AMD EPYC 9655P vs AMD EPYC 9745 Comparison

The AMD EPYC 9745 and AMD EPYC 9655P are both flagship-class server processors built on the Zen 5 architecture, yet they are engineered for fundamentally different workload profiles. The data shows a clear split: the 9655P dominates in raw compute speed and latency-sensitive tasks, while the 9745 excels in memory bandwidth-intensive and parallel data operations. With the 9745 holding an average benchmark score of 425,973 against the 9655P’s 396,673, the overall average favors the 9745 by 7.4%, but this aggregate figure masks a more nuanced reality where the 9655P wins 11 of 17 direct head-to-head comparisons.

Where Each One Wins

The AMD EPYC 9655P is the unequivocal winner in single-threaded and lightly-threaded performance, as well as in most multi-threaded compute benchmarks. Its advantage in Cinebench tests is striking and consistent: it leads by 18.5% in R15, R20, and R23 multi-core tests, and by 18.6% in R15 single-core, with R20 and R23 single-core margins also at 18.5%. This pattern extends to PassMark’s physics simulation, where the 9655P scores 26,810 versus 17,122 for the 9745, a 36.1% lead, and to prime number finding, where it scores 1,683 against 979, a massive 41.8% advantage. The 9655P’s single-thread PassMark score of 3,848 is 27.1% higher than the 9745’s 2,806, reinforcing its dominance in latency-sensitive, clock-dependent work.

Conversely, the AMD EPYC 9745 wins where sheer core count and memory subsystem parallelism matter more than clock speed. It takes PassMark data compression with 3,929,890 versus 3,478,283, a 13% margin, and data encryption with 229,447 versus 219,606, a 4.5% lead. The 9745’s most significant victory comes in extended instructions, where it scores 280,477 against 227,538, a 23.3% advantage. It also edges out the 9655P in floating-point math (761,219 vs 710,260, up 7.2%), integer math (1,224,315 vs 1,219,189, up 0.4%), and random string sorting (468,975 vs 455,310, up 3%). These wins indicate the 9745’s 128 cores and larger aggregate cache configuration provide superior throughput for workloads that can be parallelized across many threads with heavy data movement.

The Verdict

For workloads dominated by high-frequency execution, single-threaded responsiveness, and physics or logic simulations, the AMD EPYC 9655P is the definitive choice based on the benchmark data. Its consistent 18.5% lead across all Cinebench multi-core tests, combined with a 36.1% advantage in PassMark physics and a 41.8% lead in prime number finding, makes it the superior processor for database query processing, real-time analytics, and any application where per-core performance is the bottleneck. The 9655P also benefits from a higher boost clock of 4.50 GHz versus the 9745’s 3.70 GHz, which directly contributes to its single-thread superiority.

For environments that prioritize data compression, encryption, and extended instruction set throughput, the AMD EPYC 9745 is the better fit. Its 13% lead in data compression and 23.3% lead in extended instructions suggest it handles batch processing and cryptographic workloads more efficiently. The 9745’s 128 cores, compared to the 9655P’s 96, provide a theoretical thread advantage that materializes in floating-point math (7.2% ahead) and integer math (0.4% ahead). However, the 9745’s overall win count is lower (6 wins vs 11), and its average benchmark score advantage of 7.4% over the 9655P is driven by specific niche tests rather than broad compute superiority.

The data ultimately suggests that the 9655P is the more versatile processor for general-purpose server tasks, while the 9745 is specialized for high-throughput data operations. The 9655P’s 18.5% lead in Cinebench multi-core is particularly telling, as this benchmark often correlates with real-world rendering and simulation workloads. The 9745’s wins are narrower on average, with its largest margin being 23.3% in extended instructions, whereas the 9655P’s margins exceed 18% in nine separate tests.

Head-to-Head Benchmarks

The Cinebench results provide the clearest separation between the two processors. In Cinebench R15 multi-core, the 9655P scores 13,744 against the 9745’s 11,198, a delta of -18.5% for the 9745. This pattern repeats exactly in R20 multi-core (57,268 vs 46,659) and R23 multi-core (136,354 vs 111,093), all with -18.5% deltas. Single-core Cinebench tests follow the same trajectory: R15 shows 1,940 vs 1,580 (-18.6%), R20 shows 8,085 vs 6,586 (-18.5%), and R23 shows 19,250 vs 15,683 (-18.5%). The consistency of these margins indicates a fundamental clock-for-clock advantage for the 9655P, not a workload-specific anomaly.

