AMD EPYC 9655P vs AMD EPYC 9684X Comparison

Head-to-Head Benchmarks

The benchmark data presents a decisive overall picture: the AMD EPYC 9655P wins 16 of the 17 recorded head-to-head comparisons, with the AMD EPYC 9684X taking a single victory. The magnitude of the 9655P's wins is substantial across nearly every workload category.

In Cinebench tests, the 9655P is consistently about 31.9% ahead in both multi-core and single-core workloads. The R15 multi-core score is 13744 versus 10418, while R15 single-core shows 1940 against 1470, a 32% advantage. The R20 results mirror this pattern exactly at 31.9% in both tests, with scores of 57268 versus 43409 and 8085 versus 6128 respectively. R23 multi-core records 136354 versus 103355, and single-core reaches 19250 versus 14591, again a 31.9% margin. This uniformity across Cinebench versions suggests a consistent architectural performance advantage rather than workload-specific behavior.

PassMark results widen the gap further in several areas. Floating point math shows the largest difference, with the 9655P scoring 715866 against 472174 for the 9684X, a 51.6% advantage. Integer math follows at 45.1% ahead, scoring 1225251 versus 844145. Single-thread performance is 33.1% higher at 3849 versus 2891, and multi-thread performance is 32% higher at 160490 versus 121595. Data compression favors the 9655P by 29.2% (3486158 versus 2698807), while extended instructions show a 29.6% edge (230609 versus 177956). Random string sorting is 29.4% faster at 451824 versus 349126, and data encryption is 23.3% ahead at 220074 versus 178453.

The sole win for the 9684X comes in the find prime numbers test, where it scores 2020 against 1686 for the 9655P, a 16.5% advantage. This is notable because it is the only test where the older processor demonstrates superiority, and it indicates a specific sensitivity to the workload's characteristics. The physics test is the closest overall race, with the 9655P winning by just 4.7% (25847 versus 24686).

Average benchmark scores reinforce this hierarchy. The 9655P holds an average score of 397773, while the 9684X averages 266914. The 9655P sits at the 100th percentile among all CPUs in the database, while the 9684X sits at the 99th. The 9655P's nearest rival, the AMD Ryzen Threadripper PRO 9995WX, averages 412068, which is 3.5% higher than the 9655P. The 9684X's nearest rival, the Intel Xeon 6780E, averages 280438, which is 4.8% higher than the 9684X. These rival comparisons show that both processors are positioned within a competitive field, though the gap between the two EPYC parts themselves is considerable.

Architecture Differences

The architectural divide between these two processors is generational. The EPYC 9655P belongs to the EPYC 9005 series, uses Zen 5 architecture, and carries the codename Turin. It is built on a 4 nm process at TSMC with a claimed 99,780 million transistors spread across 12 dies, each measuring 70.6 mm². The EPYC 9684X belongs to the EPYC 9004 series, uses Zen 4 architecture, and carries the codename Genoa-X. It is built on a 5 nm process at TSMC with 135,240 million transistors across 12 dies, each measuring 72 mm². The newer process node and smaller die size give the 9655P a density advantage, although the 9684X has a higher total transistor count.

Cache configurations differ dramatically. The 9655P provides 80 KB of L1 cache per core, 1 MB of L2 cache per core, and 384 MB of shared L3 cache. The 9684X provides 64 KB of L1 per core, the same 1 MB of L2 per core, but a much larger 1152 MB of shared L3 cache. This threefold L3 difference is the defining feature of the 9684X, which leverages the larger cache for workloads that benefit from high-capacity on-die storage. However, the recorded benchmarks show that this cache advantage does not translate into widespread performance wins against the 9655P.

Clock speeds also favor the newer part. The 9655P has a base clock of 2.60 GHz and a boost clock of 4.50 GHz. The 9684X runs at a 2.55 GHz base and a 3.70 GHz boost. The 0.80 GHz boost advantage for the 9655P is significant for single-threaded and lightly threaded workloads, and it helps explain the consistent single-core Cinebench wins. Both processors share a 400 W TDP, the AMD Socket SP5 platform, DDR5 memory support, twelve-channel memory buses, and PCIe Gen 5 with 128 lanes on the CPU. Both support ECC memory and have no integrated graphics.

Memory bandwidth differs as well. The 9655P delivers 576.0 GB/s, while the 9684X delivers 460.8 GB/s. For memory-intensive server workloads, this higher bandwidth gives the 9655P an additional edge beyond raw compute. The 9684X launched on 2023-06-12, while the 9655P launched later on 2024-10-09. The 9655P has a launch MSRP of $10811, and the 9684X has a launch MSRP of $14756. Both processors are currently active in production, and neither has an unlocked multiplier. The part numbers differ (100-000001522 for the 9655P and 100-100000892 for the 9684X), and the market segment for both is server and workstation.

