AMD Ryzen Threadripper PRO 9985WX vs Intel Xeon 6781P Comparison

The AMD Ryzen Threadripper PRO 9985WX and Intel Xeon 6781P are both 350W, eight-channel DDR5 workstation behemoths, but benchmark data reveals two very different performance personalities. The AMD part wins 12 of 14 head-to-head comparisons, while the Intel part takes two specialized workloads. These are 99th-percentile CPUs by any measure, yet the data shows a clear overall leader.

Head-to-Head Benchmarks

The most striking pattern in the data is the near-uniform 32.5% margin in Cinebench multi-core tests. The Ryzen Threadripper PRO 9985WX scores 13,392 in Cinebench R15 multi-core versus 10,105 for the Xeon 6781P. That exact 32.5% delta repeats in Cinebench R20 (55,800 vs 42,106) and Cinebench R23 (132,859 vs 100,254). This consistency suggests a fundamental architectural advantage rather than workload-specific luck.

The largest single win for AMD comes in PassMark integer math, where the Threadripper PRO scores 872,710 against 584,834 — a 49.2% lead. This is the kind of gap that shows up in compilation, encoding, and general compute tasks. Single-thread performance also heavily favors AMD: 4,482 versus 3,152 in PassMark single-thread, a 42.2% advantage. That is a massive difference for a workstation CPU and explains why the AMD chip feels snappier in lightly-threaded applications despite having fewer cores.

Memory and data-heavy workloads follow the same trend. PassMark data compression shows a 19.3% win for AMD (2,912,972 vs 2,441,690), while data encryption gives AMD a 29.4% edge (154,824 vs 119,623). Random string sorting, a cache-latency sensitive test, goes to AMD by 22.5% (329,014 vs 268,573). Floating-point math is closer — AMD wins by 9.1% (553,348 vs 507,406) — but it is still a clear victory. The extended instructions test (SIMD-heavy) goes to AMD by 13.2% (225,340 vs 199,048), and the multithreaded PassMark score shows a 27.2% advantage (150,071 vs 117,946).

The Xeon 6781P's two wins are narrow but interesting. In PassMark find prime numbers, Intel wins 1,687 to 1,138 — a 32.5% margin in its favor, which exactly mirrors AMD's Cinebench advantage. In PassMark physics, Intel scores 17,753 versus 13,783, a 22.4% win. These are specific, compute-bound integer workloads where the Xeon's higher core count (80 vs 64) appears to overcome its clock speed disadvantage.

Looking at average benchmark scores, the AMD chip sits at 320,749 versus 315,524 for Intel — a 1.7% gap according to AMD's nearest-rival data, and -1.6% from Intel's perspective. The Threadripper PRO 9985WX's closest rival is actually the AMD Ryzen Threadripper 9980X at 321,753 (a 0.3% difference), meaning the two workstation chips under discussion are nearly tied in aggregate performance, with the Intel part also trailing the EPYC 9575F by 1.2% and the EPYC 9734 by 1.6%.

FAQ

Q: Which CPU has more cores and threads?

A: The Intel Xeon 6781P has 80 cores and 160 threads, while the AMD Ryzen Threadripper PRO 9985WX has 64 cores and 128 threads.

Q: Why does the AMD chip win most benchmarks despite fewer cores?

A: The AMD part has a much higher boost clock (5.40 GHz vs 3.80 GHz) and a larger L3 cache per core configuration, which shows up in the 42.2% single-thread win and 32.5% Cinebench multi-core margins.

Q: What benchmark does the Intel Xeon win by its biggest margin?

A: The PassMark find prime numbers test, where Intel leads by 32.5% (1,687 vs 1,138). This is a pure integer computation workload.

Q: Are both CPUs in the same market segment?

A: Yes, both are listed as Server/Workstation processors with active production status and 99th percentile rankings against all CPUs.

Q: Do both support the same memory configuration?

A: Both support DDR5 with eight-channel memory buses and 409.6 GB/s bandwidth, and both have ECC memory support.

Q: Is there a difference in PCIe lane counts?

A: Yes — the Intel Xeon 6781P provides 136 Gen 5 lanes (CPU only), while the AMD Threadripper PRO 9985WX provides 128 Gen 5 lanes (CPU only).

