AMD EPYC 9655P vs AMD EPYC 9754 Comparison

The AMD EPYC 9754 and AMD EPYC 9655P are both server-class processors built for AMD Socket SP5, yet they target different performance profiles. The 9754 is a 128-core Zen 4c part, while the 9655P is a 96-core Zen 5 design. The benchmark data reveals a clear split: the 9655P dominates nearly every workload, but the 9754 holds specific wins in data compression and encryption. This analysis examines the head-to-head results, architectural differences, and what the numbers imply for different deployment scenarios.

Head-to-Head Benchmarks

The EPYC 9655P wins 15 of the 17 head-to-head comparisons, and its victories are often decisive. In Cinebench R23 multi-core, the 9655P scores 136,354 against the 9754’s 83,939, a delta of -38.4% from the 9754’s perspective. That same -38.4% delta repeats across Cinebench R15 multi-core (13,744 vs 8,460), R20 multi-core (57,268 vs 35,254), and PassMark multithread (160,417 vs 98,752). The consistency of that delta across all multi-threaded Cinebench versions suggests a fundamental throughput advantage, not a workload-specific quirk.

Single-threaded performance shows an even larger gap. In Cinebench R23 single-core, the 9655P scores 19,250 versus 11,850, a -38.4% delta. PassMark single-thread results show the 9655P at 3,848 versus 2,328, a -39.5% delta. The 9655P’s 4.50 GHz boost clock compared to the 9754’s 3.10 GHz explains much of this, but the architecture generation difference also contributes.

The 9655P’s largest wins come in specific compute-heavy tests. In PassMark physics, it scores 26,810 versus 8,793, a -67.2% delta. PassMark find prime numbers shows 1,683 versus 604, a -64.1% delta. These aren’t marginal improvements; the 9655P is more than 2.5x faster in physics and nearly 2.8x faster in prime number finding. Floating-point math also favors the 9655P (710,260 vs 588,187, -17.2%), as does integer math (1,218,189 vs 1,026,896, -15.8%). Random string sorting goes to the 9655P at 455,310 versus 306,481, a -32.7% delta.

The 9754 does register wins, but they are narrow. In PassMark data compression, the 9754 scores 3,558,043 versus 3,478,283, a 2.3% advantage. In PassMark data encryption, it scores 231,891 versus 219,606, a 5.6% edge. These two wins suggest the 9754’s higher core count (128 vs 96) helps in specific memory-bound or parallel-friendly tasks where the 9655P’s architectural efficiency doesn’t fully compensate. However, even in extended instructions, the 9655P edges ahead (227,538 vs 224,322, -1.4%), showing that the 9754’s victories are isolated.

Looking at the aggregate benchmark score, the 9655P averages 396,673 versus the 9754’s 364,371. The nearest-rival data confirms this ordering: the 9655P sits 8.9% above the 9754 in average score, while the 9754 trails the 9655P by -8.1% from the 9655P’s perspective. Both CPUs sit at the 100th percentile against all CPUs, meaning they are top-tier parts, but the 9655P is clearly the stronger performer overall.

Architecture Differences

The two processors come from different design generations. The 9754 uses Zen 4 architecture with the Bergamo codename, while the 9655P uses Zen 5 with the Turin codename. This generational jump is visible in the process node: the 9754 is built on a 5 nm process, whereas the 9655P uses a 4 nm process. Both are fabricated by TSMC, but the transistor counts differ dramatically—the 9754 packs 71,000 million transistors across 8x 73 mm² dies, while the 9655P contains 99,780 million transistors across 12x 70.6 mm² dies.

Core and cache configurations also diverge. The 9754 has 128 cores and 256 threads, while the 9655P has 96 cores and 192 threads. Despite having fewer cores, the 9655P’s L1 cache is larger per core at 80 KB versus 64 KB, and its L3 cache is 384 MB shared versus 256 MB shared. Both have 1 MB of L2 per core. The 9655P’s base clock is 2.60 GHz versus 2.25 GHz, and its boost clock reaches 4.50 GHz versus 3.10 GHz. The 9655P also draws more power, with a TDP of 400 versus 360.

Memory bandwidth favors the 9655P as well. Both support DDR5 with a twelve-channel memory bus, but the 9655P’s bandwidth is 576.0 GB/s versus 460.8 GB/s. PCIe lanes are identical at Gen 5, 128 lanes (CPU only). Neither part has integrated graphics, and both are locked (multiplier unlocked: false). The 9655P was released later, but the exact date is not relevant to the performance data.

The process node shrink and larger die count suggest the 9655P uses a more complex layout, likely enabling higher clocks and better per-core efficiency. The 9754’s higher core count with a smaller process node suggests a density-optimized design, trading clock speed for more parallel throughput.

The Verdict

The data is unambiguous: the EPYC 9655P is the superior performer in almost every benchmark category. Its 38.4% lead across all Cinebench multi-core tests is not a fluke—it reflects a fundamental advantage in sustained multi-threaded throughput. The 9655P also wins decisively in single-threaded tests (39.5% in PassMark single-thread), physics (67.2%), and prime number finding (64.1%). For workloads that depend on raw compute, the 9655P is the clear choice.

