AMD EPYC 9475F vs AMD EPYC 9655P Comparison

The AMD EPYC 9655P and AMD EPYC 9475F are both 400 W parts in the EPYC 9005 series, sharing the same Turin codename and Zen 5 architecture on a 4 nm TSMC process. The data, however, shows two very different scaling philosophies: the 9655P pairs 96 Zen 5 cores with a 2.60 GHz base and 4.50 GHz boost clock, while the 9475F uses 48 cores at a 3.65 GHz base and 4.80 GHz boost. Benchmark results indicate a dominant performance split, with the 9655P winning all 11 head-to-head comparisons, though the 9475F remains competitive in single-threaded workloads, trailing by only 1.9% in PassMark single-thread tests.

FAQ

Q: Which processor has more physical cores and threads?

A: The AMD EPYC 9655P has 96 cores and 192 threads, exactly double the 48 cores and 96 threads found in the AMD EPYC 9475F.

Q: How much larger is the L3 cache on the 9655P?

A: The 9655P features 384 MB of shared L3 cache, which is 128 MB more than the 256 MB shared L3 cache on the 9475F.

Q: Are there differences in memory bandwidth or PCIe lane support?

A: No. Both processors support DDR5 memory over a Twelve-channel bus with a memory bandwidth of 576.0 GB/s, and both provide Gen 5 PCIe with 128 Lanes (CPU only).

Q: What is the launch MSRP for each processor?

A: The AMD EPYC 9655P has a launch MSRP of $10811, while the AMD EPYC 9475F has a launch MSRP of $7592.

Q: Which processor has the higher boost clock?

A: The AMD EPYC 9475F has a higher boost clock at 4.80 GHz, compared to the 9655P's 4.50 GHz boost clock.

Q: In how many head-to-head benchmark comparisons did each processor win?

A: The AMD EPYC 9655P won all 11 head-to-head benchmark comparisons, while the AMD EPYC 9475F won none.

Architecture Differences

Both processors are built on the same foundational architecture: Zen 5, codename Turin, on a 4 nm process node from TSMC. They share the same AMD Socket SP5, integrated graphics are N/A for both, and neither has an unlocked multiplier. The core counts diverge sharply, with the 9655P integrating 96 cores versus 48 on the 9475F. This core difference drives a corresponding difference in transistor count, as the 9655P packs 99,780 million transistors across 12x 70.6 mm² dies, while the 9475F uses 66,520 million transistors across 8x 70.6 mm² dies.

Cache hierarchy also scales with core count. Both parts use 80 KB of L1 per core and 1 MB of L2 per core, but the shared L3 cache jumps from 256 MB on the 9475F to 384 MB on the 9655P. The clock strategy is reversed: the 9475F runs a higher base clock of 3.65 GHz and a higher boost clock of 4.80 GHz, whereas the 9655P operates at a 2.60 GHz base and 4.50 GHz boost. Both maintain the same 400 W TDP, which suggests the 9475F's higher clocks come at the cost of core density. Memory support is identical—DDR5, Twelve-channel, 576.0 GB/s bandwidth, and ECC memory support—as is the PCIe configuration of Gen 5 with 128 lanes. The production status, release date, and market segment (Server/Workstation) are also identical, as both launched on 2024-10-09.

Head-to-Head Benchmarks

The PassMark benchmark suite shows a consistent pattern of 9655P dominance, with the largest margins appearing in multi-threaded and math-intensive workloads. The most significant victory comes in integer math, where the 9655P scores 1,225,251 against the 9475F's 605,696, a 102.3% advantage—effectively doubling the 9475F's throughput. Floating-point math follows a similar trend, with the 9655P at 715,866 versus 406,524, a 76.1% lead. Data encryption shows an 88.7% delta (220,074 vs 116,648), and random string sorting is 77.9% faster on the 9655P (451,824 vs 253,936). These results indicate that the 9655P's extra cores provide near-linear scaling for parallel integer and floating-point workloads.

The 9655P also wins in multi-threaded benchmarks, scoring 160,490 in PassMark multithread versus 122,476 for the 9475F, a 31% advantage. Physics simulations show a 57.2% lead (25,847 vs 16,443), and extended instructions are 33.2% faster (230,609 vs 173,169). Data compression results favor the 9655P by 61.7% (3,486,158 vs 2,156,305). The smallest margins appear in single-threaded tests: the 9655P scores 3,849 versus 3,779 in PassMark single-thread, a 1.9% edge, and a similar 1.9% delta in the singlethread test. The 9655P also edges out the 9475F in prime number finding, 1,686 vs 1,507, an 11.9% win. Across all 11 comparisons, the 9655P wins decisively, but the 9475F's higher boost clock keeps single-thread performance nearly level.

Specification Differences

The two processors differ in several key specifications, all traceable to the core count and clock trade-off. The 9655P has 96 cores and 192 threads, while the 9475F has 48 cores and 96 threads. Base clock differs by 1.05 GHz, with the 9475F at 3.65 GHz and the 9655P at 2.60 GHz; boost clock also favors the 9475F at 4.80 GHz versus 4.50 GHz. Transistor count is 33,260 million higher on the 9655P (99,780 million vs 66,520 million), and die count is 12x 70.6 mm² versus 8x 70.6 mm². L3 cache is 384 MB on the 9655P versus 256 MB on the 9475F. Launch MSRP differs as well, with the 9655P at $10811 and the 9475F at $7592.

These are the only differing fields. Both processors share the identical TDP of 400 W, the same AMD Socket SP5, the same Zen 5 architecture and Turin codename, the same 4 nm TSMC process, the same L1 and L2 cache sizes per core, the same memory support and bandwidth, the same PCIe lane count, and the same release date and production status. The part numbers also differ: 100-000001522 for the 9655P and 100-000001143 for the 9475F.

Where Each One Wins

The 9655P is the clear winner for heavy multi-threaded and parallel workloads. Its 96 cores deliver a 102.3% advantage in integer math, a 76.1% lead in floating-point math, and a 88.7% edge in data encryption, making it the appropriate choice for database compression, scientific computing, and encryption-heavy server tasks. The 31% multithread win and the 61.7% data compression win further reinforce its suitability for high-throughput data centers running many concurrent threads. For workloads that scale with core count—such as virtualization, large-scale data processing, and high-performance computing—the 9655P's benchmark results are unambiguous.

The 9475F wins in no benchmark comparisons, but its higher clocks (3.65 GHz base, 4.80 GHz boost) make it a viable option for single-threaded or lightly threaded applications where the 9655P's 1.9% single-thread lead is negligible. In workloads that are latency-sensitive or cache-bound per core, the 9475F's 48 cores with identical per-core L1 and L2 cache, combined with a lower core count, could reduce contention. However, the data shows no scenario where the 9475F outperforms the 9655P, so its use case is limited to situations where the lower launch MSRP is a factor and the core count is sufficient. For mixed workloads, the 9655P's 384 MB L3 cache and 192 threads provide a safety margin that the 9475F cannot match, despite the latter's frequency advantage. Ultimately, the 9655P is the performance leader across every measured metric, while the 9475F serves as a higher-clocked, lower-core-density alternative for power- or cost-conscious deployments that do not require maximum parallelism.