AMD EPYC 4565P vs Intel Core Ultra 9 285K Comparison

The AMD EPYC 4565P and Intel Core Ultra 9 285K sit in the same performance tier — both land at the 98th percentile of all CPUs — yet they are built for different jobs. The EPYC 4565P is a 16-core, 32-thread Zen 5 server/workstation chip on AMD Socket AM5; the Core Ultra 9 285K is a 24-core, 24-thread Arrow Lake desktop processor on Intel Socket 1851. Both carried a $589 launch MSRP. Average benchmark scores favor AMD: 95,820 for the EPYC versus 93,672 for the Intel, a 2.3% gap in AMD's favor. But the head-to-head results tell a more interesting story: Intel wins 14 of 17 tests, while AMD takes 3 — and two of those wins are by very large margins.

Where Each One Wins

The EPYC 4565P wins the workloads that reward raw integer throughput and memory-friendly data movement. Its biggest head-to-head victory is PassMark integer math, where it scores 253,794 against Intel's 173,281 — a 46.5% lead. It also wins data compression (856,815 vs 794,635, up 7.8%) and extended instructions (63,454 vs 62,300, up 1.9%). These three wins point to a clear use case: integer-heavy server code, compression pipelines, and workloads that exercise the CPU's extended instruction set. If your application spends most of its time on arithmetic on whole numbers, bulk data transforms, or archive-style operations, the EPYC is the stronger pick.

The Core Ultra 9 285K wins everything else. That includes all six Cinebench tests, floating-point math (225,800 vs 152,067, a 32.7% lead), prime number finding (545 vs 299, up 45.1%), physics (4,062 vs 3,072, up 24.4%), data encryption (58,210 vs 49,408, up 15.1%), random string sorting (95,726 vs 82,131, up 14.2%), single-thread (5,097 vs 4,738, up 7%), and the PassMark multithread score (67,737 vs 64,006, up 5.5%). The practical takeaway: for floating-point simulation, encryption, physics, and general desktop responsiveness, Intel has the edge. The Core Ultra 9 is the generalist; the EPYC is the specialist.

Architecture Differences

The two chips come from different design philosophies. The EPYC 4565P uses AMD's Zen 5 architecture (codename Grado) on a 4 nm TSMC process, while the Core Ultra 9 285K uses Intel's Arrow Lake architecture (Arrow Lake-S) on a 3 nm TSMC process. Both are fabbed by TSMC, but the process nodes differ.

Core and thread counts diverge sharply. The EPYC has 16 cores and 32 threads via simultaneous multithreading; the Intel has 24 physical cores and 24 threads — no SMT. That means Intel relies on more physical cores, while AMD gets more logical threads from fewer cores. Base clocks differ: 4.30 GHz for AMD versus 3.70 GHz for Intel. Boost clocks are identical at 5.70 GHz for both.

Cache layouts are also very different. The EPYC provides 80 KB of L1 per core, 1 MB of L2 per core, and 64 MB of shared L3. The Intel provides 192 KB of L1 per core, 3 MB of L2 per core, and 36 MB of shared L3. So Intel has much larger per-core caches, while AMD has nearly double the total L3.

Memory and I/O differ as well. Both support DDR5 on dual-channel buses, but Intel's bandwidth is higher at 102.4 GB/s versus AMD's 89.6 GB/s. Both support ECC memory. PCIe lanes favor AMD: Gen 5 with 24 CPU lanes versus Intel's Gen 5 with 20 CPU lanes. Integrated graphics differ — AMD has Radeon Graphics, Intel has Arc Xe-LPG Graphics 64EU. The EPYC is a locked multiplier part; the Core Ultra 9 is unlocked. Die size also differs: the AMD uses a 2x 70.6 mm² configuration, while the Intel is a single 243 mm² die. Release dates are roughly seven months apart: Intel launched 2024-10-23, AMD on 2025-05-12.

Head-to-Head Benchmarks

The head-to-head data shows a clear pattern: Intel wins the Cinebench suite by a consistent 5.5% margin across every test. In Cinebench R23 multi-core, Intel scores 57,584 versus AMD's 54,405; in R23 single-core, it's 8,129 versus 7,680. R20 multi-core is 24,185 against 22,850, and R15 multi-core is 5,804 versus 5,484. The single-core R15 and R20 tests show the same 5.5% gap (819 vs 774 and 3,414 vs 3,225, respectively). This uniformity suggests a per-clock advantage for Intel in lightly threaded and moderately threaded rendering workloads, not just a core-count effect.