PassMark physics and prime number tests amplify this trend. The physics test yields 26,810 for the 9655P versus 17,122 for the 9745, a 36.1% difference that suggests the 9655P’s higher base clock of 2.60 GHz (versus 2.40 GHz) and boost clock of 4.50 GHz (versus 3.70 GHz) provide outsized benefits in sequential, branch-heavy computations. The prime number test shows an even larger gap: 1,683 vs 979, a 41.8% delta. Single-thread PassMark confirms this, with 3,848 vs 2,806 (-27.1%).

The 9745’s wins are concentrated in data-oriented tasks. Its 23.3% lead in extended instructions (280,477 vs 227,538) is its largest margin, followed by data compression at 13% (3,929,890 vs 3,478,283). Floating-point math shows a 7.2% advantage (761,219 vs 710,260), while integer math is nearly tied at 0.4% (1,224,315 vs 1,219,189). Data encryption and random string sorting are narrower wins at 4.5% and 3%, respectively. Notably, the 9745 loses the PassMark multi-thread test by 18.5% (130,698 vs 160,417), indicating that its extra 32 cores do not translate into higher overall multi-threaded throughput in this metric.

FAQ

Q: Which processor has a higher average benchmark score?

A: The AMD EPYC 9745 has an average benchmark score of 425,973, which is 7.4% higher than the AMD EPYC 9655P’s 396,673. However, the 9655P wins 11 of 17 head-to-head comparisons, showing the average is skewed by the 9745’s large wins in niche tests.

Q: How does the core count differ between the two?

A: The AMD EPYC 9745 has 128 cores and 256 threads, while the AMD EPYC 9655P has 96 cores and 192 threads. Despite having 32 fewer cores, the 9655P leads in Cinebench R23 multi-core by 18.5% (136,354 vs 111,093).

Q: Which CPU is better for single-threaded performance?

A: The AMD EPYC 9655P is significantly better, with a PassMark single-thread score of 3,848 versus 2,806 for the 9745, a 27.1% advantage. Its Cinebench R23 single-core score of 19,250 is also 18.5% higher than the 9745’s 15,683.

Q: What are the clock speed differences?

A: The AMD EPYC 9655P has a base clock of 2.60 GHz and a boost clock of 4.50 GHz. The AMD EPYC 9745 has a base clock of 2.40 GHz and a boost clock of 3.70 GHz. The 9655P’s higher clocks directly contribute to its single-thread and physics test wins.

Q: Which processor has more L3 cache?

A: The AMD EPYC 9655P has 384 MB of shared L3 cache, while the AMD EPYC 9745 has 256 MB of shared L3 cache. The 9655P’s larger cache likely helps in its Cinebench and physics test victories.

Q: How do they compare in memory bandwidth?

A: Both processors support DDR5 memory with a twelve-channel bus and have identical memory bandwidth of 576.0 GB/s. Despite this parity, the 9745 wins in data compression (13% ahead) and encryption (4.5% ahead), suggesting core count matters more for these tasks.

Architecture Differences

The two processors share the same Zen 5 architecture and AMD Socket SP5 platform, but they are built on different process nodes and have distinct die configurations. The AMD EPYC 9745 is fabricated on TSMC’s 3 nm process, while the AMD EPYC 9655P uses a 4 nm process. The 9655P’s specifications list 99,780 million transistors across 12 dies, each measuring 70.6 mm², while the 9745’s transistor count and die size are not provided in the data. This process difference may explain why the 9655P achieves higher clock speeds despite having fewer cores.

Core configuration is the most significant architectural split: the 9745 uses the Zen 5c variant (codename Turin) with 128 cores and 256 threads, while the 9655P uses standard Zen 5 cores with 96 cores and 192 threads. Both have identical L1 cache (80 KB per core) and L2 cache (1 MB per core), but the L3 cache differs substantially: the 9655P has 384 MB shared, versus 256 MB shared for the 9745. This 128 MB L3 advantage for the 9655P likely contributes to its superior single-threaded and physics performance, as more data can reside closer to the cores.

Both processors feature identical memory support: DDR5 with a twelve-channel bus and a memory bandwidth of 576.0 GB/s, with ECC support enabled. They also share the same PCIe configuration: Gen 5 with 128 lanes (CPU only). The TDP is identical at 400 watts for both, and neither has an unlocked multiplier. Both are classified as Server/Workstation parts, were released on the same date (2024-10-09), and are currently marked as Active in production. The 9745’s launch MSRP is $12141, while the 9655P’s launch MSRP is $10811, though per benchmark data, the lower-priced 9655P wins more head-to-head tests.