AMD EPYC 7643P vs AMD EPYC 9335 Comparison

FAQ

Q: Which processor has more cores and threads?

A: The AMD EPYC 7643P has 48 cores and 96 threads, while the AMD EPYC 9335 has 32 cores and 64 threads. The 7643P offers 50% more cores and threads than the 9335.

Q: How do the two compare in single-threaded performance?

A: The AMD EPYC 9335 wins in single-threaded workloads, scoring 2732 versus 2655 for the 7643P, a 2.9% advantage. The 9335 also has a higher boost clock of 4.40 GHz compared to 3.60 GHz on the 7643P.

Q: Which processor supports faster memory?

A: The AMD EPYC 9335 supports DDR5 memory with a twelve-channel memory bus and 576.0 GB/s bandwidth. The AMD EPYC 7643P uses DDR4 with an eight-channel bus and 204.8 GB/s bandwidth, making the 9335 nearly three times faster in theoretical memory bandwidth.

Q: What is the manufacturing process difference?

A: The 9335 is built on a 4 nm process at TSMC, while the 7643P uses a 7 nm process, also at TSMC. The 9335's dies are 4x 70.6 mm², whereas the 7643P has 8x 81 mm² dies.

Q: Which processor has more L3 cache?

A: The AMD EPYC 7643P has 256 MB of shared L3 cache, double the 128 MB shared L3 cache on the AMD EPYC 9335. However, the 9335 has larger per-core L1 and L2 caches: 80 KB and 1 MB per core respectively, versus 64 KB and 512 KB per core on the 7643P.

Q: How do their average benchmark scores compare?

A: The AMD EPYC 9335 has an average benchmark score of 194228, placing it in the 99th percentile of all CPUs. The AMD EPYC 7643P has an average score of 144824, in the 98th percentile.

Architecture Differences

The two processors represent distinct generations of AMD's EPYC lineup. The AMD EPYC 9335 belongs to the EPYC 9005 series, codenamed Turin, and uses the Zen 5 architecture. The AMD EPYC 7643P is part of the EPYC 7003 series, codenamed Milan, built on the older Zen 3 architecture. This generational gap drives most of the behavioral differences in the benchmark data.

The manufacturing process separates them clearly. The 9335 uses TSMC's 4 nm node, while the 7643P is fabricated on TSMC's 7 nm node. The 9335 packs 33,260 million transistors across 4x 70.6 mm² dies. The 7643P contains 33,200 million transistors spread over 8x 81 mm² dies. Despite similar transistor counts, the newer node allows the 9335 to achieve higher clock speeds at a slightly lower TDP: 210 W versus 225 W.

Cache hierarchies differ substantially. The 9335 provides 80 KB of L1 and 1 MB of L2 per core, with 128 MB of shared L3. The 7643P offers 64 KB of L1 and 512 KB of L2 per core, but compensates with 256 MB of shared L3. The doubled L3 capacity on the 7643P likely contributes to its strong results in data-heavy workloads.

Memory architecture is another major divider. The 9335 supports DDR5 over a twelve-channel bus, yielding 576.0 GB/s of memory bandwidth. The 7643P uses DDR4 across an eight-channel bus, capping out at 204.8 GB/s. The 9335's memory bandwidth is roughly 2.8 times higher, which matters for memory-bound server workloads.

PCIe capabilities also differ by generation. The 9335 supports PCIe Gen 5 with 128 lanes (CPU only), while the 7643P supports PCIe Gen 4 with the same 128 lanes. Sockets differ as well: the 9335 uses AMD Socket SP5, and the 7643P uses AMD Socket SP3. Both processors support ECC memory and target the server/workstation segment. The 9335 launched on 2024-10-09, about a year after the 7643P's 2023-09-04 release.

The Verdict

The data paints a clear split between generations and core counts. The AMD EPYC 7643P wins 7 of the 11 head-to-head benchmark comparisons. Its core advantage (48 versus 32) shows up most strongly in massively parallel workloads. The 7643P leads by 47.8% in find prime numbers, 33.9% in data encryption, and 76.2% in the physics test. For workloads that scale with core count and can use large shared L3 cache, the 7643P is the stronger option.

The AMD EPYC 9335 wins 4 of the 11 comparisons, but its wins are concentrated in areas that reflect architectural efficiency. It beats the 7643P by 56.8% in extended instructions, a massive margin that suggests the Zen 5 core handles advanced instruction sets far better. It also leads in floating point math (5.3%), single-thread (2.9%), and the single-thread duplicate test. The 9335's higher boost clock (4.40 GHz versus 3.60 GHz) and newer process node explain these results.

Buyers should choose the 7643P when thread count and cache capacity dominate the workload profile. The data shows it is particularly strong in encryption, prime number calculation, physics simulation, and integer math. The 9335 becomes the better pick when per-core performance, memory bandwidth, and modern instruction support matter more. Its 576.0 GB/s memory bandwidth versus 204.8 GB/s on the 7643P is a significant advantage for memory-intensive tasks, even if the benchmark suite does not isolate that metric directly.

