AMD EPYC 9655P vs AMD EPYC 9965 Comparison
AMD EPYC 9655P
EPYC 9965
PERFORMANCE BENCHMARKS
Analysis: AMD EPYC 9655P vs AMD EPYC 9965
The AMD EPYC 9965 and AMD EPYC 9655P are both members of the EPYC 9005 series, built on the Zen 5 architecture and released on the same date. Despite sharing a generation, they are engineered for different extremes of the server market. The 9965 is a 192-core flagship, while the 9655P is a 96-core part with higher clock speeds. Benchmark data reveals a fascinating split: in threaded productivity and data-heavy workloads, the 9965 dominates, but in specific latency-sensitive tasks, the 9655P pulls ahead. This analysis walks through the benchmark results, architectural differences, and the use cases where each processor is the logical choice.
Head-to-Head Benchmarks
The most striking pattern in the data is the near-tie in Cinebench tests. Across all six Cinebench R15, R20, and R23 runs—both single-core and multi-core—the AMD EPYC 9965 wins by a margin of exactly 0.2%. For instance, in Cinebench R23 multi-core, the 9965 scores 136,661 against the 9655P’s 136,354. In single-core R23, the scores are 19,293 versus 19,250. This indicates that for rendering workloads using these specific tests, the massive core-count advantage of the 9965 is neutralized, likely because the tests scale differently or the 9655P’s higher clock speed compensates for having half the cores.
The PassMark suite tells a completely different story. The 9965’s wins are not marginal; they are decisive. In data compression, the 9965 scores 5,679,990 against 3,486,158, a delta of 62.9%. Data encryption shows a 58.3% advantage (348,449 vs. 220,074). The largest margin is in extended instructions, where the 9965 leads by 66.2% (383,298 vs. 230,609). Floating-point math also favors the 9965 heavily, with a 61.1% lead (1,153,453 vs. 715,866), and integer math is 57.2% higher (1,926,069 vs. 1,225,251). Random string sorting shows a 40.1% edge for the 9965 (633,030 vs. 451,824).
However, the 9655P wins four benchmarks, and they are revealing. The most significant is PassMark single-thread, where it scores 3,849 versus the 9965’s 3,176—a 17.5% advantage. It also wins in find prime numbers (1,686 vs. 1,208, a 28.4% lead) and physics (25,847 vs. 18,707, a 27.6% lead). The PassMark multithread score is essentially a tie, with the 9965 at 160,542 and the 9655P at 160,490, a delta of 0%. This suggests that in a generic multithreaded stress test without specialized instruction sets, the two CPUs are functionally equivalent, despite the 9965’s doubling of cores and threads.
Architecture Differences
The core architectural divergence is in the process node. The AMD EPYC 9965 is fabricated on a 3 nm process, while the AMD EPYC 9655P uses a 4 nm process. Both are from TSMC. This node difference is reflected in their core counts: the 9965 packs 192 cores and 384 threads, while the 9655P has 96 cores and 192 threads. The 9655P compensates for fewer cores with a higher base clock of 2.60 GHz and a boost clock of 4.50 GHz, compared to the 9965’s 2.25 GHz base and 3.70 GHz boost.
The cache hierarchy is identical on paper: both have 80 KB of L1 per core, 1 MB of L2 per core, and 384 MB of shared L3 cache. The 9655P’s datasheet lists 99,780 million transistors across 12 dies of 70.6 mm² each, but the 9965’s transistor count and die size are not specified. Memory support is also identical: DDR5 with a twelve-channel bus and 576.0 GB/s of bandwidth, plus ECC support. Both use AMD Socket SP5 and provide PCIe Gen 5 with 128 lanes (CPU only). Neither has integrated graphics. The 9655P is labeled with the "P" suffix, indicating a single-socket optimized part, while the 9965 is not.
Where Each One Wins
The AMD EPYC 9965 is the clear winner for throughput-oriented, parallel workloads that heavily utilize integer and floating-point math. The data shows it is 57.2% ahead in integer math and 61.1% ahead in floating-point math. Its 62.9% lead in data compression and 58.3% lead in data encryption point to strength in database, archival, and security-related tasks. The 66.2% margin in extended instructions suggests it is superior for workloads that use advanced SIMD or specialized instruction sets, such as scientific computing or financial modeling. For any task where the dataset can be split across hundreds of threads, the 9965’s raw core count delivers massive gains.
