AMD EPYC 4565P vs AMD EPYC 7F72 Comparison

The AMD EPYC 4565P and AMD EPYC 7F72 occupy the same server and workstation segment but come from different eras of EPYC design, and the recorded data shows a clear generational gap. The EPYC 4565P, a Zen 5 based part from the EPYC 4005 series, wins 13 of the 17 head-to-head benchmarks, while the Zen 2 based EPYC 7F72 takes 4. Both CPUs sit in the 96th percentile versus all CPUs in the database, but their average benchmark scores diverge: 95764 for the 4565P against 85072 for the 7F72, roughly a 12.6% aggregate advantage for the newer chip. The interesting story is not that the newer part wins, but where the older one still holds ground.

Head-to-Head Benchmarks

The single most lopsided result in the entire dataset is single-threaded performance. In Passmark single-thread the EPYC 4565P scores 4712 against 2384 for the EPYC 7F72, a 97.7% lead that reflects the 4565P's 4.30 GHz base and 5.70 GHz boost clocks against the 7F72's 3.20 GHz base and 3.70 GHz boost. That clock gap, combined with three generations of IPC improvements from Zen 2 to Zen 5, produces a doubling of per-core throughput.

Cinebench shows the same picture with remarkable consistency. The 4565P wins every Cinebench test by essentially the same margin: 21.4% in R20 multi-core (22850 vs 18828), 21.4% in R20 single-core (3225 vs 2657), 21.4% in R23 multi-core (54405 vs 44829), 21.4% in R23 single-core (7680 vs 6328), 21.4% in R15 multi-core (5484 vs 4518), and 21.5% in R15 single-core (774 vs 637). The uniformity matters: it means the 16-core Zen 5 part outruns the 24-core Zen 2 part in multi-threaded rendering despite having 8 fewer cores, and it does so while trailing nowhere in per-core work.

Passmark's pure compute subtests amplify the 4565P's advantage. Floating point math goes to the 4565P by 40.2% (152003 vs 108437), integer math by 38.4% (250683 vs 181103), and extended instructions by 37.1% (64345 vs 46936). Passmark multi-thread goes to the 4565P by 20.4% (63474 vs 52740), and data compression by 6.4% (860786 vs 808795).

The EPYC 7F72's four wins are concentrated in memory-heavy and integer-throughput workloads. Its largest is Passmark physics, where it scores 6459 against 2976, a 53.9% win that lines up directly with its eight-channel DDR4 memory bus delivering 204.8 GB/s versus the 4565P's dual-channel DDR5 at 89.6 GB/s. Random string sorting favors the 7F72 by 20.6% (102436 vs 81318), again a bandwidth- and cache-sensitive test where its 192 MB of shared L3 helps. Prime number search goes to the 7F72 by 41% (498 vs 294), and data encryption by 14.2% (56261 vs 48268).

Where Each One Wins

For the EPYC 4565P, the use-case case is broad: any workload that depends on per-core speed, latency-sensitive licensing, or instruction throughput. Simulation, CAD, compilation, rendering, and general workstation duty all favor it, as the near-total sweep of Cinebench and the 97.7% single-thread lead indicate. Its Passmark multi-thread score of 63474 also shows it handles heavily parallel jobs faster than the higher-core-count rival. Among its nearest rivals in the database, the 4565P's average score of 95764 sits within 0.2% of the AMD EPYC 9175F (95615), 0.3% behind the AMD Ryzen 9 PRO 9945 (96083), 0.6% ahead of the AMD EPYC 4564P (95183), and 1.7% behind the AMD Ryzen 9 9955HX3D (97453), placing it in a tight cluster of high-performing modern parts.

For the EPYC 7F72, the case is narrower but real. Workloads that stream large data sets through memory, or that benefit from a very large shared L3 cache, favor it: physics simulation, sorting, prime searching, and encryption all fall its way. Its rivals in the database are of its own performance tier: the Intel Core Ultra 9 290K Plus (84003, 1.3% behind), the Intel Core Ultra 9 285K (83807, 1.5% behind), the AMD EPYC 4584PX (83090, 2.4% behind), and the AMD EPYC 9135 (82980, 2.5% behind). It holds a higher average score than all four.

