AMD EPYC 9655P vs AMD Ryzen Threadripper 9970X Comparison

# AMD EPYC 9655P vs AMD Ryzen Threadripper 9970X

The AMD EPYC 9655P and AMD Ryzen Threadripper 9970X are both Zen 5 designs on 4 nm silicon, but they target fundamentally different workloads. The EPYC 9655P is a 96-core server processor with a 400 W TDP, while the Threadripper 9970X is a 32-core desktop processor with a 350 W TDP. Benchmark data shows the EPYC winning 9 of 11 head-to-head tests, with the Threadripper claiming only the single-threaded tests. The average benchmark score for the EPYC is 397,773, placing it in the 100th percentile of all CPUs, while the Threadripper averages 279,778, sitting in the 99th percentile. These are both elite processors, but the data reveals clear specialization: the EPYC dominates parallel throughput, while the Threadripper excels at latency-sensitive single-thread work.

Where Each One Wins

The EPYC 9655P is the clear winner in every multi-threaded category. In the head-to-head results, it leads by 49.4% in PassMark multithread, 98.3% in data compression, 131.1% in floating point math, 136% in random string sorting, 153.6% in data encryption, 163.3% in integer math, 174.1% in prime number finding, and 278.2% in physics. The physics result is the largest gap, reflecting the massive core-count advantage: 96 cores versus 32 cores. The EPYC also wins in extended instructions by 62%, a test that measures AVX and SIMD throughput, where the sheer number of cores overwhelms the Threadripper.

The Threadripper 9970X wins the single-thread tests, both PassMark singlethread and PassMark single_thread, with a score of 4,530 versus the EPYC's 3,849. That is a 15% advantage in raw per-core performance. This comes from a higher boost clock of 5.40 GHz versus 4.50 GHz, and a higher base clock of 4.00 GHz versus 2.60 GHz. The Threadripper also has a higher PassMark single-thread percentile within its own benchmark suite, at 4,530, which is notable because it is not just a clock speed win but also a cache and memory latency win.

In terms of workload suitability, the EPYC is for server virtualization, database processing, scientific computing, and any workload that scales linearly with cores. The Threadripper is for enthusiast desktop use, content creation, and applications that require fast single-thread responsiveness alongside moderate multi-thread capability. The data shows the EPYC is 42.2% ahead in average benchmark score (397,773 versus 279,778), but the Threadripper's single-thread advantage means it will feel snappier in interactive applications.

Architecture Differences

Both processors use the Zen 5 architecture on TSMC's 4 nm process node. The EPYC 9655P, codenamed Turin, uses 12 CCDs, each with a die size of 70.6 mm², for a total transistor count of 99,780 million. The Threadripper 9970X, codenamed Shimada Peak, uses 4 CCDs of the same 70.6 mm² die size, totaling 33,260 million transistors. The EPYC has 3x the transistors, which directly translates to its 3x core count.

Cache configurations differ substantially. The EPYC has 80 KB of L1 cache per core and 1 MB of L2 per core, with a massive 384 MB of shared L3 cache. The Threadripper has 64 KB of L1 per core and 1 MB of L2 per core, with 128 MB of L3. The EPYC's L3 is 3x larger, which helps in large working sets typical of server workloads. However, the Threadripper's per-core L1 is smaller, meaning each core has less dedicated cache, but the lower core count means less contention.

Memory support is another major differentiator. The EPYC uses twelve-channel DDR5 memory with a bandwidth of 576.0 GB/s. The Threadripper uses quad-channel DDR5 with 204.8 GB/s. The EPYC has 2.8x the memory bandwidth, which is critical for the data compression, encryption, and floating point tests where it wins by huge margins. The Threadripper's memory bandwidth is still high, but it is constrained by the quad-channel bus.

PCIe connectivity also differs. The EPYC provides 128 PCIe Gen 5 lanes (CPU only), while the Threadripper provides 80 lanes. Both support ECC memory, and neither has integrated graphics. The EPYC's socket is AMD Socket SP5, while the Threadripper uses AMD Socket sTR5. The EPYC is a server/workstation part with a launch MSRP of $10811, while the Threadripper is a desktop part with a launch MSRP of $2499. The EPYC has a locked multiplier, while the Threadripper is unlocked, allowing overclocking.

Head-to-Head Benchmarks

The largest win for the EPYC is in PassMark physics, where it scores 25,847 versus the Threadripper's 6,835, a 278.2% advantage. This test simulates rigid body physics and scales almost perfectly with core count, so the 96-core EPYC overwhelms the 32-core Threadripper. In integer math, the EPYC scores 1,225,251 versus 465,378, a 163.3% lead. This is a core-count-bound workload, and the EPYC's 192 threads versus 64 threads show a clear advantage.

In prime number finding, the EPYC scores 1,686 versus 615, a 174.1% lead. This test is sensitive to both core count and per-core performance, and the EPYC's combination of high core count and high boost clock of 4.50 GHz wins. Data encryption shows a 153.6% lead (220,074 versus 86,765), which benefits from the EPYC's larger L3 cache and higher memory bandwidth. Floating point math is 131.1% ahead (715,866 versus 309,719), again reflecting the core count and cache advantages.

