AMD EPYC 9655 vs AMD EPYC 9845 Comparison

The AMD EPYC 9845 and AMD EPYC 9655 are both 2024 flagship server processors from the EPYC 9005 series, sharing the Turin codename and Zen 5 architecture, yet they are engineered for distinctly different workloads. The 9845 is a density-optimized part built on a 3 nm process with 160 cores, while the 9655 is a higher-clocked, full-fat Zen 5 design on 4 nm with 96 cores. Benchmark data reveals a clear split: the 9845 dominates in data-heavy and encryption tasks, while the 9655 wins in latency-sensitive and single-threaded workloads. With 6 wins for the 9845 and 11 for the 9655 in head-to-head tests, the choice is not about overall superiority but about matching silicon to specific server roles.

Where Each One Wins

The benchmark results paint a clear picture of two different server personalities. The AMD EPYC 9845 is the throughput king for compute-intensive, parallel workloads that scale with core count and shared cache. Its 160 cores and 320 threads deliver massive wins in PassMark's data compression (4,680,013 vs 3,271,896), data encryption (296,808 vs 210,555), and floating-point math (978,377 vs 662,958). These are tasks where raw core volume and memory bandwidth—both chips share a 576.0 GB/s twelve-channel DDR5 interface—translate directly into higher aggregate throughput. The 9845 also leads integer math by 48.1% and extended instructions by 54.9%, making it the go-to for scientific computing, financial modeling, and any workload that can saturate hundreds of threads.

The AMD EPYC 9655, by contrast, wins in every Cinebench test, including both single-core and multi-core variants, by a consistent 2% margin. It also takes PassMark's single-thread test decisively (3,847 vs 3,144, an 18.3% lead) and wins in physics (25,947 vs 19,631, a 24.3% advantage) and prime number finding (1,598 vs 1,255, a 21.5% lead). This profile suits database query processing, virtualization hosts with many small VMs, and latency-critical applications where per-core performance and higher clock speeds matter more than raw core count. The 9655's higher base clock of 2.60 GHz and boost clock of 4.50 GHz, compared to 2.10 GHz and 3.70 GHz on the 9845, directly explain these wins in single-threaded and lightly-threaded scenarios.

The data also shows a nuance in multi-threaded PassMark results: the 9655 edges out the 9845 in PassMark multithread (156,110 vs 152,985) despite having 64 fewer cores. This suggests that for certain mixed workloads, the 9655's higher clock speeds and larger per-core resources compensate for its lower core count. Meanwhile, the 9845 wins random string sorting by 22.4%, indicating its advantage in memory-heavy, non-cache-friendly operations. In short, choose the 9845 for batch processing, compression, and encryption at scale; choose the 9655 for interactive workloads, virtualization, and applications that demand the fastest possible response per core.

Architecture Differences

The two processors diverge significantly in their physical implementation despite sharing the Zen 5 architecture and Turin codename. The 9845 uses a 3 nm process node from TSMC, while the 9655 uses a 4 nm node. This difference allows the 9845 to pack 160 cores and 320 threads into the same AMD Socket SP5 package, but it requires lower clock speeds to manage thermals. The 9845's generation is listed as "EPYC (Zen 5c (Turin))", indicating it uses the dense "c" variant of Zen 5 cores, whereas the 9655 is "EPYC (Zen 5 (Turin))" with full-size Zen 5 cores. The 9655's larger cores are reflected in its transistor count of 99,780 million, spread across 12 dies of 70.6 mm² each, while the 9845's transistor count and die size are not specified in the data.

Cache architecture also differs substantially. Both chips have 80 KB of L1 cache and 1 MB of L2 cache per core, which is identical on a per-core basis. However, the L3 cache differs: the 9845 has 320 MB shared, while the 9655 has 384 MB shared. This 64 MB difference is notable because the 9655 has fewer cores (96 vs 160), meaning each core on the 9655 has access to a larger pool of L3 cache relative to core count. This could explain the 9655's advantage in single-threaded and physics benchmarks, where cache hits are critical. The 9845's smaller L3 per core is offset by its higher core count, which helps in throughput-heavy tasks that stream data rather than reuse it.

Both processors share the same memory support (DDR5, twelve-channel, 576.0 GB/s bandwidth, ECC enabled) and PCIe configuration (Gen 5, 128 lanes CPU-only). Neither has integrated graphics, and both are locked (no unlocked multiplier). The 9845's TDP is 390 W, while the 9655's is 400 W—a minor difference that underscores the 9655's higher clock speeds. The production status for both is "Active", and both were released on the same date: 2024-10-09. These architectural choices mean the 9845 is optimized for core density in power-constrained data centers, while the 9655 maximizes per-core performance at a slightly higher power envelope.