The 9335 also sits in a higher performance percentile overall (99th versus 98th) and has a higher average benchmark score (194228 versus 144824). However, that average reflects the 9335's strong single-thread and extended instruction results, not overall throughput dominance.

Specification Differences

| Specification | AMD EPYC 9335 | AMD EPYC 7643P |

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

| Series | EPYC 9005 series | EPYC 7003 series |

| Cores | 32 | 48 |

| Threads | 64 | 96 |

| Base Clock | 3.00 GHz | 2.30 GHz |

| Boost Clock | 4.40 GHz | 3.60 GHz |

| TDP | 210 W | 225 W |

| Socket | AMD Socket SP5 | AMD Socket SP3 |

| Architecture | Zen 5 | Zen 3 |

| Codename | Turin | Milan |

| Process Node | 4 nm | 7 nm |

| Transistors | 33,260 million | 33,200 million |

| Die Size | 4x 70.6 mm² | 8x 81 mm² |

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

| L2 Cache | 1 MB (per core) | 512 KB (per core) |

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

| Memory Support | DDR5 | DDR4 |

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

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

| PCIe | Gen 5, 128 Lanes (CPU only) | Gen 4, 128 Lanes (CPU only) |

| Release Date | 2024-10-09 | 2023-09-04 |

| Launch MSRP | $3178 | $2722 |

Both processors share the same manufacturer, support ECC memory, have 128 PCIe lanes, lack integrated graphics, target the server/workstation segment, remain in active production, and have locked multipliers.

Head-to-Head Benchmarks

The largest single victory belongs to the 9335 in extended instructions, where it scores 105706 against 67398, a 56.8% advantage. This result likely reflects the Zen 5 architecture's improved handling of AVX-512 or similar instruction sets. No other comparison comes close to this margin.

The 7643P counters with several decisive wins. The physics test shows the biggest gap: 8002 versus 1905, a 76.2% lead for the 7643P. This result is striking because physics workloads often scale with core count, and the 7643P has exactly 50% more cores. The margin exceeds the core advantage, suggesting the 7643P's 256 MB L3 cache helps here.

In find prime numbers, the 7643P scores 651 versus 340, a 47.8% lead. Data encryption shows 95606 versus 63159, a 33.9% advantage. Random string sorting favors the 7643P by 27.2% (160274 versus 116608). Integer math goes to the 7643P at 390775 versus 346291, an 11.4% margin. The multithread test shows 77307 versus 65811, a 14.9% lead for the 7643P. Data compression rounds out the 7643P's wins at 1326051 versus 1203096, a 9.3% gap.

The 9335 wins the remaining tests. Floating point math scores 228123 versus 216592, a 5.3% margin. Single-thread performance is close: 2732 versus 2655, a 2.9% lead. The duplicate single-thread test shows identical numbers.

The overall pattern is consistent. The 7643P leverages its 48 cores, 96 threads, and double L3 cache to win throughput-oriented tasks. The 9335 uses its Zen 5 cores, higher clocks, and newer process node to win efficiency-oriented and instruction-heavy tasks.

Where Each One Wins

The AMD EPYC 7643P is the clear winner for parallel compute workloads. The data shows it dominating in physics simulation (76.2% ahead), prime number calculation (47.8% ahead), and encryption (33.9% ahead). These are workloads that expand with core count and benefit from large cache. The 7643P also leads in integer math, multithread scores, random string sorting, and data compression. Server workloads involving database operations, scientific computing, and batch processing would favor the 7643P based on these results.

The AMD EPYC 9335 wins where instruction efficiency and per-core performance matter. Its 56.8% lead in extended instructions makes it the better choice for workloads that use modern vector or cryptographic instruction sets. The floating point math win (5.3%) suggests advantages in scientific or financial modeling that relies on floating point throughput. The single-thread win (2.9%) points to better responsiveness in lightly threaded tasks or applications with serial bottlenecks.

The memory bandwidth difference deserves attention. The 9335's 576.0 GB/s bandwidth versus 204.8 GB/s on the 7643P means the 9335 is overwhelmingly better suited for workloads that stream large datasets. The benchmark suite does not isolate memory bandwidth directly, but the specification gap is enormous. The 9335 also supports DDR5, which brings lower latency and higher density options compared to DDR4.

The 7643P is the choice for maximizing core count and cache capacity within a single socket. Its 48 cores and 256 MB L3 cache make it a strong fit for virtualized environments, high-density container hosts, and parallel batch jobs. The 9335 is the choice for workloads where each core must do more work individually, where memory bandwidth is the limiting factor, or where the latest instruction set support is required. The 9335's higher boost clock (4.40 GHz versus 3.60 GHz) reinforces this position.

Both processors hold active production status and serve the server/workstation segment. The 9335 sits in the 99th percentile of all CPUs, while the 7643P sits in the 98th. The average benchmark scores differ by roughly 34%, with the 9335 at 194228 and the 7643P at 144824. That said, the head-to-head results show the 7643P winning most individual tests, so the average score does not tell the full story. The 9335's extreme score in extended instructions inflates its average, while the 7643P spreads its strengths across more conventional workloads.