AMD EPYC 9634 vs Intel Xeon 6747P Comparison

The AMD EPYC 9634 and Intel Xeon 6747P are both 99th-percentile server processors, but they achieve that status through very different designs. The EPYC 9634, built on Zen 4 (Genoa), offers a massive 84 cores and 168 threads, while the Xeon 6747P, based on Granite Rapids, counters with 48 cores and 96 threads but higher clock speeds. The benchmark data reveals a clear split: AMD dominates in raw throughput and integer-heavy workloads, while Intel wins in single-threaded and floating-point tasks. This analysis breaks down the numbers to help you decide which platform aligns with your specific server workloads.

Head-to-Head Benchmarks

The most striking pattern in the head-to-head results is AMD's dominance in multi-threaded and integer-scaled tests. In Cinebench R15, R20, and R23 multi-core, the EPYC 9634 wins all three by a consistent 6.2% margin. The scores tell a clear story: 9248 vs 8712 in R15, 38535 vs 36301 in R20, and 91752 vs 86432 in R23. This uniformity suggests the EPYC’s core-count advantage (84 vs 48) translates directly into sustained multi-core performance without throttling penalties.

The gap widens dramatically in PassMark integer math, where the EPYC 9634 scores 725356 against the Xeon’s 468518, a 54.8% advantage. Similarly, random string sorting shows a 44.8% win for AMD (261134 vs 180382). Data compression follows with a 22% lead (2236412 vs 1833378), and data encryption is the single largest victory: 151943 vs 90789, a 67.4% margin. These are not marginal differences; they represent fundamental throughput advantages in workloads that depend on many parallel integer operations.

However, the Xeon 6747P takes five of the fourteen head-to-head tests, and its victories are equally decisive in their own domains. PassMark single-thread shows Intel ahead by 9.6% (3236 vs 2924). Physics tests favor Intel by 8.3% (13398 vs 12291). Even in floating-point math, the Xeon edges out a 3.3% win (365904 vs 353784). Extended instructions also go to Intel by 3.5% (142557 vs 137543). The overall score comparison in the nearestRivals data shows the EPYC 9634 leading by only 2.5% in average benchmark score (244274 vs 238263), which confirms that while AMD wins more tests, Intel’s wins are in areas that matter for specific application types.

The PassMark multithread score (107944 vs 101685) shows a 6.2% win for AMD, mirroring the Cinebench results. Interestingly, the EPYC 9634 also wins the find-prime-numbers test, though by a slim 2.2% (1176 vs 1151). This is a workload where cache size and memory bandwidth often matter more than raw core count, and AMD’s 384 MB L3 cache provides an edge. In total, the EPYC 9634 secures 9 wins across the 14 tests, while Intel claims 5, but the margin in AMD’s losses is generally smaller than the margin in its biggest wins.

Architecture Differences

The two processors represent fundamentally different design philosophies. The AMD EPYC 9634 uses the Zen 4 architecture on a 5 nm TSMC process, with a die size of 12x 72 mm² and 78,840 million transistors. It is built on the EPYC 9004 series (Genoa) and uses AMD Socket SP5. The Intel Xeon 6747P is also on a 5 nm process, but fabricated by Intel, with a die size of 2x 598 mm² and no transistor count listed. It uses the Granite Rapids architecture on Intel Socket 4710.

Core and cache configurations diverge sharply. The EPYC 9634 packs 84 cores and 168 threads, with 64 KB L1 and 1 MB L2 per core, plus a massive 384 MB shared L3. The Xeon 6747P has 48 cores and 96 threads, with 112 KB L1 and 2 MB L2 per core, and 288 MB shared L3. This means the AMD chip gives you 75% more cores and threads, while Intel gives you larger per-core caches but 25% less total L3.

Memory subsystems also differ. The EPYC 9634 supports DDR5 over a twelve-channel bus, delivering 460.8 GB/s bandwidth. The Xeon 6747P also supports DDR5 but over an eight-channel bus, capping at 409.6 GB/s. Both support ECC memory. PCIe connectivity shows another gap: the EPYC provides 128 Gen 5 lanes (CPU only), while the Xeon provides 88 Gen 5 lanes. For systems that need maximum I/O expansion, this is a significant architectural advantage for AMD.

Clock speeds favor Intel. The Xeon 6747P has a 2.70 GHz base and 3.90 GHz boost, versus 2.25 GHz base and 3.70 GHz boost on the EPYC 9634. This explains Intel’s single-thread wins. However, the TDP tells a different story: Intel draws 330 W versus AMD’s 290 W, meaning the EPYC achieves higher multi-core performance while consuming less power. The release dates also differ substantially: the EPYC 9634 launched on 2022-11-09, while the Xeon 6747P arrived much later on 2025-02-23.

FAQ

Q: Which processor has higher multi-core performance?

