AMD EPYC 9734 vs AMD Ryzen Threadripper PRO 9985WX Comparison

The AMD Ryzen Threadripper PRO 9985WX and AMD EPYC 9734 are both extreme server/workstation parts, but the benchmark data shows they are built for different jobs. The Threadripper PRO 9985WX wins 12 of the 17 head-to-head tests, while the EPYC 9734 takes 5. The decisive factor is not core count—the EPYC has 112 cores to the Threadripper’s 64—but clock speed and architecture generation. The Threadripper’s average benchmark score of 318888 puts it 2.7% ahead of the EPYC’s 310619, and its 100th percentile ranking vs. the EPYC’s 99th confirms it edges out the server chip in overall throughput. But the EPYC’s wins in data compression, encryption, extended instructions, floating-point math, and random string sorting show where raw core density still matters.

Where Each One Wins

The Threadripper PRO 9985WX dominates every Cinebench test, both multi-core and single-core, with a consistent 52.8% lead across R15, R20, and R23. This is the chip for rendering, simulation, and any workload that scales with per-core performance and high boost clocks. Its 5.40 GHz boost clock versus the EPYC’s 3.00 GHz explains the single-thread gap of 98.5% in PassMark’s single-thread test (4586 vs 2310). The Threadripper also wins PassMark integer math by 7.9% (888032 vs 823150), find prime numbers by 44.3% (1196 vs 829), and PassMark multithread by 52.8% (156305 vs 102286). The physics test is the biggest landslide: 19284 vs 6747, a 185.8% advantage.

The EPYC 9734 wins where its 112 cores and twelve-channel memory bus (460.8 GB/s) provide an edge. PassMark data compression goes to the EPYC at 2900008 vs 2851752, a 1.7% margin. Data encryption sees a 5.7% win (179390 vs 169124). Extended instructions are nearly tied, with the EPYC ahead by 0.6% (205925 vs 204637). Floating-point math favors the EPYC by 1.9% (549045 vs 538413), and random string sorting by 1.4% (357638 vs 352614). These are narrow wins but consistent across memory-bandwidth-sensitive and highly parallel integer workloads.

Architecture Differences

The Threadripper PRO 9985WX uses Zen 5 architecture on a 4 nm TSMC process, codenamed Shimada Peak. It has 64 cores and 128 threads, with a base clock of 3.20 GHz and boost of 5.40 GHz. Its die size is 8x 70.6 mm² with 66,520 million transistors. The EPYC 9734 uses Zen 4c architecture on a 5 nm TSMC process, codenamed Bergamo. It packs 112 cores and 224 threads, but clocks are much lower: 2.20 GHz base and 3.00 GHz boost. Its die size is 8x 73 mm² with 71,000 million transistors.

Both chips have identical cache layouts: 64 KB L1 per core, 1 MB L2 per core, and 256 MB shared L3. Memory support is DDR5 for both, with ECC enabled on both. The EPYC uses a twelve-channel memory bus with 460.8 GB/s bandwidth, while the Threadripper uses eight channels with 409.6 GB/s. Both support PCIe Gen 5 with 128 lanes from the CPU. The Threadripper has an unlocked multiplier; the EPYC does not. The Threadripper’s socket is AMD Socket sTR5, while the EPYC uses AMD Socket SP5.

The core count difference is stark: 112 vs 64, a 75% advantage for the EPYC. Yet the Threadripper still wins the average benchmark score. This is a direct consequence of the Zen 5 vs Zen 4c split—the Threadripper’s cores run much faster, and the data shows that per-core speed matters more than core count for most of the benchmarks in this comparison. The EPYC’s 71,000 million transistors on a larger 5 nm process reflect its denser core layout, but that density does not translate into higher scores outside of specific workloads.

Head-to-Head Benchmarks

The Cinebench suite is a total sweep for the Threadripper PRO 9985WX. In Cinebench R15 multi-core, it scores 13392 vs 8763, a 52.8% lead. Single-core R15 mirrors that: 1890 vs 1237, again 52.8%. The same pattern holds for R20 multi-core (55800 vs 36516) and single-core (7877 vs 5155), and for R23 multi-core (132859 vs 86943) and single-core (18756 vs 12274). Every single Cinebench delta is exactly 52.8%, which indicates the Threadripper’s advantage is not workload-specific but a fundamental per-clock performance edge.

