AMD EPYC 9845 vs AMD Ryzen Threadripper PRO 9985WX Comparison

Where Each One Wins

The benchmark split between the AMD EPYC 9845 and the AMD Ryzen Threadripper PRO 9985WX is nearly even in raw win count, 9 versus 8, but the nature of those wins tells a clear story about workload orientation. The EPYC 9845 dominates in throughput-heavy, parallel compute tasks, while the Threadripper PRO 9985WX takes every single-threaded and render-based multi-core test in the Cinebench suite.

Starting with the EPYC 9845, its wins are concentrated in PassMark's compute and data-processing workloads. The most dramatic margin is in integer math, where the EPYC scores 1,687,531 against 872,710, a 93.4% advantage. Floating-point math follows closely at 978,377 versus 553,348, a 76.8% lead. Data encryption shows a 91.7% delta (296,808 versus 154,824), and data compression comes in at 60.7% ahead (4,680,013 versus 2,912,972). These are not marginal differences; they are near-doublings of throughput, which points to the EPYC's 160-core configuration being the decisive factor in highly parallel algorithmic work.

The Threadripper PRO 9985WX, by contrast, wins every Cinebench test, both single-core and multi-core, but by a uniform and narrow margin of 2.1%. In Cinebench R23 multi-core, it scores 132,859 versus 130,037; in R23 single-core, 18,756 versus 18,358. The pattern repeats across R15 and R20 with the same 2.1% delta. This suggests that the Threadripper's advantage in these render workloads is not about core count, since it has only 64 cores versus 160, but rather about per-core efficiency and clock behavior. The single-thread PassMark test confirms this: the Threadripper scores 4,482 versus 3,144, a 29.9% lead, which is the largest single-test margin in its favor.

There is also a middle ground. The EPYC wins PassMark multi-thread by only 1.9% (152,985 versus 150,071), a narrow margin that suggests the two processors are nearly equivalent in that specific aggregate workload. The EPYC also wins extended instructions by 39.7%, physics by 42.4%, and random string sorting by 63.5%. The Threadripper's only other win besides Cinebench and single-thread is the find-prime-numbers test, where it actually loses by 10.3%, so its win column is entirely composed of Cinebench and PassMark single-thread tests.

Architecture Differences

The two processors share the Zen 5 architecture and TSMC as the foundry, but they diverge sharply in implementation. The EPYC 9845, codenamed Turin, uses a 3 nm process node, while the Threadripper PRO 9985WX, codenamed Shimada Peak, uses 4 nm. The EPYC packs 160 cores and 320 threads, a figure that requires the denser process to fit on the server package. The Threadripper offers 64 cores and 128 threads, which is still a large count for a workstation part but less than half the EPYC's core count.

Clock speeds tell the opposite story. The EPYC has a base clock of 2.10 GHz and a boost clock of 3.70 GHz, while the Threadripper runs at a 3.20 GHz base and a 5.40 GHz boost. That 1.70 GHz difference in boost clock is the primary reason the Threadripper wins every single-threaded test by a large margin. The EPYC compensates with sheer core count, but its lower clocks cap its per-thread performance.

Cache allocations also differ. The EPYC provides 80 KB of L1 per core and 1 MB of L2 per core, with 320 MB of shared L3. The Threadripper has 64 KB of L1 per core, 1 MB of L2 per core, and 256 MB of L3. The EPYC's larger L3, 320 MB versus 256 MB, aligns with its server role where large working sets are common. The Threadripper's smaller per-core L1 is a minor difference, but the L3 gap is substantial and likely aids the EPYC in data-heavy workloads.

Memory subsystems diverge significantly. The EPYC uses twelve-channel DDR5 with a memory bandwidth of 576.0 GB/s, while the Threadripper uses eight-channel DDR5 at 409.6 GB/s. The EPYC's 166.4 GB/s bandwidth advantage is a direct contributor to its wins in data compression and encryption, which are memory-intensive. Both support ECC memory, and both provide PCIe Gen 5 with 128 lanes from the CPU, so expansion capability is identical.

The Threadripper has a notable feature the EPYC lacks: an unlocked multiplier. This means the Threadripper can be overclocked, though the benchmark data reflects stock settings. The EPYC is multiplier-locked, which is typical for server parts. The Threadripper also carries a significantly higher boost clock for single-thread work, and its TDP is lower at 350 watts versus 390 watts, despite the clock advantage. The EPYC's higher TDP reflects the cost of powering 160 cores.

The Threadripper's transistor count is listed at 66,520 million across 8 dies of 70.6 mm² each, while the EPYC's transistor and die size data are not recorded in the database. The EPYC uses the SP5 socket, the Threadripper uses sTR5, so they are not platform-compatible. Both are active production parts with no integrated graphics.

The Verdict

The choice between these two processors is determined by workload scale and thread-level efficiency. The data shows the EPYC 9845 is the correct pick for tasks that scale across many cores and benefit from massive memory bandwidth. Its 93.4% lead in integer math, 91.7% lead in encryption, and 76.8% lead in floating-point math are decisive. For server virtualization, database processing, scientific computing, or any workload that can use 160 cores and 320 threads, the EPYC's throughput advantage is overwhelming. Its 576.0 GB/s memory bandwidth and 320 MB L3 cache are purpose-built for that role.