The biggest Intel wins come outside Cinebench. Prime number finding is a rout: 545 for Intel versus 299 for AMD, a 45.1% advantage — likely reflecting Intel's larger per-core L2 and higher single-thread throughput. Floating-point math shows a 32.7% lead (225,800 vs 152,067), and physics is 24.4% ahead (4,062 vs 3,072). Encryption favors Intel by 15.1% (58,210 vs 49,408), and random string sorting by 14.2% (95,726 vs 82,131). Even the PassMark multithread aggregate, which might be expected to favor the 32-thread AMD part, goes to Intel at 67,737 versus 64,006, a 5.5% margin.

AMD's counterpunches are concentrated but devastating. The integer math score of 253,794 crushes Intel's 173,281 — a 46.5% lead that is the single largest delta in the entire comparison. Data compression at 856,815 beats Intel's 794,635 by 7.8%, and extended instructions at 63,454 edges out 62,300 by 1.9%. Notably, AMD's single-thread score of 4,738 trails Intel's 5,097 by 7%, yet AMD still wins integer-heavy tests — the Zen 5 core's integer pipeline appears to be doing heavy lifting despite the single-thread deficit.

Specification Differences

| Field | AMD EPYC 4565P | Intel Core Ultra 9 285K |

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

| Cores | 16 | 24 |

| Threads | 32 | 24 |

| Base clock | 4.30 GHz | 3.70 GHz |

| Boost clock | 5.70 GHz | 5.70 GHz |

| TDP | 170 W | 125 W |

| Socket | AMD Socket AM5 | Intel Socket 1851 |

| Architecture | Zen 5 | Arrow Lake |

| Process node | 4 nm | 3 nm |

| Die size | 2x 70.6 mm² | 243 mm² |

| L1 cache | 80 KB (per core) | 192 KB (per core) |

| L2 cache | 1 MB (per core) | 3 MB (per core) |

| L3 cache | 64 MB (shared) | 36 MB (shared) |

| Memory bandwidth | 89.6 GB/s | 102.4 GB/s |

| PCIe | Gen 5, 24 lanes | Gen 5, 20 lanes |

| Integrated graphics | Radeon Graphics | Arc Xe-LPG Graphics 64EU |

| Market segment | Server/Workstation | Desktop |

| Multiplier | Locked | Unlocked |

| Release date | 2025-05-12 | 2024-10-23 |

Both chips share DDR5 support, dual-channel memory buses, ECC memory support, TSMC as foundry, and a 98th-percentile standing. They also match on boost clock and launch MSRP.

FAQ

Q: Which CPU is faster in Cinebench R23 multi-core?

A: The Intel Core Ultra 9 285K scores 57,584 versus the EPYC 4565P's 54,405, a 5.5% lead for Intel.

Q: The EPYC has more threads (32 vs 24) — why doesn't it win multithreaded tests?

A: The PassMark multithread score still goes to Intel at 67,737 versus 64,006, and Intel wins every Cinebench multi-core test by 5.5%. Intel's 24 physical cores outperform AMD's 16 cores with SMT in these workloads.

Q: Which chip has more L3 cache?

A: The EPYC 4565P has 64 MB of shared L3, while the Core Ultra 9 has 36 MB — AMD has nearly double.

Q: Do both support ECC memory?

A: Yes, both list ECC memory support as true.

Q: Can either processor be overclocked?

A: Only the Intel Core Ultra 9 285K has an unlocked multiplier; the EPYC 4565P is locked.

Q: Which has higher memory bandwidth?

A: Intel leads at 102.4 GB/s versus AMD's 89.6 GB/s, despite both using dual-channel DDR5.

The Verdict

Pick the Intel Core Ultra 9 285K if you need a general-purpose desktop CPU. It wins 14 of 17 head-to-head tests, leads in every Cinebench workload, and holds decisive advantages in floating-point math (32.7%), prime finding (45.1%), physics (24.4%), and encryption (15.1%). Its unlocked multiplier and lower 125 W TDP make it the more flexible and efficient choice for a desktop build, and its 24 physical cores handle threaded workloads well despite lacking SMT.

Pick the AMD EPYC 4565P if your workload is dominated by integer math, data compression, or extended-instruction processing. Its 46.5% lead in integer math and 7.8% lead in compression are not marginal — they are transformative for those specific tasks. The 64 MB of L3 cache and 24 PCIe Gen 5 lanes also make it a stronger server/workstation part for memory-heavy and I/O-heavy deployments. The 2.3% higher average benchmark score (95,820 vs 93,672) reflects that specialization paying off in aggregate.

Both are 98th-percentile CPUs, so neither is a bad choice. The decision comes down to whether you need Intel's broad all-around dominance or AMD's narrow, extreme integer throughput. For a desktop that does a bit of everything, the Core Ultra 9 is the safer pick. For a server node running integer-heavy services, the EPYC 4565P is the one that turns benchmark leads into real workload wins.