The AMD EPYC 9655P wins in single-threaded performance and specific latency-bound tasks. Its 17.5% lead in PassMark single-thread is significant for applications that rely on per-core speed, such as legacy software, certain database queries, or front-end web serving. The 28.4% win in find prime numbers indicates an advantage in workloads with heavy branching and sequential dependencies. The 27.6% win in physics suggests better performance in simulation engines that do not scale perfectly with core count. The near-tie in PassMark multithread (0% delta) is a critical data point: for a generic multithreaded load without specialized math, the 9655P matches the 9965, making it a more efficient choice for such tasks.
FAQ
Q: Which processor has more cores and threads?
A: The AMD EPYC 9965 has 192 cores and 384 threads, while the AMD EPYC 9655P has 96 cores and 192 threads.
Q: Is there a performance difference in Cinebench benchmarks?
A: The difference is minimal. In all Cinebench R15, R20, and R23 tests, the AMD EPYC 9965 wins by a margin of 0.2%, with scores like 136,661 versus 136,354 in R23 multi-core.
Q: Which CPU has a higher clock speed?
A: The AMD EPYC 9655P has a higher base clock of 2.60 GHz and boost clock of 4.50 GHz, compared to the AMD EPYC 9965’s 2.25 GHz base and 3.70 GHz boost.
Q: What is the difference in manufacturing process?
A: The AMD EPYC 9965 is built on a 3 nm process, while the AMD EPYC 9655P uses a 4 nm process. Both are fabricated by TSMC.
Q: Are there any workloads where the 9655P beats the 9965?
A: Yes, the 9655P wins in PassMark single-thread (3,849 vs. 3,176), find prime numbers (1,686 vs. 1,208), and physics (25,847 vs. 18,707).
Q: How do they compare in memory bandwidth and cache?
A: They are identical. Both have 384 MB of shared L3 cache, 80 KB L1 per core, 1 MB L2 per core, and support DDR5 with a twelve-channel bus and 576.0 GB/s bandwidth.
Specification Differences
| Specification | AMD EPYC 9965 | AMD EPYC 9655P |
|----------------|----------------|-----------------|
| Cores | 192 | 96 |
| Threads | 384 | 192 |
| Base Clock | 2.25 GHz | 2.60 GHz |
| Boost Clock | 3.70 GHz | 4.50 GHz |
| TDP | 500 W | 400 W |
| Process Node | 3 nm | 4 nm |
| Transistors | Not specified | 99,780 million |
| Die Size | Not specified | 12x 70.6 mm² |
| Generation | EPYC (Zen 5c (Turin)) | EPYC (Zen 5 (Turin)) |
| Launch MSRP | $14,813 | $10,811 |
The Verdict
The data presents a clear dichotomy. The AMD EPYC 9965 is the performance king for massively parallel, compute-intensive workloads. Its 62.9% lead in data compression and 66.2% lead in extended instructions demonstrate that for any server handling big data analytics, high-frequency trading algorithms, or scientific simulations, the 9965 is the superior choice. Its 100th percentile ranking among all CPUs and an average benchmark score of 620,487, which is 18.5% higher than its nearest rival (the EPYC 9845), reinforces its position at the top of the stack. If the workload can use more than 96 cores efficiently, the 9965’s 192 cores will provide a substantial return.
The AMD EPYC 9655P, while having half the cores, is not a lesser product—it is a different tool. Its higher clock speeds (4.50 GHz boost vs. 3.70 GHz) make it the winner for single-threaded and latency-sensitive tasks. The 17.5% lead in PassMark single-thread and 27.6% lead in physics are decisive for applications like real-time analytics, certain gaming servers, or database transaction processing where per-core speed is paramount. Its average benchmark score of 397,773 is only 3.5% behind the Ryzen Threadripper PRO 9995WX, yet it offers 192 threads. The near-tie in PassMark multithread (0% delta) is the key indicator: if you cannot saturate 192 cores, the 9655P delivers comparable generic multithreaded performance with a lower TDP of 400 W versus 500 W. The choice hinges on the workload’s scaling behavior: choose the 9965 for maximum core throughput, and the 9655P for balanced performance with superior single-core speed.