizeable cache hierarchy comprising 12800 KB of L1, 160 MB of L2, and 320 MB of L3 cache. The processor supports PCIe 5.0, multithreading, NX bit, and a range of instruction sets including AVX2, FMA3, and AES. In benchmarking, it achieves a PassMark score of 152985 in multi-threaded testing and 3144 in single-threaded evaluation.

Pros
  • Supports up to 9000 GB of DDR5 ECC memory across 12 channels, accommodating extremely large in-memory workloads with built-in error correction
  • 320 MB of L3 cache distributed at 2 MB per core helps sustain throughput across all 160 cores without frequent memory fetches
  • PCIe 5.0 support enables high-bandwidth connections to compatible storage and networking devices
  • A peak memory bandwidth of 576 GB/s at speeds up to 6000 MHz ensures data can be fed to the processor at a substantial rate
  • Broad instruction set support including AVX2, FMA3, and AES allows the processor to handle diverse computational and cryptographic workloads natively
  • Built on a 3 nm manufacturing process, which allows a high core count within the constraints of the platform
Cons
  • A TDP of 390W places significant demands on server cooling infrastructure and power delivery systems
  • The locked clock multiplier means there is no option to adjust operating frequency through multiplier tuning
  • Single-threaded PassMark score of 3144 indicates moderate per-core performance for tasks that do not scale across many threads
  • No integrated graphics means additional hardware is required for any display output or GPU-dependent diagnostics
  • The overclocked PassMark result of 154312 shows minimal gain over the standard score of 152985, suggesting limited headroom beyond base configuration
Who is this for?

This processor is well-suited to large-scale server and data center environments where workloads are designed to run across many cores simultaneously. Its 160-core, 320-thread configuration makes it a practical fit for tasks such as high-performance computing, large database engines, and virtualization platforms that can distribute work efficiently across a wide pool of logical processors. The ability to address up to 9000 GB of ECC DDR5 memory across 12 channels also makes it appropriate for memory-intensive applications such as in-memory analytics or enterprise workloads that demand both capacity and reliability. Environments requiring hardware-accelerated cryptographic processing will also benefit from the native AES instruction set support.

Who is this NOT for?

This processor is not a practical choice for workloads that depend heavily on single-threaded performance, as its single-core PassMark result of 3144 reflects moderate per-core throughput that may bottleneck latency-sensitive or sequentially executed tasks. It is also unsuitable for any deployment requiring integrated graphics, since no on-chip display output is available and a discrete solution would be necessary for even basic visual diagnostics. Additionally, with a TDP of 390W, it demands a substantial power and cooling infrastructure that makes it impractical for compact, power-constrained, or thermally limited environments such as edge deployments or small-footprint server builds.

General info:

Thermal Design Power (TDP) 390W
semiconductor size 3 nm
PCI Express (PCIe) version 5
Supports 64-bit ✓
Has integrated graphics ✕

The AMD Epyc 9845 carries a Thermal Design Power (TDP) of 390W, placing it firmly in the territory of full-scale server hardware. It is manufactured on a 3 nm semiconductor process and supports PCIe 5.0, enabling high-bandwidth connectivity for compatible expansion cards and storage devices. The processor is fully 64-bit capable and does not include integrated graphics, making it suited for headless server deployments where a discrete or external graphics solution would be used if needed.

Performance:

CPU speed 160 x 2.1 GHz
CPU threads 320 threads
turbo clock speed 3.7GHz
L3 cache 320 MB
L1 cache 12800 KB
L2 cache 160 MB
L2 core 1 MB/core
clock multiplier 21
Has an unlocked multiplier ✕
L3 core 2 MB/core

The AMD Epyc 9845 operates across 160 cores at a base clock speed of 2.1 GHz, with a turbo frequency reaching up to 3.7 GHz, and delivers a total of 320 threads for handling highly parallel workloads. The cache hierarchy is substantial, with 12800 KB of L1 cache, 160 MB of L2 cache at 1 MB per core, and 320 MB of L3 cache allocated at 2 MB per core. The processor uses a clock multiplier of 21, and the multiplier is locked, meaning frequency adjustments through multiplier tuning are not supported.

Memory:

Supports ECC memory ✓
maximum memory bandwidth 576 GB/s
DDR memory version 5
RAM speed (max) 6000 MHz
maximum memory amount 9000GB
memory channels 12

The AMD Epyc 9845 supports DDR5 memory across 12 channels, with a maximum RAM speed of 6000 MHz and a peak memory bandwidth of 576 GB/s. It can address up to 9000 GB of total memory, making it capable of handling very large in-memory datasets. ECC memory support is included, providing error detection and correction functionality that is commonly required in server and enterprise environments.

Features:

uses multithreading ✓
instruction sets MMX, F16C, FMA3, AES, AVX, AVX2, SSE 4.1, SSE 4.2
Has NX bit ✓

The AMD Epyc 9845 supports multithreading, allowing each physical core to handle multiple threads simultaneously for more efficient processor utilization. It includes the NX bit, a hardware-level security feature that helps prevent certain classes of malicious code execution. The processor is compatible with a broad set of instruction sets including MMX, AVX, AVX2, FMA3, F16C, AES, SSE 4.1, and SSE 4.2, covering a wide range of computational tasks from floating-point operations to hardware-accelerated encryption.

Benchmarks:

PassMark result 152985
PassMark result (single) 3144
PassMark result (overclocked) 154312

In PassMark testing, the AMD Epyc 9845 achieves a multi-threaded score of 152,985, reflecting its capacity for heavily threaded workloads across its 160 cores. The single-threaded PassMark result stands at 3,144, indicating per-core performance in tasks that rely on sequential execution. When tested in an overclocked configuration, the score rises modestly to 154,312, representing a marginal uplift over the standard multi-threaded result.

Final Verdict

The AMD Epyc 9845 is a processor built unambiguously for large-scale enterprise and data center deployment, where its 160-core, 320-thread architecture paired with up to 9000 GB of ECC DDR5 memory support can be fully leveraged by parallelized, memory-intensive workloads. Its combination of a broad instruction set, PCIe 5.0 connectivity, and a substantial cache hierarchy makes it a technically complete platform for demanding server environments. That said, its 390W thermal footprint and moderate single-threaded performance mean it is best placed in infrastructure specifically designed to accommodate high-core-count processors rather than general-purpose or resource-constrained deployments. For organizations running workloads that scale effectively across many cores, the Epyc 9845 represents a well-specified and technically coherent choice within the enterprise CPU segment.

Popular Comparisons

AMD Epyc 9845
AMD Epyc 9845
VS
AMD Epyc 9755
AMD Epyc 9755