Head-to-Head Benchmarks

The most striking wins for the AMD EPYC 9845 come in PassMark's data encryption test, where it scores 296,808 against the 9655's 210,555—a 41% advantage. This is a massive gap that reflects the 9845's 160 cores, which can parallelize encryption algorithms far more effectively than the 9655's 96 cores. Similarly, data compression shows a 43% lead (4,680,013 vs 3,271,896), and extended instructions show a 54.9% lead (314,798 vs 203,285). The floating-point math test also goes decisively to the 9845: 978,377 vs 662,958, a 47.6% difference. Integer math follows the same pattern with a 48.1% lead (1,687,531 vs 1,139,161). These are not marginal victories; they are overwhelming advantages that make the 9845 the clear choice for any workload dominated by arithmetic or data transformation.

The AMD EPYC 9655's wins are narrower but consistent across Cinebench R15, R20, and R23, with a 2% margin in every multi-core and single-core test. For example, in Cinebench R23 multi-core, the 9655 scores 132,672 vs the 9845's 130,037, and in single-core it scores 18,730 vs 18,358. These small deltas in Cinebench suggest that the 9655's higher boost clock (4.50 GHz vs 3.70 GHz) provides a slight edge even in multi-core rendering, which is surprising given the 9845's core advantage. The 9655's bigger wins come in PassMark physics (25,947 vs 19,631, a 24.3% lead), prime number finding (1,598 vs 1,255, a 21.5% lead), and single-thread (3,847 vs 3,144, an 18.3% lead). The PassMark multithread test also goes to the 9655 by 2% (156,110 vs 152,985), reinforcing the idea that its per-core efficiency can overcome core-count deficits in certain mixed workloads.

The only test where the 9845 wins by a smaller margin is random string sorting, where it leads 538,060 vs 439,682, a 22.4% advantage. This workload likely benefits from the 9845's larger total L3 cache (320 MB) and higher core count, even though the 9655 has more L3 per core. Overall, the head-to-head data shows a clear pattern: the 9845 wins where parallelism scales, and the 9655 wins where clock speed and cache per core matter. The 9845's average benchmark score is 523,613, significantly higher than the 9655's 373,479, but this average is skewed by the 9845's huge wins in data-heavy tests. The 9655's nearest rivals include the AMD EPYC 9535 (avg score 379,408, deltaPct -1.6) and Intel Xeon 6960P (avg 365,194, deltaPct 2.3), showing it competes closely with other high-core server parts, while the 9845 sits 3.5% above the AMD EPYC 9755 (avg 505,778) but 15.6% below the AMD EPYC 9965 (avg 620,487).

Specification Differences

The key specification differences between the two processors are stark. The 9845 has 160 cores and 320 threads, while the 9655 has 96 cores and 192 threads—a 64-core and 128-thread gap. Base clock speeds differ: 2.10 GHz on the 9845 vs 2.60 GHz on the 9655. Boost clocks also favor the 9655: 3.70 GHz vs 4.50 GHz. The TDP is slightly higher on the 9655 at 400 W vs 390 W on the 9845. The process node is the biggest manufacturing difference: 3 nm for the 9845 vs 4 nm for the 9655, both from TSMC. The 9655 has a specified transistor count of 99,780 million and a die size of 12x 70.6 mm², while the 9845's transistor count and die size are not provided. L3 cache differs: 320 MB shared on the 9845 vs 384 MB shared on the 9655. L1 and L2 cache are identical per core (80 KB and 1 MB, respectively). Memory support, PCIe lanes, socket, and integrated graphics are identical. The launch MSRP is $13,564 for the 9845 and $11,852 for the 9655—the only price-related fact available, stated once here.

FAQ

Q: Which processor has more cores and threads?

A: The AMD EPYC 9845 has 160 cores and 320 threads, while the AMD EPYC 9655 has 96 cores and 192 threads.

Q: Why does the EPYC 9655 win in Cinebench multi-core tests despite having fewer cores?

A: The 9655 wins all Cinebench R15, R20, and R23 multi-core tests by 2% because its higher base clock (2.60 GHz vs 2.10 GHz) and boost clock (4.50 GHz vs 3.70 GHz) compensate for its lower core count, along with a larger 384 MB L3 cache vs 320 MB on the 9845.

Q: In which benchmarks does the EPYC 9845 have the largest advantage?

A: The 9845's biggest wins are in PassMark extended instructions (54.9% lead), floating-point math (47.6% lead), integer math (48.1% lead), data compression (43% lead), and data encryption (41% lead).

Q: What is the manufacturing process difference between the two chips?

A: The EPYC 9845 is built on a 3 nm process node from TSMC, while the EPYC 9655 uses a 4 nm node. The 9845 uses Zen 5c (Turin) cores, while the 9655 uses full Zen 5 (Turin) cores.

Q: Do both processors support the same memory and PCIe configurations?

A: Yes, both support DDR5 with a twelve-channel memory bus and 576.0 GB/s bandwidth, ECC memory, and PCIe Gen 5 with 128 lanes (CPU only). They also share the same AMD Socket SP5.

Q: Which processor has a higher single-threaded performance?

A: The EPYC 9655 wins PassMark single-thread with a score of 3,847 vs 3,144 on the 9845, an 18.3% advantage, and also wins all Cinebench single-core tests by 2%.