Architecture Differences

The AMD Ryzen Threadripper PRO 9985WX is built on TSMC's 4 nm process with a transistor count of 66,520 million spread across 8 dies, each 70.6 mm². It uses the Zen 5 architecture with the codename Shimada Peak, part of the 9000 series. In contrast, the Intel Xeon 6781P uses Intel's own 5 nm process with a die size of 2x 598 mm² (no transistor count is listed), based on the Granite Rapids architecture with the Granite Rapids-SP generation codename.

Cache hierarchies differ substantially. The AMD chip has 64 KB L1 and 1 MB L2 per core, plus a 256 MB L3 cache. The Intel chip has 112 KB L1 and 2 MB L2 per core, with a larger 336 MB shared L3 cache. This means Intel allocates more cache per core on L1/L2, while AMD's total L3 is smaller but its per-core L3 ratio is higher relative to its 64-core count.

The AMD part is a multi-die design (8x 70.6 mm² chiplets), while the Xeon uses two larger dies (2x 598 mm²). The AMD chip uses Socket sTR5, and the Intel chip uses Socket 4710. Both are monolithic in the sense of having no 3D V-Cache, and both lack integrated graphics. The AMD chip has an unlocked multiplier, while the Intel chip is locked. Their release dates differ by five months: the Xeon launched on 2025-02-23, and the Threadripper PRO on 2025-07-22.

Specification Differences

| Specification | AMD Ryzen Threadripper PRO 9985WX | Intel Xeon 6781P |

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

| Cores | 64 | 80 |

| Threads | 128 | 160 |

| Base Clock | 3.20 GHz | 2.00 GHz |

| Boost Clock | 5.40 GHz | 3.80 GHz |

| Process Node | 4 nm (TSMC) | 5 nm (Intel) |

| Foundry | TSMC | Intel |

| Die Size | 8x 70.6 mm² | 2x 598 mm² |

| Transistors | 66,520 million | Not listed |

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

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

| L3 Cache | 256 MB | 336 MB (shared) |

| Socket | AMD Socket sTR5 | Intel Socket 4710 |

| PCIe | Gen 5, 128 Lanes (CPU only) | Gen 5, 136 Lanes (CPU only) |

| Multiplier | Unlocked | Locked |

| Launch MSRP | $7999 | $8960 |

| Part Number | 100-000000722 | SRV5J |

Where Each One Wins

The AMD Ryzen Threadripper PRO 9985WX is the clear choice for general workstation compute. Its 32.5% lead across all three Cinebench multi-core tests indicates superior all-core rendering performance. The 49.2% win in integer math makes it the better pick for code compilation, scientific computing, and financial modeling. The 42.2% single-thread advantage means it will feel faster in everyday desktop tasks, CAD single-threaded operations, and legacy software that doesn't scale well across cores. For data compression, encryption, and random string sorting, AMD wins by margins between 19.3% and 29.4%, making it the stronger choice for database workloads and archival tasks. The 27.2% multithread win and 9.1% floating-point win round out a dominant profile for rendering, simulation, and machine learning inference workloads.

The Intel Xeon 6781P wins specifically in PassMark physics (17,753 vs 13,783) and PassMark find prime numbers (1,687 vs 1,138). These are heavily integer-bound, branch-heavy workloads where the Xeon's 80 cores and 160 threads provide a raw count advantage. The physics test suggests the Xeon handles rigid-body and particle simulations better, which could matter for certain engineering and scientific workloads. The prime number test indicates a strength in modular arithmetic and cryptography-adjacent computations. Additionally, the Xeon offers 136 PCIe Gen 5 lanes versus 128 for AMD, which matters for systems with many GPUs or NVMe storage devices, and it has a larger 336 MB L3 cache that could benefit workloads with massive data sets that fit within that shared pool.

For most buyers, the data points to AMD. The 32.5% Cinebench margins are decisive for rendering, the 42.2% single-thread lead helps every interactive task, and the 49.2% integer math advantage covers a broad swath of technical computing. The Xeon's wins are real but narrow in scope, and its higher launch MSRP of $8960 versus $7999 does not offset the performance gap in the benchmark data. The only scenarios where the Xeon makes sense are those specific physics simulations and prime-number workloads, or when the extra 8 PCIe lanes and 80 MB of L3 cache tip the balance for a particular system configuration.