AMD EPYC 9655P vs AMD EPYC 9845 Comparison

Both the AMD EPYC 9845 and the AMD EPYC 9655P are active members of the EPYC 9005 series, built on the Zen 5 architecture and codenamed Turin, targeting the server and workstation market segment. They share the same AMD Socket SP5 platform, support DDR5 memory across a twelve-channel bus with a 576.0 GB/s bandwidth, and offer Gen 5 PCIe with 128 lanes from the CPU, while lacking integrated graphics. Despite these similarities, the data reveals fundamentally different design philosophies: the 9845 maximizes core count for throughput, while the 9655P prioritizes higher frequencies and a distinct chip layout. Benchmark results show a clear split, with the 9655P dominating in single-threaded and multi-threaded rendering workloads, but the 9845 winning decisively in a range of PassMark compute tasks. This analysis breaks down their architectural differences and identifies where each processor excels based solely on the provided performance data.

FAQ

Q: Which processor has more cores and threads?

A: The AMD EPYC 9845 has 160 cores and 320 threads, while the AMD EPYC 9655P has 96 cores and 192 threads. The 9845 offers a substantial increase in core count, which is key for its performance in parallel workloads.

Q: How do their single-core performances compare?

A: The AMD EPYC 9655P is the clear winner in single-core performance. It achieves a Cinebench R23 single-core score of 19250 compared to the 9845's 18358, and a PassMark single-thread score of 3849 versus 3144, a lead of 18.3% in the latter test.

Q: Which CPU is better for Cinebench multi-core rendering?

A: Despite having fewer cores, the AMD EPYC 9655P wins all Cinebench multi-core tests. In Cinebench R23 multi-core, it scores 136354 against the 9845's 130037, a 4.6% advantage.

Q: In which specific areas does the AMD EPYC 9845 win?

A: The 9845 wins in all PassMark tests related to data processing and math, including data compression (34.2% ahead), data encryption (34.9% ahead), extended instructions (36.5% ahead), floating point math (36.7% ahead), integer math (37.7% ahead), and random string sorting (19.1% ahead).

Q: What are the differences in process node and L3 cache?

A: The EPYC 9845 is built on a 3 nm process from TSMC and has 320 MB of shared L3 cache. The EPYC 9655P is built on a 4 nm process from TSMC and has 384 MB of shared L3 cache, which is larger despite the older node.

Q: What is the difference in their average benchmark scores?

A: The AMD EPYC 9845 has a higher average benchmark score of 523613, while the AMD EPYC 9655P has an average score of 397773. This reflects the 9845's dominance in the math-intensive PassMark tests.

Architecture Differences

The most significant architectural difference lies in the core configuration and physical design, not the fundamental Zen 5 instruction set. The EPYC 9845 is based on the Zen 5c variant, designed for density, whereas the EPYC 9655P uses the full Zen 5 design. This is evident in their process nodes: the 9845 is manufactured on a 3 nm process from TSMC, while the 9655P uses a 4 nm process. The 9655P is the only one with listed physical dimensions, consisting of 12 chiplets each measuring 70.6 mm², containing a total of 99,780 million transistors. The 9845's die size and transistor count are not specified, but its higher core count (160 vs 96) and smaller process node indicate a denser packaging strategy.

Clock speeds also differ, with the 9655P running at a 2.60 GHz base clock and 4.50 GHz boost clock, compared to the 9845's 2.10 GHz base and 3.70 GHz boost. This substantial clock advantage for the 9655P explains its single-threaded superiority. Cache hierarchies are similar per core, with both offering 80 KB of L1 and 1 MB of L2 per core, but the total L3 cache differs: the 9845 has 320 MB shared, while the 9655P has 384 MB shared. The 9655P's larger L3 cache, despite fewer cores, suggests a different balance toward per-core resources. Both processors share the same external platform features, including the AMD Socket SP5, a twelve-channel DDR5 memory bus with 576.0 GB/s bandwidth, and 128 Gen 5 PCIe lanes, but their internal architectures are tuned for opposing goals: massive parallel throughput versus higher per-core performance.

Where Each One Wins

The benchmark results show a clear division of labor. The AMD EPYC 9655P is the winner in all six Cinebench tests, covering both single and multi-core workloads, as well as the PassMark multithread test. This indicates a distinct advantage in rendering applications and general CPU throughput tasks that are heavily optimized for frequency and per-core performance. Its 24% lead in the PassMark physics test and 25.6% lead in finding prime numbers further solidify its position for simulation and computational tasks that rely on high clock speeds and efficient single-thread execution. The 9655P’s profile is ideal for workloads where a smaller number of powerful cores is more valuable than a massive core count.

