AMD EPYC 9654

The AMD EPYC 9654 is a 96-core, 192-thread server processor built on the Zen 4 architecture and the 5 nm process node, designed for the highest-density compute workloads. Its benchmark profile reveals a processor whose average score of 29409 places it in the 86th percentile of all CPUs, yet its nearest rivals in the database are not other server parts but rather mainstream desktop and mobile processors. This unusual positioning warrants a closer look at what the data actually indicates about its performance character and intended role.

How It Compares

Against the AMD Ryzen 7 3800X, the EPYC 9654 shows a delta of just 0.1% in average benchmark score, with the EPYC scoring 29409 versus 29385. This near-identical average is deceptive, as the Ryzen 7 3800X is an 8-core desktop part from an older generation, while the EPYC 9654 is a 96-core server behemoth. The data suggests that the average benchmark score, which likely weights single-thread and multi-thread results equally, masks the EPYC’s extreme multi-core advantage.

The AMD Ryzen 9 PRO 6950HS, a mobile processor, scores 29466, giving it a -0.2% delta relative to the EPYC 9654. This mobile chip achieves parity in the average score despite having 16 threads versus the EPYC’s 192. The implication is that the average benchmark metric heavily favors single-thread performance, where the Ryzen 9 PRO’s higher boost clock compensates for its vastly lower core count.

Similarly, the AMD Ryzen 9 8945HS scores 29467, a -0.2% delta from the EPYC 9654. This is a modern laptop processor with 8 cores, and its near-identical average score to the 96-core EPYC underscores how the average benchmark score can be misleading for server parts. The data indicates that this metric does not reflect the EPYC’s true multi-threaded throughput, which is its primary purpose.

The Intel Core i7-12650H, a 10-core mobile processor, scores 29271, resulting in a +0.5% delta favoring the EPYC 9654. This is the only rival where the EPYC leads, but the margin is slim at 0.5%. The benchmark data shows that the EPYC 9654’s average score is within a 1% band of all four rivals, which are all consumer-oriented chips with far fewer cores, highlighting that the average score is not a meaningful comparison for this class of processor.

Who Should Consider It

The benchmark results indicate that the EPYC 9654 is purpose-built for multi-threaded server workloads, not for typical desktop or mobile tasks. In Cinebench R23 multi-core, the EPYC 9654 scores 101675, which is its most defining result. This score suggests that workloads such as large-scale virtualization, database processing, scientific simulations, and heavy compilation tasks, which can utilize all 192 threads, will see exceptional throughput. Any workload that can scale across dozens of cores will benefit directly from this processor.

For single-threaded tasks, the EPYC 9654 scores 14354 in Cinebench R23, which is respectable but not class-leading. Office productivity, web browsing, and light coding tasks will run adequately, but the processor’s design is not optimized for these. The data shows that its single-core performance is comparable to that of the mobile rivals listed, meaning that for a user primarily running single-threaded applications, a consumer chip would offer similar responsiveness at a fraction of the platform cost.

Creation workloads, such as video rendering or 3D animation, are a strong fit, provided the software is multi-threaded. The Cinebench R20 multi-core score of 42703 and the R15 multi-core score of 10248 both reinforce the processor’s dominance in heavily threaded rendering tasks. Conversely, gaming is not a recommended use case, as games typically rely on fewer cores and higher single-core clocks, where this processor’s 2.40 GHz base clock and 3.70 GHz boost clock are moderate by consumer standards.

Platform and Compatibility

The EPYC 9654 uses the AMD Socket SP5 platform, which is exclusive to the EPYC 9004 series. This socket supports the Zen 4 architecture and is paired with the Genoa codename. The processor requires a server-class motherboard designed for this socket, which is not interchangeable with consumer platforms.

Memory support is DDR5 with a twelve-channel memory bus, providing a peak memory bandwidth of 460.8 GB/s. This high-bandwidth configuration is critical for feeding the 96 cores and is a key differentiator from consumer platforms. ECC memory is supported, which is essential for data integrity in server environments where errors are unacceptable.

PCIe connectivity is provided through Gen 5 with 128 lanes from the CPU only. This extensive lane count allows for multiple high-speed accelerators, NVMe storage devices, and network interfaces to be connected directly to the processor. There is no integrated graphics, meaning a discrete GPU or a server management controller with its own video output is required for display.

The upgrade path is constrained to the EPYC 9004 series, as the SP5 socket is not forward-compatible with other generations. Within the series, users can select lower-core-count parts with the same platform, but moving to a different architecture would require a new motherboard and socket.

