AMD EPYC 9655 vs AMD EPYC 9684X Comparison

Head-to-Head Benchmarks

The benchmark data records 17 head-to-head comparisons between the AMD EPYC 9655 and the AMD EPYC 9684X. The EPYC 9655 wins 16 of those tests, while the EPYC 9684X takes a single victory. The overall pattern is consistent: the 9655 is faster in nearly every measured workload, often by a wide margin.

The most decisive win for the EPYC 9655 comes in PassMark floating point math, where it scores 662,958 against 472,174, a 40.4% advantage. That is the largest delta in the entire comparison. Integer math follows closely, with the 9655 scoring 1,139,161 versus 844,145, a 34.9% lead. Single-thread performance also shows a substantial gap: the 9655 records 3,847 in PassMark single-thread tests, 33.1% ahead of the 9684X's 2,891.

The Cinebench suite shows a uniform 28.4% delta across all six tests. In Cinebench R23 multicore, the 9655 scores 132,672 against 103,355. In R23 single-core, it scores 18,730 versus 14,591. The same 28.4% gap appears in R15 and R20, both multicore and single-core. This uniformity suggests a consistent architectural advantage rather than workload-specific tuning.

Data compression and encryption also favor the 9655, though by smaller margins. Compression sees a 21.2% lead (3,271,896 vs. 2,698,807), while encryption shows an 18% advantage (210,555 vs. 178,453). Extended instructions are 14.2% faster on the 9655 (203,285 vs. 177,956). Random string sorting favors the 9655 by 25.9% (439,682 vs. 349,126).

The 9684X's lone win is in PassMark find prime numbers, where it scores 2,020 versus 1,598, a 20.9% advantage. This is a notable result because it is the only test where the older processor outperforms the newer one, and the margin is substantial. The 9655 still wins the multithread test by 28.4% (156,110 vs. 121,595) and physics by a narrower 5.1% (25,947 vs. 24,686).

Where Each One Wins

The EPYC 9655 dominates across the board, but the magnitude of its wins varies by workload class. The largest advantages appear in math-heavy tests: floating point, integer math, and single-thread performance. These are workloads where clock speed and per-core efficiency matter most. The 9655 boosts to 4.50 GHz, while the 9684X tops out at 3.70 GHz, and the data reflects that gap directly.

For multithreaded rendering and encoding, the Cinebench results show a consistent 28.4% lead for the 9655. This is a broad-based win that covers both multicore and single-core scenarios, meaning the 9655 does not just scale better with more threads; it is also faster per thread. The PassMark multithread score confirms this, with the same 28.4% delta.

Memory-sensitive workloads favor the 9655 as well. The 9655 has a recorded memory bandwidth of 576.0 GB/s, compared to 460.8 GB/s for the 9684X. Data compression and random string sorting, both of which are memory-intensive, show 21.2% and 25.9% leads respectively. Encryption, which can be sensitive to both memory bandwidth and instruction efficiency, shows an 18% lead.

The 9684X wins only in prime number calculation. This is a niche workload that relies heavily on integer division and branch prediction rather than raw clock speed or memory bandwidth. The 9684X's 2,020 score against 1,598 suggests that its Zen 4 architecture handles this specific pattern better, despite being slower in almost everything else. For users whose primary workload is prime number searching, the 9684X is the better choice. For everything else in the recorded data, the 9655 is faster.

Architecture Differences

The two processors share the same socket, AMD Socket SP5, and the same memory type, DDR5 with twelve-channel support. Both have 96 cores and 192 threads, and both run at a 400 W TDP. The differences lie deeper in the architecture.

The 9655 is built on Zen 5 architecture with the codename Turin, using a 4 nm process from TSMC. The 9684X uses Zen 4 architecture with the codename Genoa-X, on a 5 nm process. The 9655 has a smaller transistor count at 99,780 million, while the 9684X packs 135,240 million transistors. Die size is similar: 12x 70.6 mm² for the 9655 versus 12x 72 mm² for the 9684X. The smaller process node and lower transistor count for the 9655 indicate a denser, more efficient design.

Cache configuration is a major differentiator. The 9655 has 80 KB of L1 cache per core and 1 MB of L2 per core, with 384 MB of shared L3. The 9684X has 64 KB of L1 per core and 1 MB of L2 per core, but its shared L3 is three times larger at 1152 MB. That massive L3 pool is the defining feature of the 9684X, designed for workloads that fit large datasets in cache.

Clock speeds differ significantly. The 9655 has a base clock of 2.60 GHz and a boost clock of 4.50 GHz. The 9684X runs at 2.55 GHz base and 3.70 GHz boost. The 9655's boost clock is 0.80 GHz higher, which explains much of its single-thread advantage.