In PassMark tests, the Threadripper wins multithread by 52.8% (156305 vs 102286) and single-thread by 98.5% (4586 vs 2310). The physics test is the outlier at 185.8% (19284 vs 6747), suggesting the Threadripper’s high boost clocks massively accelerate physics simulations that are often latency-bound. Integer math goes to the Threadripper by 7.9% (888032 vs 823150), and find prime numbers by 44.3% (1196 vs 829).

The EPYC’s wins are all single-digit margins. Data compression: 2900008 vs 2851752 (1.7% ahead). Data encryption: 179390 vs 169124 (5.7% ahead). Extended instructions: 205925 vs 204637 (0.6% ahead). Floating-point math: 549045 vs 538413 (1.9% ahead). Random string sorting: 357638 vs 352614 (1.4% ahead). These are the only five tests where the EPYC comes out on top, and none of them exceed a 5.7% delta. The Threadripper’s wins include margins of 44.3%, 52.8%, 98.5%, and 185.8%—far more decisive.

The Verdict

The data is unambiguous: the AMD Ryzen Threadripper PRO 9985WX is the faster processor in the majority of workloads. Its 12-5 win tally in head-to-head tests, combined with a 2.7% higher average benchmark score (318888 vs 310619), makes it the better choice for anyone running Cinebench-style rendering, physics simulations, integer-heavy tasks, or any single-threaded application. The 52.8% Cinebench leads are not marginal—they are generational. The Threadripper’s 5.40 GHz boost clock and Zen 5 architecture deliver results that the EPYC’s 3.00 GHz Zen 4c cores cannot match, despite the EPYC having 48 more cores.

The EPYC 9734 is not without merit. Its wins in data compression, encryption, floating-point math, and random string sorting show that certain server workloads—particularly those that saturate memory bandwidth or benefit from massive core counts—will see a slight edge. The twelve-channel memory bus (460.8 GB/s) gives it a 12.5% bandwidth advantage over the Threadripper’s eight-channel 409.6 GB/s. For workloads that are purely memory-bound and highly parallel, the EPYC’s narrow wins (all under 6%) could add up over long runs.

Choose the Threadripper if your work is rendering, simulation, physics, or any mix of single-thread and multi-thread performance. Choose the EPYC if you are running dedicated server tasks like encryption, compression, or floating-point-heavy batch processing where the extra cores and wider memory bus provide a measurable—if small—benefit. The Threadripper’s unlocked multiplier is another differentiator for overclocking, though the EPYC’s locked design is typical for server stability.

FAQ

Q: Which processor has more cores?

A: The AMD EPYC 9734 has 112 cores and 224 threads, compared to the AMD Ryzen Threadripper PRO 9985WX’s 64 cores and 128 threads.

Q: Why does the Threadripper win Cinebench if it has fewer cores?

A: The Threadripper’s 5.40 GHz boost clock and Zen 5 architecture on a 4 nm process deliver a 52.8% advantage in every Cinebench test, despite having 48 fewer cores than the EPYC’s Zen 4c design.

Q: What is the largest single benchmark margin between the two?

A: PassMark physics shows the Threadripper ahead by 185.8% (19284 vs 6747), the widest gap in the head-to-head data.

Q: Does the EPYC win any benchmark by a large margin?

A: No. The EPYC’s largest win is PassMark data encryption at 5.7% (179390 vs 169124). All five of its wins are under 6%.

Q: How do their memory systems differ?

A: The EPYC uses a twelve-channel DDR5 bus with 460.8 GB/s bandwidth, while the Threadripper uses eight-channel DDR5 at 409.6 GB/s. Both support ECC memory.

Q: Are both CPUs still in production?

A: Yes, both are listed with Active production status. The Threadripper has an unlocked multiplier; the EPYC does not.

Specification Differences

| Field | AMD Ryzen Threadripper PRO 9985WX | AMD EPYC 9734 |

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

| Cores | 64 | 112 |

| Threads | 128 | 224 |

| Base Clock | 3.20 GHz | 2.20 GHz |

| Boost Clock | 5.40 GHz | 3.00 GHz |

| TDP | 350 W | 340 W |

| Socket | AMD Socket sTR5 | AMD Socket SP5 |

| Architecture | Zen 5 | Zen 4 |

| Codename | Shimada Peak | Bergamo |

| Process Node | 4 nm | 5 nm |

| Transistors | 66,520 million | 71,000 million |

| Die Size | 8x 70.6 mm² | 8x 73 mm² |

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

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

| Multiplier Unlocked | Yes | No |

| Launch MSRP | $7999 | $9600 |