AMD EPYC 9384X vs Intel Xeon 674X Comparison

Head-to-Head Benchmarks

The benchmark database shows a clear overall winner in the head-to-head comparison: Intel Xeon 674X wins 14 of the 16 recorded tests, with AMD EPYC 9384X taking only two. The margins are not uniform, however, and the pattern of victories tells a more nuanced story than the raw win count.

Intel's most dominant victory comes in PassMark floating point math. The Xeon 674X scores 243877 against the EPYC's 174630, a 39.7% advantage. This is the largest delta in the entire comparison. Extended instructions follow closely, with Intel leading 97373 to 74363, a 30.9% gap. Single-thread performance also strongly favors Intel: the Xeon's 3933 PassMark single-thread score beats the EPYC's 3015 by 30.4%. The Cinebench single-core results repeat this pattern, with the Xeon leading by 20.9% in both R15 (1018 vs 842) and R20 (4243 vs 3510).

The multi-core Cinebench results are consistent across all three versions. In R15, the Xeon scores 7213 against 5968; in R20, 30057 against 24870; in R23, 71566 against 59215. Every one of these shows the same 20.9% lead for Intel. PassMark multithread also lands at exactly 20.9%, with scores of 84196 and 69665. Data compression shows a smaller but still solid Intel win: 1236272 vs 1119983, a 10.4% margin. Prime number finding gives Intel a 16.3% edge (693 vs 596), and random string sorting goes to Intel by 6.8% (127529 vs 119440). Integer math is the closest Intel victory, just 3.7% ahead (308968 vs 297833).

AMD's two wins are specific and notable. PassMark physics shows the EPYC 9384X ahead by 18.7%, scoring 9332 against the Xeon's 7586. This is the largest single reverse margin in the comparison. Data encryption is the other AMD win, with the EPYC scoring 72631 versus 61195, a 15.7% advantage. These two results indicate that AMD's architecture holds clear strengths in physics simulation workloads and encryption tasks, even while trailing broadly elsewhere.

The overall average benchmark score reflects the general trend: Intel Xeon 674X sits at 143103, while AMD EPYC 9384X averages 120427. Intel sits in the 98th percentile of all CPUs in the database, while the EPYC is in the 97th percentile. Notably, the Xeon's average score is very close to its nearest rival, the Intel Xeon 6732P, which averages 143444, a delta of only -0.2%. The EPYC's nearest rival, the Intel Xeon w7-3565X, averages 118307, putting the EPYC 1.8% ahead of that part.

FAQ

Q: Which processor wins the most head-to-head benchmarks?

A: The Intel Xeon 674X wins 14 of the 16 recorded head-to-head tests. The AMD EPYC 9384X wins only 2.

Q: What is the largest performance gap in either direction?

A: The largest gap is in PassMark floating point math, where the Intel Xeon 674X scores 243877 versus the AMD EPYC 9384X's 174630, a 39.7% advantage for Intel.

Q: Are there any workloads where the AMD EPYC 9384X beats the Intel Xeon 674X?

A: Yes. The EPYC wins PassMark physics by 18.7% (9332 vs 7586) and PassMark data encryption by 15.7% (72631 vs 61195).

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

A: The Intel Xeon 674X leads by 30.4% in PassMark single-thread (3933 vs 3015) and by 20.9% in both Cinebench R15 single-core (1018 vs 842) and R20 single-core (4243 vs 3510).

Q: What are the average benchmark scores for each CPU?

A: The Intel Xeon 674X has an average benchmark score of 143103, while the AMD EPYC 9384X has an average of 120427.

Q: Which CPU has the higher percentile ranking in the database?

A: The Intel Xeon 674X is in the 98th percentile of all CPUs, while the AMD EPYC 9384X is in the 97th percentile.

Architecture Differences

The two processors come from fundamentally different design philosophies, and the benchmark data reflects those choices. Intel's Xeon 674X is built on Granite Rapids architecture, using a 5 nm process node from Intel's own foundry. The die size is listed as 2x 598 mm². AMD's EPYC 9384X uses Zen 4 architecture under the Genoa-X codename, also on a 5 nm node but fabricated by TSMC, with a die size of 8x 72 mm². AMD lists 90,160 million transistors, while Intel's transistor count is not recorded.

Core and thread counts differ significantly. The Intel part has 28 cores and 56 threads, while the AMD part has 32 cores and 64 threads. Despite having fewer cores, the Xeon 674X wins most multi-threaded tests, which points to higher per-core efficiency or clock advantages. The base clocks are close, with Intel at 3.00 GHz and AMD at 3.10 GHz, but the boost clocks diverge sharply: Intel boosts to 4.90 GHz, while AMD tops out at 3.90 GHz. This 1.00 GHz boost advantage likely explains Intel's strong single-thread and floating-point results.