Memory bandwidth also differs: the 9655 records 576.0 GB/s, while the 9684X records 460.8 GB/s. Both support ECC memory and PCIe Gen 5 with 128 lanes. The 9655 has no integrated graphics, and the 9684X's integrated graphics field is not recorded. Both are active production parts, with the 9655 released in October 2024 and the 9684X in June 2023.

FAQ

Q: Which processor has the higher boost clock?

A: The AMD EPYC 9655 boosts to 4.50 GHz, while the AMD EPYC 9684X boosts to 3.70 GHz. The 9655 also has a slightly higher base clock at 2.60 GHz versus 2.55 GHz.

Q: How much larger is the L3 cache on the 9684X?

A: The 9684X has 1152 MB of shared L3 cache, which is three times the 384 MB found on the 9655. Both have 1 MB of L2 per core, but the 9655 has 80 KB of L1 per core versus 64 KB on the 9684X.

Q: What is the memory bandwidth difference?

A: The 9655 records 576.0 GB/s of memory bandwidth, while the 9684X records 460.8 GB/s. That is a 25% higher bandwidth figure for the 9655, which contributes to its wins in compression and string sorting tests.

Q: In which test does the 9684X beat the 9655?

A: The 9684X wins only in PassMark find prime numbers, scoring 2,020 against the 9655's 1,598, a 20.9% advantage. In every other recorded head-to-head test, the 9655 is faster.

Q: Are both processors compatible with the same socket and memory?

A: Yes, both use AMD Socket SP5 and support DDR5 memory with twelve-channel memory buses. Both also support ECC memory and PCIe Gen 5 with 128 lanes.

Q: What is the single-thread performance gap?

A: In PassMark single-thread tests, the 9655 scores 3,847 versus 2,891 for the 9684X, a 33.1% lead. The Cinebench R23 single-core test shows a 28.4% gap, with the 9655 at 18,730 and the 9684X at 14,591.

The Verdict

The data is unambiguous. The AMD EPYC 9655 outperforms the AMD EPYC 9684X in 16 of 17 recorded benchmarks. Its advantages range from a modest 5.1% in physics to a commanding 40.4% in floating point math. The 9655 leads by 28.4% across all Cinebench tests, 33.1% in single-thread performance, and 34.9% in integer math. It also has a higher boost clock, more memory bandwidth, and a smaller process node.

The 9684X retains one clear strength: its 1152 MB L3 cache. This gives it a 20.9% win in prime number calculation, a workload that benefits from large cache residency. For any application that repeatedly scans a dataset larger than 384 MB but smaller than 1152 MB, the 9684X could be competitive. However, the recorded benchmark suite shows that this advantage does not translate to broader wins.

The 9655 is the better choice for general server workloads, rendering, compression, encryption, and math-heavy tasks. The 9684X is the better choice only for the specific pattern of prime number searching, where its cache size overcomes its lower clock speed. Given that the 9655 also records a higher average benchmark score of 373,479 versus 266,914 for the 9684X, and both parts carry a 400 W TDP, the 9655 delivers more performance per watt in nearly every measured scenario.

The 9655 is released more recently, and its launch MSRP is $11852. The 9684X had a launch MSRP of $14756. The data favors the newer part on both performance and launch price, though neither of these figures should be interpreted as a purchasing recommendation on price alone.

Specification Differences

| Specification | AMD EPYC 9655 | AMD EPYC 9684X |

|---|---|---|

| Series | EPYC 9005 series | EPYC 9004 series |

| Architecture | Zen 5 | Zen 4 |

| Codename | Turin | Genoa-X |

| Generation | EPYC (Zen 5 (Turin)) | EPYC (Zen 4 (Genoa)) |

| Process Node | 4 nm | 5 nm |

| Transistors | 99,780 million | 135,240 million |

| Die Size | 12x 70.6 mm² | 12x 72 mm² |

| Base Clock | 2.60 GHz | 2.55 GHz |

| Boost Clock | 4.50 GHz | 3.70 GHz |

| L1 Cache | 80 KB (per core) | 64 KB (per core) |

| L3 Cache | 384 MB (shared) | 1152 MB (shared) |

| Memory Bandwidth | 576.0 GB/s | 460.8 GB/s |

| Release Date | 2024-10-09 | 2023-06-12 |

| Launch MSRP | $11852 | $14756 |

| Part Number | 100-000000674 | 100-100000892 |

Both processors share the same core count (96), thread count (192), TDP (400 W), socket (AMD Socket SP5), memory type (DDR5), memory bus (twelve-channel), ECC support, PCIe configuration (Gen 5, 128 lanes), and market segment (Server/Workstation). The 9655 has a higher average benchmark score of 373,479 and a 100th percentile ranking, while the 9684X has an average of 266,914 and a 99th percentile ranking.