The 9754’s wins are real but limited. A 2.3% edge in data compression and a 5.6% edge in data encryption suggest that the 128-core configuration provides a slight benefit in specific memory-oriented or cryptographic tasks. However, these margins are small enough that they may not justify the 9754’s lower performance elsewhere. The 9754’s aggregate benchmark score of 364,371 is 8.1% below the 9655P’s 396,673, and the 9655P is also 6.2% above the non-P EPYC 9655, indicating it is the stronger of the two 9655 variants.

Who should pick which? The 9655P is the default recommendation for most server workloads—database processing, scientific computing, virtualization, and any task that benefits from high single-thread and multi-thread performance. The 9754 could be considered for environments where data compression or encryption is the dominant workload and where the extra 32 cores provide a measurable, albeit small, benefit. But even then, the 9655P’s 15-2 win record suggests the 9754 is a niche part.

Specification Differences

| Specification | AMD EPYC 9754 | AMD EPYC 9655P |

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

| Cores | 128 | 96 |

| Threads | 256 | 192 |

| Base Clock | 2.25 GHz | 2.60 GHz |

| Boost Clock | 3.10 GHz | 4.50 GHz |

| TDP | 360 | 400 |

| Architecture | Zen 4 | Zen 5 |

| Codename | Bergamo | Turin |

| Process Node | 5 nm | 4 nm |

| Transistors | 71,000 million | 99,780 million |

| Die Size | 8x 73 mm² | 12x 70.6 mm² |

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

| L2 Cache | 1 MB (per core) | 1 MB (per core) |

| L3 Cache | 256 MB (shared) | 384 MB (shared) |

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

| Launch MSRP | $11900 | $10811 |

Both CPUs share the same socket (AMD Socket SP5), memory support (DDR5, twelve-channel), ECC support, PCIe configuration (Gen 5, 128 lanes), market segment (Server/Workstation), and production status (Active). Neither has integrated graphics or an unlocked multiplier.

FAQ

Q: Which CPU has more cores?

A: The AMD EPYC 9754 has 128 cores and 256 threads, while the AMD EPYC 9655P has 96 cores and 192 threads.

Q: Why does the 9655P win most benchmarks despite having fewer cores?

A: The 9655P uses Zen 5 architecture on a 4 nm process, with higher base (2.60 GHz) and boost (4.50 GHz) clocks, larger L1 per core (80 KB), and more L3 cache (384 MB). These factors offset its lower core count.

Q: In which benchmarks does the 9754 beat the 9655P?

A: The 9754 wins in PassMark data compression (3,558,043 vs 3,478,283, a 2.3% edge) and PassMark data encryption (231,891 vs 219,606, a 5.6% edge).

Q: What is the performance gap in multi-core workloads?

A: The 9655P leads by 38.4% in all Cinebench multi-core tests (R15, R20, R23) and PassMark multithread, with scores of 13,744 vs 8,460 (R15), 57,268 vs 35,254 (R20), 136,354 vs 83,939 (R23), and 160,417 vs 98,752 (PassMark).

Q: How do their memory bandwidths compare?

A: The 9655P has a memory bandwidth of 576.0 GB/s, while the 9754 has 460.8 GB/s. Both use DDR5 with a twelve-channel bus.

Q: Which CPU has a higher average benchmark score?

A: The 9655P has an average benchmark score of 396,673, compared to the 9754’s 364,371. The 9655P is 8.9% above the 9754 in the nearest-rival comparison.

Where Each One Wins

The 9655P wins in 15 of 17 head-to-head tests, making it the dominant part for general compute. Its largest victories are in physics (67.2% faster), prime number finding (64.1% faster), and single-threaded workloads (39.5% faster). These results point to workloads like simulation, scientific modeling, financial risk analysis, and any application with heavy branching or sequential dependencies. The 9655P also excels in floating-point math (17.2% faster) and integer math (15.8% faster), covering most numerical computing scenarios. For virtualized environments or containerized microservices that rely on high per-core throughput, the 9655P’s clock speed advantage is decisive.

The 9754’s wins are confined to data compression (2.3% faster) and data encryption (5.6% faster). These are tasks that scale with core count and memory bandwidth, and the 9754’s 128 cores provide a slight edge. However, the margins are small—under 6%—so the 9754 is only preferable if these specific tasks are the primary workload and the 9655P’s other advantages are irrelevant. The 9754’s lower TDP (360 vs 400) might also be a consideration in power-constrained racks, but that is not a performance metric.

For most buyers, the 9655P is the data-driven choice. Its aggregate score is 8.9% higher than the 9754’s, and it wins every Cinebench test by a wide margin. The 9754 remains a viable option only for niche encryption or compression pipelines where its 2-5% advantage matters more than the 38% deficit in multi-core rendering or simulation tasks. The benchmark data makes the tradeoff clear: the 9655P is the performance king, while the 9754 is a specialized density play.