AMD EPYC 4564P vs AMD EPYC 7F72 Comparison

FAQ

Q: Which processor has the higher average benchmark score?

A: The AMD EPYC 4564P leads with an average benchmark score of 95,183, while the AMD EPYC 7F72 scores 85,072. That puts the 4564P about 12% ahead in overall aggregate performance, and both CPUs sit at the 96th percentile among all CPUs in the database.

Q: How do the two chips compare in single-threaded performance?

A: The EPYC 4564P is far ahead. In PassMark single-thread testing it scores 4,292 versus 2,384 for the 7F72, an 80% advantage. Cinebench R23 single-core shows the same story: 7,848 for the 4564P against 6,328 for the 7F72, a 24% lead.

Q: Are there workloads where the EPYC 7F72 wins?

A: Yes. The 7F72 wins three head-to-head tests: PassMark data encryption (56,261 vs 50,708, a 9.9% edge), PassMark find prime numbers (498 vs 365, a 26.7% edge), and PassMark physics (6,459 vs 3,392, a 47.5% edge). The physics result is its largest win.

Q: What are the core and thread counts of each processor?

A: The EPYC 4564P has 16 cores and 32 threads, while the EPYC 7F72 has 24 cores and 48 threads. Despite having fewer cores, the 4564P wins the majority of benchmark comparisons.

Q: Which processor has the higher boost clock?

A: The EPYC 4564P boosts to 5.70 GHz, while the EPYC 7F72 tops out at 3.70 GHz. The 4564P also has a higher base clock at 4.50 GHz versus 3.20 GHz.

Q: What memory technologies do the two processors support?

A: The EPYC 4564P uses DDR5 memory on a dual-channel bus with 83.2 GB/s bandwidth. The EPYC 7F72 uses DDR4 on an eight-channel bus with 204.8 GB/s bandwidth. Both support ECC memory.

Architecture Differences

The two processors come from different generations of AMD's EPYC roadmap. The EPYC 4564P is built on Zen 4 architecture with the codename Raphael, part of the EPYC 4004 series. The EPYC 7F72 uses Zen 2 architecture with the codename Rome. This generational gap explains most of the performance divergence.

The manufacturing process differs significantly. The 4564P is fabricated on a 5 nm process at TSMC, while the 7F72 uses TSMC's 7 nm process. Transistor counts reflect this: the 4564P packs 13,140 million transistors across a 2x 71 mm² die, while the 7F72 has 3,800 million transistors on a 74 mm² die. The newer node allows the 4564P to reach much higher clock speeds while maintaining a lower power envelope.

Clock speeds are a major differentiator. The 4564P runs at 4.50 GHz base and 5.70 GHz boost, compared to 3.20 GHz base and 3.70 GHz boost for the 7F72. That is a 1.30 GHz gap at base and a 2.00 GHz gap at boost, which directly drives the 4564P's dominance in single-threaded workloads.

Cache layouts also differ. The 4564P has 64 KB of L1 cache per core, 1 MB of L2 per core, and 64 MB of shared L3 cache. The 7F72 has 96 KB of L1 per core, 512 KB of L2 per core, and a much larger 192 MB of shared L3 cache. The 7F72's larger L3 pool helps in certain multi-threaded and data-heavy tasks, but the 4564P's higher clocks and newer architecture compensate in most tests.

Socket and platform support are incompatible. The 4564P uses AMD Socket AM5, while the 7F72 uses AMD Socket SP3. The 4564P supports PCIe Gen 5 with 28 CPU lanes, whereas the 7F72 supports PCIe Gen 4. Memory channels also differ: dual-channel DDR5 for the 4564P versus eight-channel DDR4 for the 7F72. The 7F72 has the higher theoretical memory bandwidth at 204.8 GB/s versus 83.2 GB/s, a result of its wider memory bus.

The 4564P includes integrated Radeon Graphics, while the 7F72 has no integrated graphics. Both processors target the server and workstation market segment, and both are currently in active production. The 4564P was released on 2024-05-20, while the 7F72 launched earlier on 2020-04-13.

Head-to-Head Benchmarks

The EPYC 4564P wins 14 of the 17 head-to-head comparisons, and the margins are often substantial. Cinebench results are uniformly in its favor. In Cinebench R23 multi-core, the 4564P scores 55,592 against 44,829 for the 7F72, a 24% advantage. The same 24% delta appears in Cinebench R20 multi-core (23,348 vs 18,828) and Cinebench R15 multi-core (5,603 vs 4,518). Single-core Cinebench results show a nearly identical pattern: R23 single-core is 7,848 vs 6,328, R20 single-core is 3,296 vs 2,657, and R15 single-core is 791 vs 637, all with a 24% or 24.2% delta.

PassMark tests tell a more nuanced story. The largest single win for the 4564P is in PassMark single-thread performance, where it scores 4,292 versus 2,384, an 80% delta. This is the most lopsided result in the entire comparison and underscores the clock speed advantage. PassMark extended instructions also favor the 4564P heavily: 63,136 vs 46,936, a 34.5% lead. Floating point math shows a 30% edge (141,017 vs 108,437), and integer math is 26.3% ahead (228,761 vs 181,103). PassMark multithread gives the 4564P a 22% win (64,357 vs 52,740).

