AMD EPYC 4584PX vs AMD EPYC 9135 Comparison

The AMD EPYC 4584PX and AMD EPYC 9135 are two 16-core server processors that end up nearly tied in overall average benchmark score, with the 4584PX edging ahead by a mere 0.1% (83090 vs 82980). Despite this statistical dead heat, the data reveals two very different performance profiles: the 4584PX dominates single-threaded and encryption workloads, while the 9135 counters with superior floating-point math and physics simulations. The 4584PX wins 12 of the 17 head-to-head benchmark comparisons, but the 9135's wins come in categories that often matter more for scientific computing. Both processors sit in the 96th percentile of all CPUs, indicating top-tier performance regardless of which specific strengths a workload emphasizes.

Head-to-Head Benchmarks

The most striking pattern in the benchmark data is the consistency of the 4584PX's lead across Cinebench tests. In all six Cinebench R15, R20, and R23 benchmarks—both single-core and multi-core—the 4584PX wins by exactly 4.9%. This uniform margin suggests a fundamental clock-speed advantage rather than architecture-specific gains. The 4584PX scores 5193 vs 4952 in Cinebench R15 multi-core, 21640 vs 20637 in R20 multi-core, and 51524 vs 49136 in R23 multi-core. Single-core results follow the same pattern: 733 vs 699 in R15, 3055 vs 2913 in R20, and 7274 vs 6936 in R23.

The 4584PX's largest victory comes in PassMark's find prime numbers test, where it scores 441 versus the 9135's 292—a massive 51% advantage. This workload, which is heavily dependent on integer arithmetic and low-latency cache access, highlights the 4584PX's 3D V-Cache advantage. Data encryption also strongly favors the 4584PX, which scores 45902 versus 40295, a 13.9% lead. The 4584PX additionally wins PassMark multi-thread (58117 vs 57170, +1.7%), single-thread (3795 vs 3672, +3.3%), and data compression (741648 vs 739277, +0.3%).

The 9135 fights back in several key areas. Its most decisive win is PassMark physics, where it scores 5477 versus 4574—a 16.5% margin that reflects the newer Zen 5 architecture's improved floating-point scheduling. Floating-point math also favors the 9135: 126679 vs 121460, a 4.1% advantage. Extended instructions go to the 9135 as well (55822 vs 53774, +3.7%), as does random string sorting (90064 vs 87690, +2.6%). Integer math is essentially a tie at 202962 vs 201919, a negligible 0.5% edge for the 9135.

Architecture Differences

The two processors represent different generations and design philosophies. The 4584PX belongs to the EPYC 4004 series built on Zen 4 architecture with the codename Raphael, manufactured on TSMC's 5 nm process. The 9135 comes from the EPYC 9005 series using Zen 5 (codenamed Turin) on a 4 nm process. Both use TSMC as the foundry, with transistor counts of 17,840 million for the 4584PX and 16,630 million for the 9135. Die sizes are nearly identical at 2x 71 mm² for the 4584PX and 2x 70.6 mm² for the 9135.

Cache configurations differ dramatically. The 4584PX features 128 MB of shared L3 cache plus an additional 1x 64MB 3D V-Cache slice, alongside 64 KB L1 per core and 1 MB L2 per core. The 9135 offers only 64 MB of shared L3 cache with no V-Cache, though its L1 is larger at 80 KB per core with the same 1 MB L2 per core. This cache disparity directly explains the 4584PX's 51% lead in prime number finding and its encryption advantage.

Clock speeds also diverge significantly. The 4584PX runs at 4.20 GHz base and 5.70 GHz boost, while the 9135 operates at 3.65 GHz base and 4.30 GHz boost. This 1.4 GHz boost clock difference underpins the 4584PX's consistent 4.9% Cinebench victories. However, the 9135 compensates with a substantially higher 200W TDP versus 120W for the 4584PX, enabling sustained throughput in floating-point-heavy workloads.

Platform support marks another major divide. The 4584PX uses AMD Socket AM5 with dual-channel DDR5 memory and 83.2 GB/s bandwidth, while the 9135 uses Socket SP5 with twelve-channel DDR5 and 576.0 GB/s memory bandwidth. PCIe connectivity is likewise asymmetric: 28 lanes (Gen 5) for the 4584PX versus 128 lanes (Gen 5) for the 9135. The 4584PX includes integrated Radeon Graphics; the 9135 has no integrated graphics.

Where Each One Wins

The 4584PX is the clear choice for latency-sensitive and single-threaded workloads. Its 3D V-Cache delivers exceptional results in prime number computation (51% faster) and data encryption (13.9% faster), workloads that benefit from larger cache residency. The 4.9% Cinebench advantage across every test indicates superior responsiveness in both single-threaded and multi-threaded rendering tasks. For applications that prioritize low-latency data access, database lookups, or financial calculations, the 4584PX's cache architecture provides a measurable edge.