Where Each One Wins

The 9655P is the clear choice for the majority of compute-heavy scenarios based on the recorded data. Its floating point and integer math advantages of 51.6% and 45.1% respectively make it the stronger option for scientific computing, financial modeling, and any workload with heavy arithmetic throughput. The 31.9% multi-core Cinebench advantage indicates superior scaling across all cores, which matters for rendering, simulation, and batch processing. Single-thread performance is also 33.1% higher, so workloads with serial components or low thread counts will see meaningful gains. Data compression and encryption workloads favor the 9655P by 29.2% and 23.3% respectively, making it the better fit for storage servers, database compression tasks, and secure communication infrastructure. The higher memory bandwidth of 576.0 GB/s versus 460.8 GB/s further supports memory-bound applications.

The 9684X has a narrow but real niche. Its 16.5% win in the find prime numbers test suggests that certain algorithmic patterns, likely those with heavy cache reuse and branch-heavy loops, can benefit from its 1152 MB L3 cache. Workloads that fit large working sets into cache without needing repeated memory fetches could see an advantage, even though the overall benchmark suite shows the 9655P winning broadly. The physics test is close at 4.7%, indicating that some mixed workloads are nearly a wash. The 9684X also has a higher transistor count at 135,240 million, which may matter for specific instruction-level behaviors, though the benchmark data does not reveal a broad benefit from this.

For general server deployment, the 9655P dominates in the recorded metrics. Its average benchmark score of 397773 is 49% higher than the 9684X's 266914, and its 100th percentile ranking versus the 99th for the 9684X places it at the top of the database. The 9684X should be considered only for specialized cache-sensitive workloads where its 1152 MB L3 cache provides a concrete advantage, as demonstrated by the prime numbers result. In every other measured category, the 9655P is faster, often by a wide margin.

FAQ

Q: Which processor is faster in multi-core Cinebench tests?

A: The AMD EPYC 9655P wins all multi-core Cinebench tests by 31.9%. R15 multi-core scores 13744 versus 10418, R20 scores 57268 versus 43409, and R23 scores 136354 versus 103355.

Q: Does the larger L3 cache of the EPYC 9684X help it win any benchmarks?

A: Yes, the 9684X wins the passmark find prime numbers test with a score of 2020 versus 1686 for the 9655P, a 16.5% advantage. This suggests the 1152 MB L3 cache helps in specific cache-sensitive workloads.

Q: How do the two processors compare in single-thread performance?

A: The 9655P is 33.1% ahead in passmark single-thread (3849 versus 2891) and 32% ahead in Cinebench R15 single-core (1940 versus 1470). The boost clock difference of 4.50 GHz versus 3.70 GHz contributes to this gap.

Q: What are the memory bandwidth capabilities of each processor?

A: The 9655P provides 576.0 GB/s of memory bandwidth, while the 9684X provides 460.8 GB/s. Both support DDR5 memory with a twelve-channel memory bus.

Q: What is the average benchmark score for each processor?

A: The 9655P has an average benchmark score of 397773, while the 9684X has an average of 266914. The 9655P ranks at the 100th percentile among all CPUs, and the 9684X ranks at the 99th.

Q: Which processor has a higher process node?

A: The 9655P is built on a 4 nm process at TSMC, while the 9684X is built on a 5 nm process at TSMC. The 9655P uses 99,780 million transistors across 12 dies of 70.6 mm² each, while the 9684X uses 135,240 million transistors across 12 dies of 72 mm² each.

Specification Differences

The two processors differ in several key specifications. The 9655P uses Zen 5 architecture with the Turin codename from the EPYC 9005 series, while the 9684X uses Zen 4 architecture with the Genoa-X codename from the EPYC 9004 series. The process node is 4 nm for the 9655P and 5 nm for the 9684X. Transistor counts are 99,780 million for the 9655P and 135,240 million for the 9684X, with die sizes of 12x 70.6 mm² and 12x 72 mm² respectively.

Base clock speeds are 2.60 GHz for the 9655P and 2.55 GHz for the 9684X. Boost clocks are 4.50 GHz and 3.70 GHz respectively. L1 cache is 80 KB per core on the 9655P and 64 KB per core on the 9684X. L2 cache is identical at 1 MB per core. L3 cache is 384 MB shared on the 9655P and 1152 MB shared on the 9684X. Memory bandwidth is 576.0 GB/s on the 9655P and 460.8 GB/s on the 9684X. Release dates are 2024-10-09 for the 9655P and 2023-06-12 for the 9684X. Launch MSRP values are $10811 for the 9655P and $14756 for the 9684X. Part numbers are 100-000001522 for the 9655P and 100-100000892 for the 9684X. Both processors share 96 cores, 192 threads, 400 W TDP, AMD Socket SP5, DDR5 memory support, twelve-channel memory buses, ECC memory support, PCIe Gen 5 with 128 lanes, no integrated graphics, no unlocked multiplier, active production status, and the server/workstation market segment.