AMD EPYC 9275F vs AMD Ryzen Threadripper PRO 9965WX Comparison

The AMD Ryzen Threadripper PRO 9965WX and AMD EPYC 9275F are both 24-core, 48-thread Zen 5 parts on TSMC's 4 nm node, but they are built for different sockets and different jobs. Head-to-head benchmark data shows the Threadripper PRO 9965WX winning 15 of 17 tests, with the EPYC 9275F taking only two. The Threadripper's average benchmark score is 146814, which places it 10.2% ahead of the EPYC's 133174 average, and both chips sit at the 98th percentile among all CPUs. This is not a close fight in raw throughput, but the EPYC's two wins point to specific workloads where its architecture advantages matter more than raw clock speed.

Head-to-Head Benchmarks

The Threadripper PRO 9965WX wins every Cinebench test by exactly 12.6%. In Cinebench R23 multi-core, it scores 80976 against the EPYC's 71927; in single-core, 11431 versus 10154. The same 12.6% margin appears in R15 and R20, both multi-core and single-core variants. This consistency across all Cinebench versions indicates a clock-speed and IPC advantage that scales uniformly across rendering workloads.

PassMark results tell a similar story but with larger swings. The biggest single-thread gap is in PassMark single-thread, where the Threadripper scores 4555 versus the EPYC's 3810 — a 19.6% lead. Integer math also favors the Threadripper heavily: 365422 versus 317777, a 15% difference. Data encryption shows a 14.7% edge (71873 vs 62664), and random string sorting is 14.1% higher (164382 vs 144037). Multithread performance is 12.7% ahead (95346 vs 84620), floating-point math is 9% ahead (220036 vs 201888), data compression is 8.4% ahead (1314596 vs 1212560), and extended instructions are 8.1% ahead (102589 vs 94889).

The EPYC 9275F wins two PassMark tests. The most dramatic is find prime numbers: the EPYC scores 991 versus the Threadripper's 682, meaning the Threadripper trails by 31.2%. Physics is the other EPYC win, with 12089 versus 11278, a 6.7% margin. These are the only two places where the EPYC's architecture overcomes the Threadripper's clock advantage. The prime-number result is particularly striking because the EPYC wins by a wide margin despite losing most other compute-heavy tests.

Architecture Differences

Both CPUs are Zen 5 on TSMC's 4 nm process, but the silicon layouts diverge significantly. The Threadripper PRO 9965WX uses 4x 70.6 mm² dies with 33,260 million transistors, while the EPYC 9275F uses 8x 70.6 mm² dies with 66,520 million transistors — exactly double the die count and nearly double the transistor count. The EPYC's extra dies provide a much larger L3 cache: 256 MB shared versus 128 MB shared on the Threadripper. L1 cache also differs: 80 KB per core on the EPYC versus 64 KB per core on the Threadripper. L2 cache is identical at 1 MB per core.

Memory subsystems are another major split. The Threadripper uses eight-channel DDR5 with 409.6 GB/s of bandwidth, while the EPYC uses twelve-channel DDR5 with 576.0 GB/s. Both support ECC memory and both have 128 Gen 5 PCIe lanes from the CPU. The sockets are different — sTR5 for the Threadripper, SP5 for the EPYC — which means they are not interchangeable in a system.

Clock speeds favor the Threadripper. Its base clock is 4.20 GHz and boost is 5.40 GHz, versus 4.10 GHz base and 4.80 GHz boost on the EPYC. TDP is also higher on the Threadripper at 350 W versus 320 W. The Threadripper has an unlocked multiplier; the EPYC does not. Release dates differ: the Threadripper launched on 2025-07-22, the EPYC on 2024-10-09. The launch MSRP for the Threadripper PRO 9965WX is $2899, and for the EPYC 9275F it is $3439.

The Verdict

The data is unambiguous for most workloads: the Threadripper PRO 9965WX is the faster CPU. It wins every Cinebench test, every PassMark test except two, and holds a 10.2% average-score advantage over the EPYC 9275F. If you are building a workstation for rendering, encoding, compilation, or general compute, the Threadripper is the pick based on benchmark results alone. Its 19.6% single-thread lead and 15% integer-math lead mean interactive and latency-sensitive tasks will feel noticeably faster.

The EPYC 9275F is not a bad chip — it wins prime-number finding by 31.2% and physics by 6.7% — but those are narrow niches. Its twelve-channel memory bus and 576.0 GB/s bandwidth, combined with 256 MB of L3, give it a clear edge in memory-bound server workloads that the PassMark physics and prime-number tests happen to capture. For a dual-socket server rack or a memory-bandwidth-hungry database workload, the EPYC's architecture is the rational choice. For a single-socket workstation where clock speed and per-thread performance matter, the Threadripper dominates.