A: The AMD EPYC 9634 wins every Cinebench multi-core test by 6.2%, scoring 91752 in R23 versus the Xeon 6747P’s 86432. Its 84 cores and 168 threads provide a clear throughput advantage over Intel’s 48 cores and 96 threads.

Q: Does the Intel Xeon 6747P have any benchmark wins?

A: Yes. It wins PassMark single-thread (3236 vs 2924, a 9.6% lead), physics (13398 vs 12291, 8.3% lead), floating-point math (365904 vs 353784, 3.3% lead), and extended instructions (142557 vs 137543, 3.5% lead).

Q: Which CPU offers better memory bandwidth?

A: The AMD EPYC 9634 provides 460.8 GB/s over a twelve-channel DDR5 bus. The Intel Xeon 6747P offers 409.6 GB/s over an eight-channel bus. This 51.2 GB/s difference can matter in memory-bound workloads.

Q: What is the core count difference?

A: The EPYC 9634 has 84 cores and 168 threads. The Xeon 6747P has 48 cores and 96 threads. AMD provides 36 more cores and 72 more threads, a 75% increase in each metric.

Q: How much larger is the L3 cache on the AMD part?

A: The EPYC 9634 has 384 MB of shared L3 cache, while the Xeon 6747P has 288 MB. That is a 96 MB advantage for AMD, which likely contributes to its 22% lead in data compression.

Q: Which processor consumes more power?

A: The Intel Xeon 6747P has a TDP of 330 W, which is 40 W higher than the AMD EPYC 9634’s 290 W. Despite this, Intel loses most multi-threaded tests, indicating lower performance-per-watt.

Specification Differences

| Specification | AMD EPYC 9634 | Intel Xeon 6747P |

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

| Cores | 84 | 48 |

| Threads | 168 | 96 |

| Base Clock | 2.25 GHz | 2.70 GHz |

| Boost Clock | 3.70 GHz | 3.90 GHz |

| TDP | 290 W | 330 W |

| Socket | AMD Socket SP5 | Intel Socket 4710 |

| Process Node | 5 nm (TSMC) | 5 nm (Intel) |

| Die Size | 12x 72 mm² | 2x 598 mm² |

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

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

| L3 Cache | 384 MB (shared) | 288 MB (shared) |

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

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

| PCIe Lanes | Gen 5, 128 (CPU only) | Gen 5, 88 (CPU only) |

| Release Date | 2022-11-09 | 2025-02-23 |

| Launch MSRP | $10304 | $6497 |

The Verdict

The data points to a clear conclusion: the AMD EPYC 9634 is the superior choice for throughput-oriented server workloads. It wins 9 of 14 head-to-head benchmarks, including all multi-core Cinebench tests and the largest-margin wins in integer math (54.8%), encryption (67.4%), and random string sorting (44.8%). Its 84 cores, 384 MB L3 cache, and twelve-channel memory bus provide a structural advantage that Intel’s higher clocks cannot overcome in most scenarios. The average benchmark score (244274 vs 238263) confirms AMD’s overall lead, even if it is modest at 2.5%.

The Intel Xeon 6747P is not without merit, but its wins are confined to specific areas: single-thread performance (3236 vs 2924) and floating-point math (365904 vs 353784). These are valuable for workloads that are latency-sensitive or heavily vectorized, but they do not compensate for the massive integer and multi-thread deficits. The Xeon also consumes 40 W more power while offering fewer cores and less memory bandwidth, which weakens its case for dense virtualization or large-scale data processing. The EPYC 9634 is the better all-around server CPU according to the benchmark data.

Where Each One Wins

AMD EPYC 9634 wins in: multi-threaded rendering (Cinebench R15, R20, R23 all by 6.2%), data compression (22% lead), data encryption (67.4% lead), integer math (54.8% lead), random string sorting (44.8% lead), and prime number finding (2.2% lead). These results make it the obvious pick for database workloads, scientific computing, virtualization hosts, and any application that scales with core count and large L3 cache. The 128 PCIe Gen 5 lanes also make it superior for storage servers or GPU-dense nodes requiring extensive I/O.

Intel Xeon 6747P wins in: single-thread tasks (9.6% lead), physics simulations (8.3% lead), floating-point math (3.3% lead), and extended instruction sets (3.5% lead). These victories suggest it is better suited for engineering simulations, financial modeling that relies on double-precision arithmetic, and legacy single-threaded applications that cannot utilize many cores. The higher base and boost clocks (2.70/3.90 GHz vs 2.25/3.70 GHz) provide a tangible benefit in latency-critical environments, even if the overall throughput ceiling is lower.

The choice ultimately depends on whether your workload scales across many cores or depends on per-thread speed. If you are building a high-core-count server for parallel processing, the AMD EPYC 9634 is the data-backed winner. If your applications are single-threaded or floating-point bound and you need maximum clock speed, the Intel Xeon 6747P offers specific, if narrower, advantages.