The Threadripper PRO 9985WX is the pick for workloads that are bound by single-thread performance or that use render engines where per-core clock speed matters more than raw core count. Its 29.9% lead in PassMark single-thread and its consistent 2.1% lead across all Cinebench tests, including multi-core, make it the better choice for interactive workstation use, 3D rendering, and software development where compile times depend on both clock speed and core count. The fact that it wins Cinebench R23 multi-core despite having 96 fewer cores is remarkable and points to a highly efficient implementation.

For users who need both high single-thread and high multi-thread performance, the Threadripper's Cinebench multi-core wins are relevant, but its PassMark multi-thread score is only 1.9% behind the EPYC, which indicates near-parity in that aggregate metric. The EPYC's wins in PassMark physics, extended instructions, and random string sorting further cement its position for computational workloads. The Threadripper's lower TDP of 350 watts versus 390 watts, and its unlocked multiplier, make it more flexible for enthusiast workstation builds. The EPYC's 100th percentile ranking versus the Threadripper's 99th percentile is a narrow overall distinction, but the average benchmark scores differ substantially: 523,613 for the EPYC versus 320,749 for the Threadripper, a gap driven by the EPYC's massive wins in high-weight PassMark tests.

FAQ

Q: Which processor has more cores?

A: The AMD EPYC 9845 has 160 cores and 320 threads, while the AMD Ryzen Threadripper PRO 9985WX has 64 cores and 128 threads.

Q: Why does the Threadripper win single-threaded benchmarks?

A: The Threadripper PRO 9985WX has a boost clock of 5.40 GHz versus 3.70 GHz for the EPYC 9845. In the PassMark single-thread test, the Threadripper scores 4,482 versus 3,144, a 29.9% lead.

Q: Which processor has higher memory bandwidth?

A: The EPYC 9845 uses twelve-channel DDR5 with 576.0 GB/s bandwidth, while the Threadripper PRO 9985WX uses eight-channel DDR5 with 409.6 GB/s.

Q: Are these processors overclockable?

A: The Threadripper PRO 9985WX has an unlocked multiplier, while the EPYC 9845 is multiplier-locked.

Q: How do they compare in Cinebench R23 multi-core?

A: The Threadripper PRO 9985WX scores 132,859 versus 130,037 for the EPYC 9845, a 2.1% lead for the Threadripper despite having 96 fewer cores.

Q: Which processor wins more benchmark tests overall?

A: The EPYC 9845 wins 9 of the 17 head-to-head tests, while the Threadripper PRO 9985WX wins 8. The EPYC's wins are larger in magnitude, with deltas up to 93.4%, while the Threadripper's Cinebench wins are all 2.1%.

Head-to-Head Benchmarks

The largest win for the EPYC 9845 comes in PassMark integer math, where it scores 1,687,531 against the Threadripper's 872,710, a 93.4% advantage. This is the kind of workload that scales almost linearly with core count, and the EPYC's 160 cores versus 64 cores explains the near-doubling. Data encryption is nearly as lopsided: 296,808 versus 154,824, a 91.7% lead. Encryption algorithms are often parallelizable and memory-bound, which favors the EPYC's twelve-channel memory subsystem.

Floating-point math shows a 76.8% lead for the EPYC (978,377 versus 553,348), and random string sorting shows a 63.5% lead (538,060 versus 329,014). Data compression comes in at 60.7% (4,680,013 versus 2,912,972). These are all substantial margins that demonstrate the EPYC's dominance in data-processing and scientific workloads. The physics test shows a 42.4% lead (19,631 versus 13,783), and extended instructions show a 39.7% lead (314,798 versus 225,340). Even the find-prime-numbers test, which is often single-thread-bound, goes to the EPYC by 10.3% (1,255 versus 1,138), a surprising result given the Threadripper's clock advantage.

The PassMark multi-thread test is the closest race: the EPYC wins 152,985 versus 150,071, a 1.9% margin. This suggests that the aggregate multi-thread score balances the EPYC's core count against the Threadripper's clock speed, resulting in near-parity.

The Threadripper PRO 9985WX's wins are concentrated in the Cinebench suite, where it takes R15, R20, and R23 in both single-core and multi-core variants, each by exactly 2.1%. The R23 multi-core scores are 132,859 versus 130,037, and the R23 single-core scores are 18,756 versus 18,358. The R20 multi-core is 55,800 versus 54,615, and R15 multi-core is 13,392 versus 13,107. The single-core R15 is 1,890 versus 1,850, and R20 single-core is 7,877 versus 7,710.

The Threadripper's largest win is in PassMark single-thread, where it scores 4,482 versus 3,144, a 29.9% lead. This is the only test where the Threadripper exceeds a 2.1% margin, and it aligns perfectly with the boost clock difference. The data shows a clear division: the EPYC wins by large margins in parallel compute, the Threadripper wins by consistent small margins in render workloads and by a large margin in single-thread performance.