The TDP difference of 30 W (350 W vs 320 W) is minor, but the socket difference is not — you cannot swap these between platforms. Choose the Threadripper for sTR5 workstations, choose the EPYC for SP5 servers. The EPYC's higher launch MSRP of $3439 versus $2899 for the Threadripper does not buy more performance in most benchmarks; it buys platform features like twelve-channel memory and a larger L3 cache.

Specification Differences

| Field | AMD Ryzen Threadripper PRO 9965WX | AMD EPYC 9275F |

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

| Socket | AMD Socket sTR5 | AMD Socket SP5 |

| Codename | Shimada Peak | Turin |

| Base Clock | 4.20 GHz | 4.10 GHz |

| Boost Clock | 5.40 GHz | 4.80 GHz |

| TDP | 350 W | 320 W |

| Transistors | 33,260 million | 66,520 million |

| Die Size | 4x 70.6 mm² | 8x 70.6 mm² |

| L1 Cache | 64 KB (per core) | 80 KB (per core) |

| L3 Cache | 128 MB (shared) | 256 MB (shared) |

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

| Memory Bandwidth | 409.6 GB/s | 576.0 GB/s |

| Release Date | 2025-07-22 | 2024-10-09 |

| Launch MSRP | $2899 | $3439 |

| Multiplier Unlocked | Yes | No |

| Part Number | 100-000000724 | 100-000001144 |

Both CPUs share Zen 5 architecture, 4 nm TSMC process, 24 cores, 48 threads, 1 MB L2 per core, DDR5 memory support, ECC memory support, and 128 Gen 5 PCIe lanes. The EPYC has more dies, more transistors, more L3, and more memory bandwidth; the Threadripper has higher clocks, a lower launch MSRP, and an unlocked multiplier.

FAQ

Q: Which CPU wins more benchmark tests?

A: The AMD Ryzen Threadripper PRO 9965WX wins 15 of 17 head-to-head benchmarks, while the AMD EPYC 9275F wins 2.

Q: What is the single-thread performance difference?

A: In PassMark single-thread, the Threadripper PRO 9965WX scores 4555 versus the EPYC 9275F's 3810, a 19.6% lead. Cinebench R23 single-core also shows a 12.6% edge (11431 vs 10154).

Q: Does the EPYC 9275F win any tests?

A: Yes. It wins PassMark find prime numbers (991 vs 682, a 31.2% margin) and PassMark physics (12089 vs 11278, a 6.7% margin).

Q: How does memory bandwidth compare?

A: The EPYC 9275F has a twelve-channel memory bus with 576.0 GB/s bandwidth, while the Threadripper PRO 9965WX has eight-channel memory with 409.6 GB/s.

Q: Which CPU has more L3 cache?

A: The EPYC 9275F has 256 MB shared L3, double the Threadripper PRO 9965WX's 128 MB shared L3.

Q: Are these CPUs on the same socket?

A: No. The Threadripper PRO 9965WX uses AMD Socket sTR5, and the EPYC 9275F uses AMD Socket SP5.

Where Each One Wins

The Threadripper PRO 9965WX is the clear winner for general-purpose workstation compute. It leads in all Cinebench versions (R15, R20, R23) by 12.6%, which covers rendering, 3D animation, and video encoding. PassMark integer math (15% ahead) and floating-point math (9% ahead) make it the better choice for scientific computing, financial modeling, and engineering simulation. Data encryption (14.7% ahead) and data compression (8.4% ahead) suit it for security workloads and archive management. Its 19.6% single-thread lead and 12.7% multithread lead mean it excels at both interactive tasks and heavily threaded batch jobs. The unlocked multiplier is a bonus for overclocking on compatible sTR5 boards.

The EPYC 9275F wins in exactly two measured areas. PassMark find prime numbers shows a 31.2% advantage, which translates to prime-number generation and cryptographic key search workloads. PassMark physics shows a 6.7% advantage, indicating better performance in physics simulations and collision detection. These wins align with the EPYC's larger L3 cache (256 MB vs 128 MB) and higher memory bandwidth (576.0 GB/s vs 409.6 GB/s), which benefit workloads that repeatedly access large datasets. For a server running memory-resident databases, large-scale virtualization, or physics-based HPC jobs, the EPYC's platform features are the right fit — but for raw benchmark performance per dollar, the Threadripper is the stronger choice.