AMD EPYC 9384X vs Intel Xeon 658X Comparison

The AMD EPYC 9384X and Intel Xeon 658X represent two distinct philosophies in the server and workstation space. The benchmark data shows a clear split: the Intel Xeon 658X dominates synthetic multi-threaded and single-threaded workloads, while the AMD EPYC 9384X excels in specialized data processing and integer-heavy tasks. The Intel part wins 12 of the 17 head-to-head benchmarks, yet the AMD part counters with significant margins in specific areas like encryption and physics, where its advantage reaches as high as 44.2%. Both processors sit at the 97th percentile among all CPUs, with average benchmark scores of 120427 for the AMD and 116060 for the Intel, making this a contest between raw throughput and specialized acceleration.

Where Each One Wins

The Intel Xeon 658X is the generalist winner across the broadest set of workloads. In every Cinebench test—R15, R20, and R23, both multi-core and single-core—the Intel part leads by a consistent 5.2% margin. This uniformity suggests a fundamental architectural advantage in rendering and content creation tasks. The Intel processor also takes PassMark’s floating-point math test by 17%, scoring 210480 against the AMD’s 174630, and it wins the extended instructions test by 12.1% (84626 vs 74363). For users running physics simulations that rely on floating-point precision, the Intel part’s 6470 score in PassMark physics is notably lower than the AMD’s 9332, but the Intel still wins the single-thread test by 19.1% (3728 vs 3015) and the find-prime-numbers test by 8.2% (649 vs 596).

The AMD EPYC 9384X, conversely, is the specialist. Its 38.7% lead in data encryption (72631 vs 52357) is the largest margin in the entire comparison, indicating a strong hardware acceleration path for cryptographic workloads. It also wins integer math by 12.8% (297833 vs 263995), random string sorting by 15.9% (119440 vs 103028), and data compression by 5.5% (1119983 vs 1062062). The most striking outlier is PassMark physics, where the AMD part scores 9332 against the Intel’s 6470—a 44.2% advantage. This suggests that while the Intel part is faster for general floating-point math, the AMD’s architecture handles the specific physics test much more efficiently, likely due to its massive cache structure.

Architecture Differences

The two chips diverge sharply in design. The AMD EPYC 9384X uses 32 cores and 64 threads based on Zen 4 architecture, manufactured on a 5 nm process at TSMC, with a total transistor count of 90,160 million spread across 8 dies of 72 mm² each. Its cache hierarchy includes 64 KB of L1 and 1 MB of L2 per core, plus a massive 768 MB of shared L3 cache. This explains its dominance in cache-sensitive workloads like encryption, compression, and string sorting, where large datasets can reside on-die. The AMD part has a base clock of 3.10 GHz and a boost clock of 3.90 GHz, with a TDP of 320 watts, and it supports twelve-channel DDR5 memory providing 460.8 GB/s of bandwidth. It uses the AMD Socket SP5 and offers 128 PCIe Gen 5 lanes.

The Intel Xeon 658X is built on Granite Rapids architecture, also on a 5 nm process but fabricated by Intel itself. It has fewer cores—24 cores and 48 threads—but compensates with higher clocks: 3.00 GHz base and 4.90 GHz boost, which is a full 1.0 GHz higher than the AMD’s boost. The Intel die size is 2x 598 mm², which is significantly larger physical silicon. Its cache is smaller in aggregate: 112 KB of L1 and 2 MB of L2 per core, with 144 MB of shared L3. Memory support is eight-channel DDR5, yielding 409.6 GB/s of bandwidth, which is 51.2 GB/s less than the AMD. The Intel part uses the Intel Socket 4710, also has 128 PCIe Gen 5 lanes, and its TDP is lower at 250 watts. The Intel processor is multiplier-unlocked, while the AMD is not, and the Intel part carries the part number SA2D2.

The Verdict

Based strictly on the data, the Intel Xeon 658X is the default choice for users who prioritize raw speed across a wide variety of tasks, especially those dominated by Cinebench rendering, floating-point math, and single-threaded performance. Its consistent 5.2% lead in all Cinebench tests and 19.1% lead in single-thread performance makes it the better engine for software that is not heavily optimized for massive cache or integer operations. The Intel part also achieves this with a lower TDP of 250 watts compared to the AMD’s 320 watts, which is a measurable efficiency advantage. The Intel part’s launch MSRP is $1699, and it is the newer release, dated 2026-02-01, versus the AMD’s 2023-06-12 launch.

