AMD EPYC 9174F vs AMD Ryzen Threadripper 3990X Comparison

Head-to-Head Benchmarks

The benchmark data presents an unusual head-to-head picture: the AMD Ryzen Threadripper 3990X wins seven of the eight shared tests, yet the AMD EPYC 9174F takes the single most decisive victory in raw single-thread performance. The Cinebench suite is a clean sweep for the 3990X, with identical 46.5% margins across R15, R20, and R23 multicore tests. In Cinebench R23 multicore, the 3990X scores 68124 against the EPYC's 46489 — a gap that reflects the 64-core versus 16-core core-count disparity. Single-core Cinebench results follow the same pattern: the 3990X leads by 46.6% in both R15 (969 vs 661) and R20 (4039 vs 2756), and by 46.5% in R23 (9617 vs 6563). These consistent margins across every Cinebench iteration indicate a structural advantage, not a workload-specific quirk.

The Geekbench results break the trend. In Geekbench multicore, the 3990X wins again, but by a narrower 20.4% (20912 vs 17363). This reduced margin suggests that Geekbench's multicore scaling is less sensitive to the 3990X's massive thread count. More striking is Geekbench single-core, where the EPYC 9174F wins by 34.4% (2244 vs 1472). This is the only test the EPYC takes, and it is a substantial reversal — the EPYC's Zen 4 architecture delivers a single-thread score that is roughly half again as large as the 3990X's. The data shows a clear split: the 3990X dominates in thread-heavy rendering workloads, while the EPYC asserts supremacy in lightly threaded tasks that reward per-core efficiency.

Notably, the 3990X's average benchmark score across its full suite is 12786, placing it at the 68th percentile of all CPUs. The EPYC's average is 12536, at the 67th percentile. These near-identical aggregate scores obscure the very different performance profiles beneath them. The 3990X's nearest rivals include the Intel Core i3-1210U (deltaPct -0.1), Intel Core i5-8400 (0.2), and Intel Xeon Gold 6348 (0.3). The EPYC's nearest rivals include the Intel Core i5-1230U (-0.2) and Intel Core i3-1215U (-0.5). Both processors sit in a dense cluster of similarly-scoring parts, which underscores how average scores can mask the divergent strengths revealed in individual tests.

Where Each One Wins

The 3990X is the clear choice for multicore rendering and content creation. Its Cinebench R23 multicore score of 68124 is 46.5% ahead of the EPYC's 46489, and the same margin holds across older Cinebench versions. This consistency across R15, R20, and R23 indicates that the 3990X's advantage scales uniformly with thread count — expected for a 64-core/128-thread part. The 3990X also wins Geekbench multicore by 20.4%, though the smaller margin suggests that Geekbench does not fully saturate all 128 threads. For workloads that are explicitly parallel — video rendering, 3D scene compilation, scientific simulation — the 3990X's 7-of-8 head-to-head win record makes it the data-backed pick.

The EPYC 9174F wins exactly one test, but it is a decisive one. Geekbench single-core at 2244 beats the 3990X's 1472 by 34.4%. This is not a marginal edge; it is a categorical difference in per-thread performance. The EPYC's Zen 4 architecture, with its higher 4.40 GHz boost clock versus the 3990X's 4.30 GHz, contributes to this advantage, though the architectural generational leap is the more significant factor. For single-threaded applications — legacy software, scripting, database queries with limited parallelism, or interactive response times — the EPYC is the superior part. The data shows no other test where the EPYC leads, so its use case is narrow but well-defined.

The 3990X also claims the single-core Cinebench tests, but those results are counterintuitive given the Geekbench outcome. In Cinebench R23 single-core, the 3990X scores 9617 versus the EPYC's 6563 — a 46.5% lead that contradicts the Geekbench single-core result. This discrepancy highlights that "single-core" is not a monolithic metric; different benchmark suites measure different instruction paths and memory access patterns. The 3990X's older Zen 2 architecture apparently handles Cinebench's single-threaded workload more efficiently relative to its multi-core performance, while the EPYC excels in Geekbench's single-threaded test. Buyers should match the benchmark to their actual application rather than rely on a single "single-core" number.

The Verdict

For users whose workloads are dominated by parallel rendering or compute, the AMD Ryzen Threadripper 3990X is the data-driven choice. It wins every Cinebench multicore test by 46.5% and Geekbench multicore by 20.4%. The 3990X's 64 cores and 128 threads, combined with 256 MB of L3 cache, deliver a multicore performance envelope that the 16-core EPYC cannot approach. Its average benchmark score of 12786 is also slightly higher than the EPYC's 12536, and it carries a 68th percentile rank versus the EPYC's 67th. The 3990X is a desktop part, launched with a $3990 MSRP, and its unlocked multiplier offers overclocking flexibility that the locked EPYC lacks.