Cache configurations show a major difference in capacity and organization. Intel provides 112 KB of L1 per core, 2 MB of L2 per core, and 144 MB of shared L3. AMD provides 64 KB of L1 per core, 1 MB of L2 per core, and a massive 768 MB of shared L3. The AMD L3 is over five times larger than Intel's, which may contribute to its physics and encryption wins, but it does not overcome Intel's clock advantage in most other workloads.

Memory subsystems also differ. Both support DDR5 and ECC memory, but Intel uses an eight-channel memory bus with 409.6 GB/s bandwidth, while AMD uses a twelve-channel bus with 460.8 GB/s bandwidth. The AMD part has a raw bandwidth advantage of 51.2 GB/s. PCIe support is identical on paper: Gen 5 with 128 lanes (CPU only) for both.

Socket and platform choices are incompatible. Intel uses Socket 4710, while AMD uses Socket SP5. The Intel part has an unlocked multiplier, while the AMD part does not. Intel's release date is recorded as February 2026, while AMD's is June 2023. Intel lists a launch MSRP of $2199, and AMD lists a launch MSRP of $5529. The Intel part number is SA2CZ, while AMD's part number is not recorded. Intel's integrated graphics are listed as N/A, and AMD's are not specified.

The Verdict

The benchmark data points to a clear recommendation for most workloads: the Intel Xeon 674X is the stronger processor in this comparison. It wins 14 of 16 tests, with particularly large margins in floating point math (39.7%), extended instructions (30.9%), and single-thread performance (30.4%). Its multi-core Cinebench results are consistently 20.9% ahead across R15, R20, and R23. For general compute, rendering, data compression, and integer workloads, the Intel part is the better choice.

The AMD EPYC 9384X should be selected for two specific scenarios. First, physics simulation workloads: the EPYC leads by 18.7% in PassMark physics, which is the largest AMD win. Second, data encryption: the EPYC leads by 15.7% in PassMark data encryption. If either of these workload types dominates a deployment, the AMD part has a measurable and meaningful advantage.

The AMD part also offers more cores (32 vs 28) and more threads (64 vs 56), plus a larger L3 cache (768 MB vs 144 MB) and higher memory bandwidth (460.8 GB/s vs 409.6 GB/s). These hardware advantages do not translate into benchmark wins in most tests, but they may matter for specific workloads not captured in the recorded benchmark suite. The EPYC's older release date (June 2023 vs February 2026) and higher launch MSRP ($5529 vs $2199) are also recorded differences, though the benchmark data does not indicate any performance-based justification for the price gap.

For users who need maximum single-thread speed, floating-point throughput, or broad multi-core performance, the data favors Intel. For users who prioritize physics simulation or encryption throughput, or who require the larger core count and cache capacity of the AMD part, the EPYC 9384X is the justified pick.

Specification Differences

| Field | Intel Xeon 674X | AMD EPYC 9384X |

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

| Cores | 28 | 32 |

| Threads | 56 | 64 |

| Base Clock | 3.00 GHz | 3.10 GHz |

| Boost Clock | 4.90 GHz | 3.90 GHz |

| TDP | 270 W | 320 W |

| Socket | Intel Socket 4710 | AMD Socket SP5 |

| Architecture | Granite Rapids | Zen 4 |

| Codename | Granite Rapids | Genoa-X |

| Generation | Xeon 600 (Granite Rapids-WS) | EPYC (Zen 4 (Genoa)) |

| Foundry | Intel | TSMC |

| Die Size | 2x 598 mm² | 8x 72 mm² |

| Transistors | Not recorded | 90,160 million |

| L1 Cache | 112 KB (per core) | 64 KB (per core) |

| L2 Cache | 2 MB (per core) | 1 MB (per core) |

| L3 Cache | 144 MB (shared) | 768 MB (shared) |

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

| Memory Bandwidth | 409.6 GB/s | 460.8 GB/s |

| Release Date | February 2026 | June 2023 |

| Launch MSRP | $2199 | $5529 |

| Multiplier Unlocked | Yes | No |

| Part Number | SA2CZ | Not recorded |

Both processors share several specifications: 5 nm process node, DDR5 memory support, ECC memory support, PCIe Gen 5 with 128 lanes (CPU only), server/workstation market segment, and active production status. The Intel part's integrated graphics are listed as N/A, while the AMD part's are not specified.