Architecture Differences

The two chips bookend three Zen generations. The EPYC 4565P uses the Zen 5 architecture under the codename Grado, built on TSMC's 4 nm process, with 16,630 million transistors across a 2x 70.6 mm² dual-die layout. The EPYC 7F72 uses Zen 2, codename Rome, on TSMC's 7 nm process, with 3,800 million transistors on a single 74 mm² die. The node advantage is the root of the frequency gap: 4.30 GHz base and 5.70 GHz boost on the 4565P versus 3.20 GHz base and 3.70 GHz boost on the 7F72.

Cache design philosophy differs sharply. The 7F72 pairs a smaller per-core allocation, 96 KB of L1 and 512 KB of L2 per core, with a massive 192 MB of shared L3, a server-oriented hierarchy sized to feed many cores from memory-resident data sets. The 4565P inverts this: 80 KB of L1 and a full 1 MB of L2 per core, but only 64 MB of shared L3. Per-core L2 is double, aggregate L3 is a third the size.

Feature support follows the same split. The 4565P offers PCIe Gen 5 with 24 CPU lanes and integrated Radeon Graphics; the 7F72 offers PCIe Gen 4 and no integrated graphics, consistent with its expectation of discrete server infrastructure. Both support ECC memory. The 4565P has a locked multiplier, as does the 7F72.

Specification Differences

The differing fields, side by side:

• Cores/threads: 16/32 on the 4565P versus 24/48 on the 7F72
• Base clock: 4.30 GHz versus 3.20 GHz; boost: 5.70 GHz versus 3.70 GHz
• TDP: 170 W versus 240 W
• Socket: AMD Socket AM5 versus AMD Socket SP3
• Architecture: Zen 5 (Grado, 4 nm) versus Zen 2 (Rome, 7 nm)
• Transistors: 16,630 million versus 3,800 million; die: 2x 70.6 mm² versus 74 mm²
• Cache: 80 KB L1 and 1 MB L2 per core with 64 MB shared L3 versus 96 KB L1 and 512 KB L2 per core with 192 MB shared L3
• Memory: DDR5 dual-channel at 89.6 GB/s versus DDR4 eight-channel at 204.8 GB/s
• PCIe: Gen 5, 24 CPU lanes versus Gen 4
• Graphics: integrated Radeon Graphics versus none
• Release date: 2025-05-12 versus 2020-04-13
• Launch MSRP: the 4565P lists a launch MSRP of $589; no launch MSRP is recorded for the 7F72
• Part numbers: 100-000001559 versus 100-000000141100-000000141WOF

Both are active production parts, both target the server and workstation segment, and both sit in the 96th percentile versus all CPUs in the database.

FAQ

Q: Which CPU is faster overall?

A: The EPYC 4565P. It wins 13 of 17 head-to-head benchmarks with an average score of 95764 versus 85072, and it leads by about 21% in every Cinebench test, both single- and multi-core, despite having fewer cores.

Q: Does the EPYC 7F72 win anything?

A: Yes, four tests: Passmark physics by 53.9% (6459 vs 2976), prime number search by 41% (498 vs 294), random string sorting by 20.6% (102436 vs 81318), and data encryption by 14.2% (56261 vs 48268). All are bandwidth- or cache-favored workloads.

Q: Why does the older chip win memory-bound tests?

A: The EPYC 7F72 has an eight-channel DDR4 bus delivering 204.8 GB/s plus 192 MB of shared L3, while the 4565P is limited to dual-channel DDR5 at 89.6 GB/s with 64 MB of shared L3.

Q: How big is the single-thread gap?

A: The 4565P scores 4712 in Passmark single-thread against 2384 for the 7F72, a 97.7% lead, driven by 5.70 GHz boost clocks on Zen 5 versus 3.70 GHz on Zen 2.

Q: Do they use the same socket?

A: No. The 4565P uses AMD Socket AM5; the 7F72 uses AMD Socket SP3. They are not interchangeable platforms, and memory support differs accordingly (DDR5 versus DDR4).

Q: How does each compare to its own nearest rivals?

A: The 4565P sits within 1.7% of rivals like the EPYC 9175F, Ryzen 9 PRO 9945, EPYC 4564P, and Ryzen 9 9955HX3D. The 7F72 outscores its nearest tier, including the Intel Core Ultra 9 290K Plus and 285K, by 1.3% to 2.5%.