The EPYC also wins in random string sorting by 136% (451,824 versus 191,445), data compression by 98.3% (3,486,158 versus 1,757,998), and extended instructions by 62% (230,609 versus 142,342). The multithread test shows a 49.4% lead (160,490 versus 107,399), which is the smallest multi-threaded gap. This is because the PassMark multithread test does not scale perfectly with core count, but the EPYC still wins decisively.

The Threadripper wins only the single-thread tests, with a score of 4,530 versus 3,849, a 15% lead. This is the only category where the Threadripper's higher clocks (5.40 GHz boost versus 4.50 GHz) and lower core count (which reduces thermal and power contention) provide an edge. In the context of the EPYC's nearest rivals, the EPYC is 3.5% behind the AMD Ryzen Threadripper PRO 9995WX, 4.8% ahead of the AMD EPYC 9535, 6.5% ahead of the AMD EPYC 9655, and 6.6% behind the AMD EPYC 9745. The Threadripper 9970X is 0.2% behind the Intel Xeon 6780E, 2% behind the AMD EPYC 9565, 2.2% behind the Intel Xeon 696X, and 2.5% behind the AMD EPYC 9555P.

FAQ

Q: Which processor has more cores?

A: The AMD EPYC 9655P has 96 cores and 192 threads, while the AMD Ryzen Threadripper 9970X has 32 cores and 64 threads.

Q: Which processor has higher single-thread performance?

A: The AMD Ryzen Threadripper 9970X scores 4,530 in PassMark single-thread, which is 15% higher than the EPYC 9655P's 3,849. This is due to its higher boost clock of 5.40 GHz versus 4.50 GHz.

Q: How much memory bandwidth does each processor support?

A: The EPYC 9655P supports twelve-channel DDR5 with 576.0 GB/s, while the Threadripper 9970X supports quad-channel DDR5 with 204.8 GB/s.

Q: What is the largest benchmark gap between the two?

A: The largest gap is in PassMark physics, where the EPYC scores 25,847 versus the Threadripper's 6,835, a 278.2% advantage.

Q: Do both processors support ECC memory?

A: Yes, both the EPYC 9655P and the Threadripper 9970X support ECC memory.

Q: Which processor has a larger L3 cache?

A: The EPYC 9655P has 384 MB of shared L3 cache, while the Threadripper 9970X has 128 MB of L3.

The Verdict

The data is unambiguous: the AMD EPYC 9655P is the superior processor for multi-threaded workloads. It wins 9 of 11 head-to-head tests, with margins ranging from 49.4% (multithread) to 278.2% (physics). Its 96 cores, 192 threads, 384 MB of L3, and 576.0 GB/s of memory bandwidth make it the dominant choice for server virtualization, database workloads, scientific simulation, and any task that can utilize more than 64 threads. The 100th percentile ranking and average benchmark score of 397,773 confirm its position at the top of the CPU hierarchy.

The AMD Ryzen Threadripper 9970X is the better choice for single-threaded performance. Its 15% lead in PassMark single-thread (4,530 versus 3,849) comes from a 5.40 GHz boost clock and a 4.00 GHz base clock, both significantly higher than the EPYC's 2.60 GHz base and 4.50 GHz boost. For desktop users who run a mix of interactive applications, content creation tools, and some multi-threaded rendering, the Threadripper provides a more balanced experience. Its 32 cores and 64 threads are still substantial, and the unlocked multiplier allows overclocking, which is not possible on the EPYC.

The EPYC's nearest rival data shows it is 3.5% behind the Threadripper PRO 9995WX, but also 6.5% ahead of the EPYC 9655. The Threadripper's nearest rival data shows it is 0.2% behind the Intel Xeon 6780E, and 2.5% behind the AMD EPYC 9555P. These deltas are small, indicating that both processors are competitive within their respective segments.

For a server or workstation deployment where core count is king, the EPYC 9655P is the clear winner. For a desktop workstation where per-core responsiveness matters, the Threadripper 9970X is the right pick. The data does not support a single "best" processor; it supports two different tools for two different jobs.

Specification Differences

| Specification | AMD EPYC 9655P | AMD Ryzen Threadripper 9970X |

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

| Cores | 96 | 32 |

| Threads | 192 | 64 |

| Base Clock | 2.60 GHz | 4.00 GHz |

| Boost Clock | 4.50 GHz | 5.40 GHz |

| TDP | 400 W | 350 W |

| Socket | AMD Socket SP5 | AMD Socket sTR5 |

| Codename | Turin | Shimada Peak |

| Transistors | 99,780 million | 33,260 million |

| Die Size | 12x 70.6 mm² | 4x 70.6 mm² |

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

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

| L3 Cache | 384 MB (shared) | 128 MB |

| Memory Bus | Twelve-channel | Quad-channel |

| Memory Bandwidth | 576.0 GB/s | 204.8 GB/s |

| PCIe | Gen 5, 128 Lanes (CPU only) | Gen 5, 80 Lanes (CPU only) |

| Market Segment | Server/Workstation | Desktop |

| Release Date | 2024-10-09 | 2025-07-29 |

| Launch MSRP | $10811 | $2499 |

| Multiplier Unlocked | No | Yes |

| Part Number | 100-000001522 | 100-000001594 |