AMD EPYC 9474F vs AMD EPYC 9654P Comparison

The AMD EPYC 9654P and the AMD EPYC 9474F are both members of the EPYC 9004 series, built on the Zen 4 architecture and the Genoa codename. They share the same 5 nm process from TSMC, the same AMD Socket SP5, and identical memory and PCIe capabilities. Yet the recorded data reveals a clear performance hierarchy: the 9654P wins every head-to-head benchmark, while the 9474F counters with higher clock speeds and a higher average benchmark score across its own test set. This analysis digs into what those numbers mean for server and workstation workloads.

FAQ

Q: Which processor has more cores?

A: The AMD EPYC 9654P has 96 cores and 192 threads, while the AMD EPYC 9474F has 48 cores and 96 threads.

Q: What are the clock speed differences?

A: The 9654P has a base clock of 2.40 GHz and a boost clock of 3.70 GHz. The 9474F runs at a base of 3.60 GHz and boosts to 4.10 GHz.

Q: Which has larger L3 cache?

A: The 9654P has 384 MB of shared L3 cache, while the 9474F has 256 MB.

Q: How do they compare in Cinebench R23 multi-core?

A: The 9654P scores 98,875, the 9474F scores 86,790, giving the 9654P a 13.9% advantage.

Q: Which has a higher average benchmark score?

A: The 9474F has an average benchmark score of 25,103, while the 9654P averages 24,465. However, the 9474F's recorded set does not include Geekbench results, which the 9654P does.

Q: Are they on the same socket?

A: Yes, both use AMD Socket SP5.

Architecture Differences

Both processors are built on the same Zen 4 core architecture and the Genoa chiplet design, but they diverge significantly in core count, cache, and physical resources. The 9654P packs 96 cores and 192 threads, while the 9474F halves that to 48 cores and 96 threads. This core disparity is reflected in the transistor counts: the 9654P contains 78,840 million transistors across 12 chiplets of 72 mm² each, whereas the 9474F uses 52,560 million transistors on 8 chiplets of the same size. The L3 cache also scales with the core count: the 9654P offers 384 MB shared, the 9474F 256 MB. Both share the same per-core L1 (64 KB) and L2 (1 MB) caches.

Clock speeds tell a different story. The 9474F runs at a base of 3.60 GHz and boosts to 4.10 GHz, while the 9654P is slower at 2.40 GHz base and 3.70 GHz boost. This makes the 9474F the higher-clocked part, but the 9654P compensates with double the cores and 50% more L3. Both have the same 360 W TDP, twelve-channel DDR5 memory support with 460.8 GB/s bandwidth, and 128 PCIe Gen 5 lanes. The architecture is identical, so the differences are purely in resource allocation and clock tuning.

Head-to-Head Benchmarks

The head-to-head data covers six Cinebench tests: R15, R20, and R23, each with multi-core and single-core variants. In every single test, the 9654P wins by exactly 13.9%. For example, in Cinebench R23 multi-core, the 9654P scores 98,875 against the 9474F's 86,790. In R23 single-core, the 9654P scores 13,958 versus 12,252. The same 13.9% delta appears in R15 and R20, both multi and single core. This uniformity is striking: the 9654P's advantage is consistent across all recorded Cinebench workloads, regardless of thread count.

What makes this notable is that the 9474F has higher base and boost clocks. A 13.9% single-core win for the 9654P, despite a 0.4 GHz lower boost clock, suggests that the larger L3 cache or the core architecture's efficiency under these tests plays a decisive role. The data does not show any test where the 9474F overtakes the 9654P. The 9654P also has Geekbench scores (22,141 multi-core, 1,981 single-core) in the database, while the 9474F has no Geekbench results, so the head-to-head comparison is limited to Cinebench.

Specification Differences

The two processors differ in several key specifications. The 9654P has 96 cores and 192 threads, the 9474F has 48 cores and 96 threads. Base clock: 2.40 GHz versus 3.60 GHz. Boost clock: 3.70 GHz versus 4.10 GHz. Transistor count: 78,840 million versus 52,560 million. Die size: 12x 72 mm² versus 8x 72 mm². L3 cache: 384 MB versus 256 MB. The launch MSRP for the 9654P is $10,625, and for the 9474F it is $6,780. All other fields, including TDP (360 W), socket, memory support, PCIe lanes, and production status, are identical.

The Verdict

The recorded data is unambiguous: the AMD EPYC 9654P wins every head-to-head benchmark, and by a consistent 13.9% margin. This makes it the stronger performer in all Cinebench tests, both multi-threaded and single-threaded. The 9474F, despite its higher clock speeds, cannot match the 9654P in any recorded test. The 9654P's double core count and larger L3 cache appear to outweigh the clock disadvantage.

However, the 9474F does have a higher average benchmark score: 25,103 versus 24,465. This is based on its own set of six Cinebench tests, which excludes Geekbench. The 9654P's average includes two Geekbench scores, which are lower than its Cinebench results, pulling its average down. So the higher average for the 9474F is a consequence of the different test sets, not a sign of overall superiority. Both processors share the same 77th percentile ranking among all CPUs, indicating they sit at a similar performance tier in the broader database.

For a buyer choosing between these two, the 9654P is the clear winner in raw performance as measured by the head-to-head tests. The 9474F offers higher clocks and a lower launch MSRP, but the data shows no workload where it outperforms the 9654P. The 9654P is the better choice for any application that can leverage its 96 cores and 384 MB of L3 cache, which includes most server and workstation workloads.

Where Each One Wins

The 9654P wins all six head-to-head Cinebench tests, so it is the victor in every recorded comparison. The 9474F has zero head-to-head wins. That said, the 9474F does hold a higher average benchmark score (25,103) than the 9654P (24,465), but this is not a direct comparison because the test sets differ. The 9474F also has higher base and boost clocks, which could theoretically benefit lightly threaded tasks, but the single-core Cinebench results contradict that: the 9654P wins single-core by the same 13.9% margin.

In practical terms, the 9654P is the go-to for multi-threaded workloads like rendering, scientific computing, and virtualization, where its core count and cache size shine. The 9474F, with its higher clocks, might be considered for latency-sensitive applications that rely on frequency, but the data does not support that advantage. The 9474F's only recorded edge is its average benchmark score, which is an artifact of its missing Geekbench results. For any workload represented by the Cinebench suite, the 9654P is the definitive winner.