AMD EPYC 7F72 vs AMD EPYC 9135 Comparison

The AMD EPYC 7F72 and AMD EPYC 9135 are both 96th-percentile server processors, yet the benchmark data reveals they are engineered for fundamentally different workloads. The 7F72, a 24-core Zen 2 part from the Rome generation, holds an average benchmark score of 85,072, while the 9135, a 16-core Zen 5 part from the Turin generation, posts an average of 82,980. The head-to-head results show the 9135 winning 12 of 17 tests, but the 7F72’s five victories are decisive in specific, latency-sensitive tasks. This analysis walks through the exact scores, deltas, and architectural implications.

Head-to-Head Benchmarks

The most striking pattern in the data is the 9135’s dominance in Cinebench across every version. In Cinebench R15 multicore, the 9135 scores 4,952 against the 7F72’s 4,518, an 8.8% advantage. The same margin appears in R15 single-core (699 vs 637, 8.9% delta) and persists through R20 multicore (20,637 vs 18,828, 8.8%) and R20 single-core (2,913 vs 2,657, 8.8%). Cinebench R23 shows identical proportionality: multicore 49,136 vs 44,829 (8.8%) and single-core 6,936 vs 6,328 (8.8%). This consistency suggests a per-clock efficiency advantage in the 9135’s Zen 5 architecture rather than a raw core-count effect.

The PassMark suite tells a more complex story. The 9135 wins PassMark single-thread by a massive 35.1% margin (3,672 vs 2,384), which is the largest delta in either direction across all tests. It also leads in floating-point math (126,679 vs 108,437, 14.4% ahead), integer math (202,962 vs 181,103, 10.8% ahead), and extended instructions (55,822 vs 46,936, 15.9% ahead). The multithread score favors the 9135 by 7.7% (57,170 vs 52,740), which is notable given the 7F72 has 24 cores against the 9135’s 16.

The 7F72’s wins are concentrated in memory and cryptographic workloads. Data encryption is its largest victory: 56,261 vs 40,295, a 39.6% advantage. Find prime numbers shows an even larger relative gap at 70.5% (498 vs 292), though the absolute scores are small. Random string sorting goes to the 7F72 by 13.7% (102,436 vs 90,064), and data compression by 9.4% (808,795 vs 739,277). The 7F72 also wins physics simulation with 6,459 vs 5,477, a 17.9% edge.

The overall win count (5 for the 7F72, 12 for the 9135) understates the 7F72’s specialty. The 9135’s wins are broad but often moderate, while the 7F72’s wins are narrow but extreme. The 39.6% encryption gap and 70.5% prime-number gap are outliers that suggest architectural strength in specific instruction patterns, not general superiority.

The Verdict

The data clearly separates these processors by workload type. The AMD EPYC 9135 is the all-around performer for compute-heavy tasks: it wins every Cinebench test by 8.8–8.9%, leads in PassMark multithread, single-thread, integer math, floating-point math, and extended instructions. Its 3,672 single-thread score is 35.1% above the 7F72’s 2,384, making it the obvious choice for applications that depend on per-core responsiveness or lightly threaded code. The 9135’s 2.5% average benchmark gap over the 7F72 (82,980 vs 85,072) is small, but its consistency across general-purpose benchmarks gives it the edge for mixed server workloads.

The AMD EPYC 7F72 is the specialist. Its 39.6% lead in data encryption and 70.5% lead in find-prime-numbers indicate strong performance in cryptographic and integer-iteration workloads. The 13.7% advantage in random string sorting and 9.4% in data compression further point to memory-intensive data processing. For servers dedicated to encryption, compression, or sorting, the 7F72 offers measurable gains despite its older architecture. The physics win (17.9%) adds a scientific-computing angle, though it is isolated.

The percentile data shows both at the 96th percentile of all CPUs, meaning neither is a weak choice. The 9135’s nearest rivals include the EPYC 4584PX (0.1% higher average score) and Intel Core Ultra 9 285K (1% higher), while the 7F72’s rivals include the same Intel parts (1.3% and 1.5% higher, respectively). In a head-to-head, the 9135 is the default recommendation for general servers; the 7F72 is the pick only when its specific benchmark wins match the deployment’s primary task.

Architecture Differences

The two processors come from different generations and nodes. The 7F72 uses Zen 2 architecture on the Rome codename, built on a 7 nm process at TSMC with 3,800 million transistors on a 74 mm² die. The 9135 uses Zen 5 architecture on the Turin codename, built on a 4 nm process at TSMC with 16,630 million transistors across two 70.6 mm² dies. The transistor count on the 9135 is over four times higher, which explains its efficiency gains despite fewer cores.

Core configuration differs substantially. The 7F72 has 24 cores and 48 threads; the 9135 has 16 cores and 32 threads. That is 50% more cores and threads for the 7F72, yet the 9135 still wins most multithreaded benchmarks. The 9135 compensates with higher clocks: 3.65 GHz base and 4.30 GHz boost versus 3.20 GHz base and 3.70 GHz boost for the 7F72. The 9135’s boost clock is 0.60 GHz higher, and its base clock is 0.45 GHz higher.