Data compression and sorting tasks are closer. PassMark data compression shows the 4564P ahead by 6.1% (858,105 vs 808,795). Random string sorting is nearly tied: 103,202 vs 102,436, a 0.7% edge for the 4564P.

The 7F72 takes three wins, and they are revealing. PassMark physics is its standout: 6,459 vs 3,392, a 47.5% margin. This likely reflects the 7F72's higher core count and larger cache. PassMark find prime numbers goes to the 7F72 by 26.7% (498 vs 365). PassMark data encryption is the narrowest of its wins: 56,261 vs 50,708, a 9.9% edge. These three wins suggest the 7F72 retains strengths in specific computational patterns that favor its 24-core configuration and 192 MB L3 cache.

Specification Differences

| Specification | AMD EPYC 4564P | AMD EPYC 7F72 |

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

| Cores | 16 | 24 |

| Threads | 32 | 48 |

| Base Clock | 4.50 GHz | 3.20 GHz |

| Boost Clock | 5.70 GHz | 3.70 GHz |

| TDP | 170 W | 240 W |

| Socket | AMD Socket AM5 | AMD Socket SP3 |

| Architecture | Zen 4 | Zen 2 |

| Codename | Raphael | Rome |

| Process Node | 5 nm | 7 nm |

| Transistors | 13,140 million | 3,800 million |

| Die Size | 2x 71 mm² | 74 mm² |

| L1 Cache | 64 KB (per core) | 96 KB (per core) |

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

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

| Memory Support | DDR5 | DDR4 |

| Memory Bus | Dual-channel | Eight-channel |

| Memory Bandwidth | 83.2 GB/s | 204.8 GB/s |

| PCIe | Gen 5, 28 Lanes | Gen 4 |

| Integrated Graphics | Radeon Graphics | None |

| Release Date | 2024-05-20 | 2020-04-13 |

| Launch MSRP | $699 | Not provided |

The Verdict

The data points decisively toward the AMD EPYC 4564P for most workloads. It wins 14 of 17 head-to-head benchmarks and holds a 12% higher average benchmark score (95,183 vs 85,072). Its single-thread performance is transformative: an 80% lead in PassMark single-thread and a 24% lead across all Cinebench single-core tests. For database workloads, web serving, or any application that relies on per-core responsiveness, the 4564P is the clear choice.

The 4564P also wins multi-threaded comparisons despite having only 16 cores versus 24 cores. Cinebench R23 multi-core shows a 24% advantage, and PassMark multithread gives it a 22% edge. The combination of Zen 4 architecture, 5 nm process, and 5.70 GHz boost clock more than compensates for the 8-core deficit. Power consumption is also lower: 170 W TDP for the 4564P against 240 W for the 7F72.

The EPYC 7F72 remains relevant for specific niches. Its 47.5% win in PassMark physics and 26.7% win in find prime numbers indicate strengths in certain scientific or simulation workloads that exploit its 24 cores, 48 threads, and 192 MB L3 cache. The eight-channel memory interface with 204.8 GB/s bandwidth also makes it attractive for memory-bandwidth-sensitive applications, even though the benchmark data does not show a broad advantage from this feature.

For most buyers, the 4564P is the stronger processor. It is newer, faster in nearly every measured test, and more efficient. The 7F72 is a legacy part from the Zen 2 generation that still wins in three specific PassMark categories, but those wins are insufficient to overcome the 4564P's broad dominance.

Where Each One Wins

AMD EPYC 4564P:

• Single-threaded performance: 80% ahead in PassMark single-thread, 24% ahead in all Cinebench single-core tests.
• Multi-core rendering: 24% ahead in Cinebench R15, R20, and R23 multi-core tests.
• Floating-point math: 30% ahead (141,017 vs 108,437).
• Integer math: 26.3% ahead (228,761 vs 181,103).
• Extended instruction workloads: 34.5% ahead (63,136 vs 46,936).
• General multithreaded throughput: 22% ahead in PassMark multithread.
• Data compression: 6.1% ahead.
• Random string sorting: 0.7% ahead.
• Overall average score: 95,183 vs 85,072.

AMD EPYC 7F72:

• Physics simulation: 47.5% ahead (6,459 vs 3,392), its strongest result.
• Prime number finding: 26.7% ahead (498 vs 365).
• Data encryption: 9.9% ahead (56,261 vs 50,708).
• Memory bandwidth potential: 204.8 GB/s versus 83.2 GB/s, though benchmark data does not translate this into broad wins.
• L3 cache capacity: 192 MB versus 64 MB, which likely contributes to its physics and prime number advantages.

The 4564P is the general-purpose champion. The 7F72 is a specialist that excels in physics-heavy simulation, prime number computation, and encryption tasks. Workloads that match those patterns may justify choosing the 7F72 despite its overall lower average score. Everything else points to the 4564P.