FAQ

Q: What is the core and thread count of the AMD EPYC 9654?

A: The EPYC 9654 has 96 cores and 192 threads.

Q: What is the boost clock speed of the EPYC 9654?

A: The boost clock is 3.70 GHz, with a base clock of 2.40 GHz.

Q: What type of memory does this processor support?

A: It supports DDR5 memory with a twelve-channel bus, achieving 460.8 GB/s of bandwidth.

Q: How many PCIe lanes does the CPU provide?

A: The EPYC 9654 provides 128 PCIe Gen 5 lanes from the CPU only.

Q: What is the TDP of the EPYC 9654?

A: The thermal design power is 360 W.

Q: When was the EPYC 9654 released?

A: The release date is 2022-11-09.

Benchmark Performance

The benchmark results for the EPYC 9654 show a clear pattern: multi-threaded performance is exceptional, while single-threaded performance is merely adequate. In Cinebench R23 multi-core, the score of 101675 is the standout figure. Comparing this to the average benchmark scores of its rivals, which are all below 29500, the multi-core result is over three times higher than any rival’s average. However, the rivals’ average scores are not directly comparable to a single multi-core run, so a more direct analysis uses the single-core scores.

In Cinebench R23 single-core, the EPYC 9654 scores 14354. The nearest rivals do not have listed single-core scores in the data, but their average scores suggest they are single-thread-competitive. The deltaPct values for the average scores are all within 0.5%, indicating that in a metric that balances single and multi-thread performance, the EPYC 9654 is statistically tied with the Ryzen 7 3800X, Ryzen 9 PRO 6950HS, Ryzen 9 8945HS, and Core i7-12650H.

This is a paradox: a 96-core processor with a 101675 multi-core score is rated as having the same average performance as an 8-core laptop chip. The data implies that the average benchmark score is dominated by single-thread results, which are similar across these processors. The single-core scores in Cinebench R15 (1446), R20 (6028), and R23 (14354) are all moderate, confirming that the EPYC 9654 does not excel in single-threaded tasks.

Single-Thread vs Multi-Thread Behavior

The split between single-thread and multi-thread performance is stark. The multi-core scores of 10248 (R15), 42703 (R20), and 101675 (R23) indicate that the processor scales exceptionally well across its 96 cores. These results are multiples of the single-core scores, demonstrating that the Zen 4 architecture and the 5 nm process node allow for efficient utilization of all threads.

The single-thread scores, however, are unremarkable. The 1446 in R15, 6028 in R20, and 14354 in R23 are consistent with a 3.70 GHz boost clock, which is modest compared to consumer processors that boost above 5 GHz. This means that for workloads that rely on a single thread, such as legacy software or lightly threaded games, the EPYC 9654 will perform no better than a mid-range desktop CPU from the same era.

For real-world workloads, this behavior implies that the processor is ideal for throughput-oriented tasks like batch processing, server consolidation, and parallel rendering. It is not suitable for latency-sensitive, single-threaded applications. The data suggests that users should match the workload to the processor’s strengths: if a task can use more than 32 threads, the EPYC 9654 will likely outperform any consumer chip; if it uses fewer than 8 threads, a consumer chip would be more cost-effective and faster.

Power and Thermals

The EPYC 9654 has a TDP of 360 W, which classifies it as a high-power server processor. This TDP reflects the energy required to operate 96 cores at their base clock of 2.40 GHz. The 5 nm process node from TSMC helps manage power efficiency, but the sheer core count necessitates substantial power delivery and thermal management.

A 360 W TDP implies the need for a robust cooling solution. Standard air coolers for consumer sockets are inadequate; the processor requires a server-grade heatsink and fan assembly, or liquid cooling, to maintain safe operating temperatures. The data does not specify thermal limits, but the TDP alone indicates that the cooling solution must be capable of dissipating significant heat continuously.

The platform’s power delivery system must also be designed for this TDP, with server motherboards for the SP5 socket featuring multiple power phases and high-current connectors. The 128 PCIe Gen 5 lanes and twelve-channel DDR5 memory further contribute to the platform’s power draw, as each memory channel and PCIe lane requires additional power. In a fully loaded server, the total system power consumption will be substantially higher than the CPU’s 360 W TDP, and the data implies that this processor is intended for data centers with adequate power and cooling infrastructure, not for typical desktop environments.