The AMD EPYC 9174F is the pick for single-threaded performance. Its Geekbench single-core score of 2244 is 34.4% ahead of the 3990X, a margin that will translate to faster response times in applications that cannot leverage many cores. The EPYC also brings server-grade features absent from the 3990X: twelve-channel DDR5 memory with 460.8 GB/s bandwidth versus quad-channel DDR4 at 102.4 GB/s, ECC memory support, and 128 PCIe Gen 5 lanes versus 64 Gen 4 lanes. For a server or workstation running lightly threaded database workloads or interactive applications, the EPYC's single-core supremacy and platform features justify its $3850 launch MSRP.

The verdict is not about which processor is "better" overall — their average scores are nearly identical — but about which workload profile matters more. The 3990X is a rendering monster with a 7-1 head-to-head win record. The EPYC is a precision instrument for single-threaded tasks and memory-bandwidth-hungry server applications. If the choice is between the two with no other context, the data favors the 3990X for its broader multicore dominance. If the application is known to be single-threaded, the EPYC is the only rational selection.

FAQ

Q: Which processor has the higher average benchmark score?

A: The AMD Ryzen Threadripper 3990X has an average benchmark score of 12786, while the AMD EPYC 9174F averages 12536. The 3990X also ranks at the 68th percentile of all CPUs, versus the EPYC's 67th.

Q: How large is the multicore performance gap in Cinebench R23?

A: The 3990X scores 68124 in Cinebench R23 multicore, which is 46.5% higher than the EPYC's 46489. The same 46.5% margin appears in Cinebench R15 and R20 multicore tests.

Q: In which benchmark does the EPYC 9174F beat the 3990X?

A: The EPYC wins Geekbench single-core with a score of 2244 against the 3990X's 1472, a 34.4% advantage. This is the only head-to-head test the EPYC wins out of eight.

Q: What memory and PCIe differences exist between the two?

A: The 3990X supports quad-channel DDR4 with 102.4 GB/s bandwidth and 64 PCIe Gen 4 lanes. The EPYC supports twelve-channel DDR5 with 460.8 GB/s bandwidth and 128 PCIe Gen 5 lanes. The EPYC also supports ECC memory; the 3990X does not.

Q: Are both processors currently in production?

A: Yes, both the AMD Ryzen Threadripper 3990X and the AMD EPYC 9174F have an "Active" production status. The 3990X was released on 2020-02-06, and the EPYC on 2022-11-09.

Q: Do the processors share the same socket?

A: No. The 3990X uses AMD Socket TRX4, while the EPYC uses AMD Socket SP5. The 3990X has an unlocked multiplier; the EPYC's multiplier is locked.

Architecture Differences

The two processors represent different generations of AMD architecture. The 3990X is built on Zen 2, codenamed Castle Peak, using TSMC's 7 nm process with 30,400 million transistors across 8x 74 mm² dies. The EPYC uses Zen 4, codenamed Genoa, on a 5 nm process with 52,560 million transistors across 8x 72 mm² dies. The newer 5 nm node gives the EPYC a transistor density advantage, but the 3990X compensates with a 4x core-count advantage: 64 cores and 128 threads versus 16 cores and 32 threads.

Cache hierarchies differ substantially. The 3990X has 64 KB of L1 and 512 KB of L2 per core, with a 256 MB L3 cache. The EPYC also has 64 KB of L1 per core but doubles L2 to 1 MB per core, with 256 MB of shared L3. The larger per-core L2 on the EPYC supports its single-thread performance, while the shared L3 sizes are identical. The 3990X's base clock is 2.90 GHz with a 4.30 GHz boost; the EPYC's base clock is 4.10 GHz with a 4.40 GHz boost. The EPYC's higher clocks, combined with Zen 4's architectural improvements, explain its Geekbench single-core lead despite having fewer cores.

Memory subsystems are fundamentally different. The 3990X runs quad-channel DDR4 with 102.4 GB/s bandwidth; the EPYC runs twelve-channel DDR5 with 460.8 GB/s — a 4.5x bandwidth advantage. This memory bandwidth difference is critical for server workloads that stream large datasets. PCIe connectivity also favors the EPYC: 128 Gen 5 lanes versus 64 Gen 4 lanes. The EPYC supports ECC memory, a requirement for many server deployments; the 3990X does not. The TDPs reflect their market segments: the 3990X draws 280 W as a desktop part, while the EPYC draws 320 W as a server part. The EPYC's socket is SP5, and the 3990X uses TRX4. Both are manufactured by TSMC, but the EPYC's 5 nm process is two generations ahead of the 3990X's 7 nm.