Conversely, the AMD EPYC 9845 is the definitive winner in PassMark data processing and mathematical workloads. It holds a commanding lead in integer math (37.7%), floating point math (36.7%), extended instructions (36.5%), and data encryption (34.9%). It also wins in data compression (34.2%) and random string sorting (19.1%). These results indicate that the 9845’s massive core count is the deciding factor in parallelizable tasks involving large datasets, heavy encryption, and complex mathematical calculations. For a server handling database workloads, virtualization, or scientific computing that scales across hundreds of threads, the 9845 is the superior choice, effectively trading per-core speed for sheer parallel processing capability.

Specification Differences

• Cores: 160 (9845) vs 96 (9655P)
• Threads: 320 (9845) vs 192 (9655P)
• Base Clock: 2.10 GHz (9845) vs 2.60 GHz (9655P)
• Boost Clock: 3.70 GHz (9845) vs 4.50 GHz (9655P)
• TDP: 390 W (9845) vs 400 W (9655P)
• Process Node: 3 nm (9845) vs 4 nm (9655P)
• Transistors: Not specified (9845) vs 99,780 million (9655P)
• Die Size: Not specified (9845) vs 12x 70.6 mm² (9655P)
• L3 Cache: 320 MB shared (9845) vs 384 MB shared (9655P)
• Architecture Generation: EPYC (Zen 5c (Turin)) (9845) vs EPYC (Zen 5 (Turin)) (9655P)
• Launch MSRP: $13564 (9845) vs $10811 (9655P)

Head-to-Head Benchmarks

The most striking result is the 9845's dominance in the PassMark math and data benchmarks. In integer math, the 9845 scores 1687531 against the 9655P's 1225251, a substantial 37.7% lead. Similarly, in floating point math, it wins with 978377 versus 715866, a 36.7% advantage. The 9845 also excels in extended instructions, scoring 314798 compared to 230609, a 36.5% gap, and in data encryption, where it leads 296808 to 220074, a 34.9% difference. Data compression shows a similar story, with the 9845 winning 4680013 to 3486158, a 34.2% lead. Even in random string sorting, a less math-intensive task, the 9845 wins with 538060 versus 451824, a 19.1% advantage.

However, the 9655P fights back decisively in other areas. In the PassMark single-thread test, it wins with a score of 3849 against 3144, a significant 18.3% lead. This advantage carries over to the physics test, where it scores 25847 versus 19631, a 24% lead, and in finding prime numbers, where it wins 1686 to 1255, a 25.6% advantage. The 9655P also sweeps the Cinebench suite, winning the R15, R20, and R23 multi-core and single-core tests by a consistent 4.6% margin. For example, in Cinebench R23 multi-core, it scores 136354 against the 9845's 130037, and in single-core it scores 19250 versus 18358. The PassMark multithread test is also won by the 9655P, with a score of 160490 against 152985, a 4.7% lead. This split shows that while the 9845 is a compute monster for specific workloads, the 9655P is a more balanced performer with a significant edge in frequency-sensitive applications.

The Verdict

The choice between these two processors hinges entirely on workload characteristics. For users running heavily parallel, math-intensive, or data-processing applications, the AMD EPYC 9845 is the clear winner. Its 160 cores and 320 threads deliver a 34-38% advantage in integer math, floating point math, encryption, and data compression. The data indicates that any workload that can fully utilize its core count will see a substantial performance boost, making it the superior choice for scientific computing, financial modeling, and data analytics. Its higher average benchmark score of 523613 reflects its strength in these areas, placing it 3.5% ahead of the AMD EPYC 9755 in its nearest rival list.

Conversely, the AMD EPYC 9655P is the better pick for general-purpose server tasks, rendering, and applications that benefit from high clock speeds. It wins all Cinebench tests, including multi-core, and offers an 18.3% lead in single-thread performance. The 9655P’s higher boost clock of 4.50 GHz makes it more responsive for interactive workloads and tasks with lower thread counts. Its 24% lead in the PassMark physics test and 25.6% lead in prime number finding highlight its strengths in simulation and scientific workloads that require fast per-core calculation. While it has fewer cores, its 4.7% lead in the PassMark multithread test suggests that its higher frequency and larger L3 cache can overcome its core deficit in some multi-threaded scenarios. Ultimately, the 9845 is for raw parallel throughput, while the 9655P is for balanced, high-frequency performance.