The AMD EPYC 9384X is the choice for specialized workloads where its 768 MB of L3 cache and 32 cores can be fully utilized. The 44.2% lead in PassMark physics and 38.7% lead in data encryption are not marginal—they are transformative for those specific applications. If a workload involves heavy encryption, large-scale integer math, or data compression, the AMD part’s 5.5% to 38.7% advantages over the Intel part justify its selection despite losing the majority of benchmarks. The AMD part also holds a slight edge in average benchmark score (120427 vs 116060), indicating that its wins are in heavier-weight tasks. However, the Intel part is the more balanced performer, and its 12 benchmark wins out of 17 make it the safer recommendation for mixed-use servers.

FAQ

Q: Which processor has more cores and threads?

A: The AMD EPYC 9384X has 32 cores and 64 threads, while the Intel Xeon 658X has 24 cores and 48 threads.

Q: What is the largest performance difference between the two?

A: The largest margin is in PassMark physics, where the AMD EPYC 9384X leads by 44.2% with a score of 9332 versus the Intel Xeon 658X’s 6470.

Q: Does the Intel Xeon 658X win any single-threaded benchmarks?

A: Yes, the Intel part wins the PassMark single-thread test with a score of 3728, which is a 19.1% lead over the AMD’s 3015. It also wins all three Cinebench single-core tests by 5.2% each.

Q: Which processor has higher memory bandwidth?

A: The AMD EPYC 9384X has higher memory bandwidth at 460.8 GB/s, compared to the Intel Xeon 658X’s 409.6 GB/s.

Q: Are both processors unlocked for overclocking?

A: No, only the Intel Xeon 658X is multiplier-unlocked. The AMD EPYC 9384X is not unlocked.

Q: How does the cache size compare?

A: The AMD EPYC 9384X has a much larger L3 cache of 768 MB shared, while the Intel Xeon 658X has 144 MB shared. The Intel part has larger per-core L1 and L2 caches: 112 KB and 2 MB per core, versus the AMD’s 64 KB and 1 MB per core.

Head-to-Head Benchmarks

The Cinebench suite shows a uniform pattern: the Intel Xeon 658X wins every test by exactly 5.2%. In Cinebench R15 multi-core, the Intel scores 6296 against the AMD’s 5968. The R20 multi-core result is 26235 for Intel versus 24870 for AMD, and R23 multi-core shows 62466 for Intel versus 59215 for AMD. The single-core results follow suit: R15 single-core is 888 for Intel versus 842 for AMD, R20 single-core is 3703 versus 3510, and R23 single-core is 8818 versus 8359. This consistency indicates that the Intel architecture provides a uniform clock-for-clock advantage in Cinebench workloads, likely due to its higher 4.90 GHz boost clock versus the AMD’s 3.90 GHz.

The PassMark results tell a different story. The AMD EPYC 9384X wins data encryption by a landslide, scoring 72631 against the Intel’s 52357, a 38.7% delta. In PassMark physics, the AMD’s 9332 dwarfs the Intel’s 6470, a 44.2% delta. The AMD also wins random string sorting with 119440 versus 103028 (15.9% delta) and integer math with 297833 versus 263995 (12.8% delta). Its win in data compression is narrower: 1119983 versus 1062062, a 5.5% delta. Conversely, the Intel wins floating-point math decisively at 210480 versus 174630 (17% delta), extended instructions at 84626 versus 74363 (12.1% delta), and find-prime-numbers at 649 versus 596 (8.2% delta). The PassMark multithread test goes to Intel at 73490 versus 69665 (5.2% delta), and single-thread goes to Intel at 3728 versus 3015 (19.1% delta). The data shows that while the Intel part wins more tests, the AMD part’s wins are often by larger margins in specialized areas, particularly encryption and physics, where its 768 MB L3 cache and 32-core configuration provide an outsized benefit.