AMD EPYC 9135

The AMD EPYC 9135 is a 16-core, 32-thread server processor built on the Zen 5 architecture, codenamed Turin, and manufactured on TSMC’s 4 nm process. It occupies the 97th percentile among all CPUs in the benchmark database, with an average benchmark score of 83020. This places it in a competitive tier where it edges out some established workstation and server parts while trailing slightly behind newer high-end mobile and data center offerings.

How It Compares

Against the AMD Ryzen 9 8940HX, the EPYC 9135 holds a marginal 0.8% advantage in average benchmark score (83020 vs 82332). This is a narrow lead, indicating that despite the EPYC’s server-oriented design, its raw computational throughput is nearly identical to a high-end mobile processor. The data suggests the two are effectively interchangeable in multi-threaded workloads, though the EPYC brings substantially different platform features that the benchmark score alone does not capture.

The AMD EPYC 7443P is a previous-generation server part, and the 9135 outperforms it by 1.7% (83020 vs 81661). This is a modest generational gain, but it is significant because the 7443P likely has a different core count and memory configuration. The data shows the Zen 5 architecture in the 9135 delivers better per-core efficiency, allowing a 16-core part to surpass a rival that may rely on more cores or higher clocks to achieve its score.

Relative to the AMD Ryzen 9 9955HX, the EPYC 9135 trails by 2.3% (83020 vs 84952). This is the largest deficit among the listed rivals, but it remains a close race. The 9955HX is a mobile part designed for high sustained performance, and its slight lead suggests that for purely compute-bound tasks, a high-end laptop CPU can rival a server chip. However, the EPYC’s advantage lies in memory bandwidth, PCIe lanes, and scalability, which are not reflected in this aggregate score.

The AMD EPYC 7F72 is another server processor, and the 9135 is 2.4% behind it (83020 vs 85072). This is the smallest margin of difference in absolute terms, but it is the only rival where the 9135 loses by more than two percentage points. The 7F72 appears to be a higher-clocked or higher-core-count part, and the data indicates that the 9135’s 16-core configuration, while efficient, cannot fully close the gap against a more aggressively configured server chip.

Power and Thermals

The EPYC 9135 carries a TDP of 200 watts. This places it in a power class that requires robust cooling, but it is not extreme for a server processor. The data implies that a capable air cooler or a moderate liquid cooling solution would suffice for most chassis, though dense rack deployments would still need to account for the thermal output in airflow planning. The 200 W TDP is a single data point, but it suggests a balance between performance and power draw that is typical for mainstream EPYC parts.

Given the 200 W envelope, the thermal design assumes a server environment with high-static-pressure fans and adequate ventilation. The Zen 5 architecture’s 4 nm process helps keep power density manageable, but the dual-die design (2x 70.6 mm²) means heat is generated across two separate silicon pieces. The data does not include specific thermal throttle points, but the TDP class indicates that the processor will sustain its base and boost clocks under standard server cooling, provided the chassis can move enough air.

Single-Thread vs Multi-Thread Behavior

The EPYC 9135 shows a distinct split between single-thread and multi-thread performance. In Cinebench R23, the single-core score is 6936, while the multi-core score is 49136, yielding a ratio of about 7.1x. This indicates that the processor scales well across its 16 cores, with minimal overhead from thread synchronization. The single-core score itself is high, suggesting that even lightly threaded workloads will benefit from the Zen 5 architecture’s instruction-level parallelism.

In Cinebench R20, the single-core score is 2913 and multi-core is 20637, a ratio of 7.1x, consistent with R23. The older R15 test shows a single-core score of 699 and multi-core of 4952, a ratio of 7.1x as well. This consistency across Cinebench versions indicates that the scaling efficiency is stable regardless of the benchmark’s age. The single-thread performance is strong enough to handle legacy applications that rely on one or two threads, while the multi-thread performance ensures that modern parallel workloads, such as video rendering or scientific simulations, see near-linear gains.

The PassMark single-thread score of 3672 reinforces this picture. It is a solid result, though not the highest in the database, meaning the EPYC 9135 will not be a bottleneck for interactive tasks like database queries or web serving that are latency-sensitive. The multi-thread score of 57808 shows that the processor’s aggregate throughput is its primary strength, making it better suited for batch processing and heavy compute loads than for real-time single-thread responsiveness.