The 9135 excels in throughput-oriented scientific and simulation workloads. Its 16.5% lead in PassMark physics suggests better handling of rigid-body dynamics and collision detection, common in engineering simulation. The 4.1% floating-point advantage points to strengths in scientific computing, financial modeling, and any workload relying heavily on non-integer arithmetic. The 9135's twelve-channel memory subsystem, with 576.0 GB/s bandwidth versus 83.2 GB/s, makes it the superior platform for memory-bandwidth-bound applications like large-scale data analytics or in-memory databases.

The 9135 also wins in extended instruction workloads (3.7% faster) and random string sorting (2.6% faster), indicating better optimization for modern SIMD instruction sets and text-processing tasks. Integer math is effectively a tie, meaning neither processor holds a meaningful advantage in general integer-heavy enterprise applications.

FAQ

Q: Which processor has the higher average benchmark score?

A: The AMD EPYC 4584PX has a marginally higher average benchmark score of 83090 versus 82980 for the 9135, a difference of only 0.1% that falls well within normal run-to-run variance.

Q: Why does the 4584PX perform so much better in prime number finding?

A: The 4584PX includes a 1x 64MB 3D V-Cache slice in addition to its 128 MB shared L3 cache, compared to the 9135's 64 MB shared L3 with no V-Cache. This additional cache allows more of the working set to remain on-die, resulting in a 51% performance advantage in PassMark's prime number test.

Q: What memory bandwidth does each processor support?

A: The 4584PX supports dual-channel DDR5 memory with 83.2 GB/s bandwidth, while the 9135 supports twelve-channel DDR5 with 576.0 GB/s bandwidth—a nearly 7x difference in theoretical peak memory throughput.

Q: Which processor has higher clock speeds?

A: The 4584PX runs at 4.20 GHz base and 5.70 GHz boost, while the 9135 runs at 3.65 GHz base and 4.30 GHz boost. The 4584PX's 1.4 GHz higher boost clock contributes to its consistent 4.9% lead across all Cinebench benchmarks.

Q: Are both processors based on the same architecture?

A: No. The 4584PX uses Zen 4 architecture (codenamed Raphael) on a 5 nm process, while the 9135 uses Zen 5 architecture (codenamed Turin) on a 4 nm process. Despite the architectural generational gap, the 4584PX wins more benchmarks due to its clock and cache advantages.

Q: How many PCIe lanes does each processor provide?

A: The 4584PX provides 28 PCIe Gen 5 lanes (CPU only), while the 9135 provides 128 PCIe Gen 5 lanes (CPU only). The 9135's 4.5x lane count makes it far more suitable for systems requiring extensive high-speed I/O.

The Verdict

The benchmark data makes a clear recommendation: choose the AMD EPYC 4584PX if your workloads are latency-sensitive, cache-dependent, or single-threaded. Its 3D V-Cache delivers a 51% win in prime number finding and 13.9% in encryption, while its higher boost clock produces a 4.9% advantage in every Cinebench test. The 4584PX also achieves these wins at 120W TDP versus the 9135's 200W, making it the more efficient choice for standard server workloads.

Choose the AMD EPYC 9135 for scientific computing, simulation, or memory-bandwidth-intensive applications. Its 16.5% physics win and 4.1% floating-point advantage stem from the newer Zen 5 architecture, while its twelve-channel memory interface with 576.0 GB/s bandwidth and 128 PCIe lanes provide far greater platform headroom for large-scale deployments. The 9135 also handles extended instruction workloads 3.7% better and random string sorting 2.6% better.

The overall average benchmark scores are nearly identical—83090 vs 82980—so the decision hinges entirely on workload characteristics. If your applications fit in the 4584PX's larger cache and benefit from higher clock speeds, it is the better performer. If your applications demand memory bandwidth, floating-point throughput, or massive I/O connectivity, the 9135 is the correct choice despite losing more individual benchmark comparisons.

Specification Differences

| Specification | AMD EPYC 4584PX | AMD EPYC 9135 |

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

| Series | EPYC 4004 series | EPYC 9005 series |

| Base Clock | 4.20 GHz | 3.65 GHz |

| Boost Clock | 5.70 GHz | 4.30 GHz |

| TDP | 120 W | 200 W |

| Socket | AMD Socket AM5 | AMD Socket SP5 |

| Architecture | Zen 4 | Zen 5 |

| Codename | Raphael | Turin |

| Generation | EPYC (Zen 4 (Raphael)) | EPYC (Zen 5 (Turin)) |

| Process Node | 5 nm | 4 nm |

| Transistors | 17,840 million | 16,630 million |

| Die Size | 2x 71 mm² | 2x 70.6 mm² |

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

| L3 Cache | 128 MB (shared) | 64 MB (shared) |

| V-Cache | 1x 64MB Slice | None |

| Memory Bus | Dual-channel | Twelve-channel |

| Memory Bandwidth | 83.2 GB/s | 576.0 GB/s |

| PCIe | Gen 5, 28 Lanes (CPU only) | Gen 5, 128 Lanes (CPU only) |

| Integrated Graphics | Radeon Graphics | N/A |

| Release Date | 2024-05-20 | 2024-10-09 |

| Launch MSRP | $699 | $1214 |

| Part Number | 100-000001481 | 100-000001150 |