Cache hierarchies are also distinct. The 7F72 provides 96 KB L1 per core, 512 KB L2 per core, and a massive 192 MB shared L3. The 9135 provides 80 KB L1 per core, 1 MB L2 per core, and only 64 MB shared L3. The 7F72’s L3 is three times larger, which explains its advantage in data compression and sorting tasks that benefit from large working sets. The 9135’s larger L2 per core (1 MB vs 512 KB) supports its single-thread performance.

Memory infrastructure differs sharply. The 7F72 supports DDR4 with an eight-channel bus and 204.8 GB/s bandwidth. The 9135 supports DDR5 with a twelve-channel bus and 576.0 GB/s bandwidth — nearly three times the memory bandwidth. The 9135 also moves to PCIe Gen 5 with 128 lanes (CPU only), while the 7F72 stays on PCIe Gen 4 with unspecified lane count. Both support ECC memory. The 7F72 uses Socket SP3; the 9135 uses Socket SP5, so they are not interchangeable in existing platforms.

FAQ

Q: Which processor has the higher average benchmark score?

A: The AMD EPYC 7F72 has an average benchmark score of 85,072, which is 2.5% higher than the AMD EPYC 9135’s 82,980. Both sit at the 96th percentile of all CPUs.

Q: How large is the single-thread performance gap?

A: In PassMark single-thread, the 9135 scores 3,672 against the 7F72’s 2,384, a 35.1% advantage. This is the largest single-test delta between the two processors.

Q: Does the 7F72 win any benchmark by a significant margin?

A: Yes. The 7F72 leads in data encryption by 39.6% (56,261 vs 40,295) and in find-prime-numbers by 70.5% (498 vs 292). It also wins random string sorting by 13.7% and data compression by 9.4%.

Q: What memory bandwidth does each processor support?

A: The 9135 supports DDR5 with a twelve-channel bus and 576.0 GB/s bandwidth. The 7F72 supports DDR4 with an eight-channel bus and 204.8 GB/s bandwidth.

Q: Are the core counts different?

A: Yes. The 7F72 has 24 cores and 48 threads, while the 9135 has 16 cores and 32 threads. Despite fewer cores, the 9135 wins the PassMark multithread test by 7.7%.

Q: What is the release timeline for these parts?

A: The 7F72 was released on 2020-04-13, while the 9135 was released on 2024-10-09. The 9135 carries a launch MSRP of $1214; the 7F72 has no listed launch MSRP.

Where Each One Wins

The 9135 wins in every Cinebench test (R15, R20, R23, both single and multicore) by 8.8–8.9%, plus PassMark single-thread (35.1%), multithread (7.7%), integer math (10.8%), floating-point math (14.4%), and extended instructions (15.9%). This makes it the clear choice for general compute workloads: rendering, simulation, code compilation, and any mixed server task where single-core speed matters.

The 7F72 wins in data encryption (39.6%), find-prime-numbers (70.5%), random string sorting (13.7%), data compression (9.4%), and physics (17.9%). These are specific but substantial wins. For database compression, cryptographic services, or scientific workloads that involve prime-number iteration, the 7F72 offers measurable performance advantages that the 9135 cannot match.

The architectural rationale is visible in the specs. The 7F72’s 192 MB L3 cache supports its sorting and compression wins, while its higher core count (24 vs 16) likely aids the encryption and physics results. The 9135’s higher clocks (4.30 GHz boost vs 3.70 GHz) and newer Zen 5 core explain its single-thread and math victories. The 9135’s 576.0 GB/s memory bandwidth versus 204.8 GB/s for the 7F72 does not save it in memory-latency-sensitive tests, where the 7F72’s larger cache wins.

Specification Differences

| Field | AMD EPYC 7F72 | AMD EPYC 9135 |

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

| Cores | 24 | 16 |

| Threads | 48 | 32 |

| Base Clock | 3.20 GHz | 3.65 GHz |

| Boost Clock | 3.70 GHz | 4.30 GHz |

| TDP | 240 W | 200 W |

| Socket | AMD Socket SP3 | AMD Socket SP5 |

| Architecture | Zen 2 | Zen 5 |

| Codename | Rome | Turin |

| Process Node | 7 nm | 4 nm |

| Transistors | 3,800 million | 16,630 million |

| Die Size | 74 mm² | 2x 70.6 mm² |

| L1 Cache | 96 KB (per core) | 80 KB (per core) |

| L2 Cache | 512 KB (per core) | 1 MB (per core) |

| L3 Cache | 192 MB (shared) | 64 MB (shared) |

| Memory Support | DDR4 | DDR5 |

| Memory Bus | Eight-channel | Twelve-channel |

| Memory Bandwidth | 204.8 GB/s | 576.0 GB/s |

| PCIe | Gen 4 | Gen 5, 128 Lanes (CPU only) |

| Release Date | 2020-04-13 | 2024-10-09 |

| Launch MSRP | None listed | $1214 |