Who Should Consider It

The EPYC 9135 is best suited for server and workstation workloads that leverage many cores and high memory bandwidth. The 16-core, 32-thread configuration, combined with a 200 W TDP, makes it a strong candidate for virtualization hosts running multiple moderate-sized virtual machines. The PassMark multi-thread score of 57808 indicates that it can handle concurrent workloads without significant contention, and the data compression score of 737167 suggests it excels at data-heavy tasks like backup compression or log processing.

For content creation, the Cinebench R23 multi-core score of 49136 is competitive with high-end desktop processors. Video editors and 3D artists using CPU-based rendering will see solid performance, though the lack of integrated graphics means a discrete GPU is mandatory. The high single-thread scores (6936 in R23) also mean that UI interactions and plugin processing in creative applications will remain responsive.

Office and general enterprise use is a fit, but the 200 W TDP and server socket make it overkill for typical productivity tasks. The data shows strong floating-point math performance (PassMark score of 125125) and integer math (204258), which are useful for financial modeling or engineering simulations. However, the processor’s primary value is in sustained multi-threaded throughput, not in bursty, low-load office work.

Benchmark Performance

The Cinebench R23 multi-core score of 49136 places the EPYC 9135 in a strong position against its rivals. It is 0.8% ahead of the Ryzen 9 8940HX’s average, which translates to a negligible real-world difference in rendering workloads. Against the EPYC 7443P, the 1.7% lead in average score suggests that the 9135’s newer architecture more than compensates for any core count disadvantage. The 2.3% deficit to the Ryzen 9 9955HX is small enough that the EPYC could win or lose depending on the specific workload’s memory access patterns.

In Cinebench R20, the multi-core score of 20637 and single-core of 2913 show a balanced profile. The PassMark suite provides additional granularity: floating-point math (125125) and integer math (204258) are both high, while extended instructions (54795) and data encryption (40941) indicate strong cryptographic performance. The find prime numbers score of 299 is notably low, which is typical for a processor without specialized prime-number acceleration, but it does not detract from overall capability.

The average benchmark score of 83020, with nearest rivals ranging from 81661 to 85072, positions the EPYC 9135 in a narrow band of performance. The deltaPct values (0.8%, 1.7%, -2.3%, -2.4%) show that no rival is more than 2.4% ahead or 1.7% behind, making this a tightly contested segment. The data suggests that purchasing decisions among these CPUs should be based on platform features and price, rather than raw compute performance, as the differences are within measurement noise for most applications.

Platform and Compatibility

The EPYC 9135 uses the AMD Socket SP5 platform, which is designed for dual-socket or single-socket server configurations. The memory support is DDR5 with a twelve-channel bus, providing a memory bandwidth of 576.0 GB/s. This is a substantial advantage over consumer platforms, as the high bandwidth is critical for memory-bound workloads like large in-memory databases or high-performance computing. The ECC memory support ensures data integrity in mission-critical environments.

PCIe Gen 5 support with 128 lanes (CPU only) is a standout feature. This allows for extensive expansion options, including multiple high-speed NVMe drives, GPUs, or network interface cards. The 128 lanes are more than most rivals offer, making the EPYC 9135 suitable for storage servers or GPU compute nodes where I/O throughput is a bottleneck. The processor is not multiplier-unlocked, so overclocking is not possible, but the boost clock of 4.30 GHz is already high for a server part.

The architecture is Zen 5, codenamed Turin, with a 4 nm process from TSMC. The transistor count is 16,630 million across two dies, each measuring 70.6 mm². The cache hierarchy includes 80 KB of L1 per core, 1 MB of L2 per core, and 64 MB of shared L3. This cache configuration is generous and helps mitigate the latency of DDR5 memory. The release date is October 9, 2024, and the processor is currently active in production. The launch MSRP is $1214, which positions it as a mid-range EPYC offering, though the upgrade path is clear: the SP5 socket supports future Zen 5 and possibly later generations, allowing for in-place upgrades without a motherboard change.