AMD Ryzen Threadripper PRO 3975WX

The AMD Ryzen Threadripper PRO 3975WX is a 32-core, 64-thread Zen 2 desktop processor in the 3000 series, built on TSMC’s 7 nm process with a 4x 74 mm² die design and 15,200 million transistors. Its launch MSRP is $2749, and it was released on 2020-07-13. The benchmark data paints a clear picture: this is a multi-threaded computing tool, with far stronger multi-core scores than single-core scores, and its 72nd-percentile aggregate ranking reflects that split.

Who Should Consider It

Users running heavily multi-threaded workloads should consider the 3975WX first. The Cinebench R23 multi-core score of 52,895 and Geekbench multi-core score of 14,828 are the standout results in the data. Applications that scale across many cores, such as those modeled by Cinebench R15, R20, and R23 multi-thread tests, will be the main beneficiaries. The 32 cores and 64 threads give workloads a large parallel surface, and the 128 MB L3 cache provides substantial capacity for data-heavy processing.

Gaming is not the primary strength indicated by the benchmark results. The single-core scores in Cinebench and Geekbench are moderate relative to the multi-core scores. Many gaming workloads depend on strong single-thread performance, and the data does not place the 3975WX in that leading single-core class. It can still run games, but users whose main requirement is high single-thread performance will find the chip’s aggregate profile less compelling.

Office and lightly threaded productivity workloads are likewise not the ideal match. The single-thread scores — Cinebench R15 single-core 752, R20 single-core 3,136, R23 single-core 7,467, and Geekbench single-core 1,648 — are respectable but not exceptional. For tasks that use one or a few threads, the 3975WX’s large core count is mostly idle. The processor is best suited to creators, engineers, analysts, and other users who can keep many cores busy simultaneously.

Power and Thermals

The TDP is 280W, placing the 3975WX in a high-power desktop class. That power envelope, combined with a 7 nm process and a 4x 74 mm² die layout, requires a serious cooling solution. Sustained multi-threaded loads will stress the cooler, and the data implies a high-end cooling tier rather than a modest stock-style cooler.

The 280W TDP is a central system design constraint. The processor is actively in production, but that does not reduce the thermal burden. Builders should plan for robust cooling and adequate case airflow to handle long all-core workloads. The absence of integrated graphics also means a discrete GPU is required, which adds another component to the overall thermal and power picture.

Platform and Compatibility

The 3975WX uses the AMD Socket WRX8 and is part of the Zen 2 Castle Peak generation. It supports DDR4 memory with an eight-channel bus and 204.8 GB/s of memory bandwidth, and ECC memory is enabled. This memory configuration is a major advantage for large, memory-intensive workloads.

PCIe support is Gen 4 with 128 lanes available from the CPU. That is a large amount of I/O bandwidth for expansion cards, NVMe storage, and other peripherals. The processor has no integrated graphics listed, so a discrete GPU is necessary. The multiplier is not unlocked, meaning overclocking is not a targeted feature in the data.

The upgrade path is defined by the WRX8 socket. The processor is listed as active production, and its release date is 2020-07-13. The cache hierarchy is 64 KB L1 per core, 512 KB L2 per core, and 128 MB L3. This platform is built for high-end desktop workstations rather than mainstream systems, and any compatibility decisions should start with the WRX8 socket requirement.

How It Compares

The 3975WX’s average benchmark score is 13,534, and the nearest rivals in the data are all notable because their aggregate scores nearly match that number despite very different specifications.

Against the Intel Core i3-12100F, the 3975WX is 0.1% ahead on aggregate average score, with the rival averaging 13,516. This negligible delta shows that an overall average across Cinebench and Geekbench tests can conceal a major workload-shape difference. The two processors are not competitors in the same class, but their averaged scores land close together.

The Intel Core i7-1250U averages 13,462, putting the 3975WX 0.5% ahead. Again, the aggregate gap is tiny. The 3975WX’s extreme multi-thread results are offset by its more moderate single-thread results, allowing a very different processor to appear similar on paper.

The Intel Core i3-10105 averages 13,430, and the 3975WX leads by 0.8%. This is the largest positive delta in the rival group, but it is still small in absolute percentage terms. The benchmark data places these parts within a narrow aggregate band, even though the underlying workload behavior is entirely different.

The AMD Ryzen 5 5500U averages 13,770, putting the 3975WX 1.7% behind on the aggregate metric. This is the only negative delta in the rival set. It illustrates a key point: an average across mixed single-thread and multi-thread benchmarks does not reward the 3975WX for its 32-core/64-thread configuration when single-core results pull the mean down.

Benchmark Performance

The 3975WX produces a Cinebench R15 multi-core score of 5,331 and a single-core score of 752. In Cinebench R20, the multi-core score rises to 22,215 while the single-core score is 3,136. Cinebench R23 shows 52,895 multi-core and 7,467 single-core. Geekbench results are 14,828 multi-core and 1,648 single-core.

The multi-core scores are consistently many thousands of points above the single-core scores. The R23 multi-core result of 52,895 versus the R23 single-core result of 7,467 is the largest gap in the dataset, showing the processor’s strength in fully parallel workloads. The Geekbench multi-core score of 14,828 is also roughly nine times the single-core score of 1,648, though the ratio should not be treated as a direct performance claim.

The overall average benchmark score is 13,534, and the processor sits at the 72nd percentile of all CPUs in the database. That percentile is modest for a 32-core part, but it is a consequence of the averaging method: the huge multi-core scores are blended with the much lower single-core scores. The nearest rival deltas are 0.1%, 0.5%, 0.8%, and -1.7%, all relative to aggregate averages. On any heavily multi-threaded test, the 3975WX’s own scores demonstrate a far larger advantage than the aggregate deltas suggest.

FAQ

Q: Does the AMD Ryzen Threadripper PRO 3975WX support ECC memory?

A: Yes, the data lists ECC memory as enabled, with DDR4 memory support over an eight-channel bus.

Q: What socket does the 3975WX use?

A: It uses the AMD Socket WRX8.

Q: How much PCIe connectivity does it provide?

A: The processor provides PCIe Gen 4 with 128 lanes available from the CPU.

Q: Is the multiplier unlocked?

A: No, the multiplier is not unlocked.

Q: When was the 3975WX released?

A: The release date is 2020-07-13.

Q: Does the 3975WX include integrated graphics?

A: No integrated graphics is listed in the data, so a discrete GPU is required.

Single-Thread vs Multi-Thread Behavior

The 3975WX behaves like two different processors depending on the workload. In single-thread tests, the scores are Cinebench R15 752, R20 3,136, R23 7,467, and Geekbench 1,648. These are usable scores, but they do not match the scale of the multi-thread results.

In multi-thread tests, the same processor reaches Cinebench R15 5,331, R20 22,215, R23 52,895, and Geekbench 14,828. The gap between the two sets of results is the defining feature of this chip. Software that can use many cores and threads will extract nearly all of its capability, while single-threaded software will leave most of the silicon idle.

The aggregate benchmark average of 13,534 sits between these extremes, which is why the 72nd percentile ranking does not fully reflect the multi-core capability. The single-thread versus multi-thread split matters more than the overall average. For rendering, compiling, batch processing, and similar workloads that parallelize well, the 3975WX is a high-end multi-thread performer. For single-thread-bound tasks, the processor’s advantage is far less visible in the data.