AMD EPYC 9565 vs AMD EPYC 9655 Comparison

Head-to-Head Benchmarks

The AMD EPYC 9655 and AMD EPYC 9565 are both Zen 5 Turin parts on the SP5 platform, but the benchmark data shows a clear division of labor. Across 17 head-to-head tests, the 9655 takes 15 wins, while the 9565 claims 2. The margins are not uniform, and the two 9565 victories are substantial enough to matter for specific workloads.

Starting with the Cinebench suite, the 9655 wins every single test by exactly 15.4%. That consistency is striking. Cinebench R15 multicore shows 13373 versus 11585, R20 multicore shows 55722 versus 48273, and R23 multicore shows 132672 versus 114937. Single-core results follow the same pattern: R15 single-core is 1887 versus 1635, R20 single-core is 7866 versus 6814, and R23 single-core is 18730 versus 16226. The 9655's higher boost clock of 4.50 GHz versus 4.30 GHz explains much of the single-thread edge, while the 96-core count versus 72 cores drives the multicore advantage. The fact that both the single-core and multicore deltas are identical at 15.4% suggests the per-core frequency difference and the core-count scaling are perfectly aligned in this workload.

PassMark multithread also lands at 156110 versus 135221, a 15.4% margin, reinforcing the Cinebench pattern. PassMark single-thread is closer: 3847 versus 3696, a 4.1% advantage for the 9655. That smaller gap makes sense, the single-thread test does not scale with core count, so the difference is purely clock speed and IPC, and the 9655's boost advantage is modest.

The larger deltas appear in workloads that scale with core count or memory subsystem throughput. PassMark integer math shows the biggest gap: 1139161 versus 717948, a 58.7% advantage for the 9655. With 96 cores versus 72, the 9655 has 33% more cores, yet it wins integer math by nearly 59%, indicating the workload benefits from more than just raw core count. PassMark data encryption is another outlier: 210555 versus 141936, a 48.3% edge. Encryption workloads often stress cache and memory bandwidth, and the 9655 benefits from the same 384 MB L3 cache and 576.0 GB/s twelve-channel DDR5 bandwidth, but with more cores to feed. PassMark physics shows 25947 versus 18036, a 43.9% margin, and random string sorting shows 439682 versus 291941, a 50.6% margin. These are all multithreaded, cache-sensitive workloads where the extra 24 cores of the 9655 translate into outsized gains.

Data compression is another clear 9655 win: 3271896 versus 2579631, a 26.8% edge. Floating-point math also favors the 9655 at 662958 versus 549422, a 20.7% margin, though this is smaller than the integer math gap. The 9565's two wins are notable. PassMark extended instructions goes to the 9565 at 209595 versus 203285, a 3% edge. This is a modest margin, but it indicates that the 9565's higher base clock of 3.15 GHz versus 2.60 GHz gives it an advantage in workloads that do not scale perfectly across cores and are sensitive to sustained clock speed rather than boost frequency. The larger 9565 win is in find prime numbers: 2422 versus 1598, a 34% advantage. Prime number finding is typically a single-threaded or lightly threaded integer workload, and the 9565's combination of higher base clock and lower core count means less thermal and power contention per core. The result is a workload where the 9565 is decisively faster, despite having fewer cores.

The average benchmark scores reflect the overall positioning: the 9655 averages 373479, while the 9565 averages 285471. That is a 30.8% difference in average score. The 9655 sits at the 100th percentile of all CPUs in the database, while the 9565 sits at the 99th percentile. Both are at the very top of the performance stack, but the 9655 is the higher-performing part in aggregate.

Where Each One Wins

The 9655 is the clear choice for heavily multithreaded, core-scaled workloads. The data shows it winning integer math by 58.7%, data encryption by 48.3%, physics by 43.9%, and random string sorting by 50.6%. These are the kinds of workloads found in database processing, scientific computing, virtualization hosts, and large-scale data analytics. The 96 cores and 192 threads provide raw parallel throughput that the 72-core 9565 cannot match. The Cinebench results, all at exactly 15.4% in favor of the 9655, confirm that rendering and 3D modeling workloads will see a consistent, predictable advantage. For any environment where the scheduler can keep all cores busy, the 9655 is the better part.

The 9565 wins in two specific areas. The first is extended instructions, where it beats the 9655 by 3%. This is a narrow margin, but it suggests that the 9565's higher base clock helps in workloads that use AVX or other extended instruction sets without fully saturating all cores. The second and more significant win is find prime numbers, where the 9565 is 34% faster. This workload is sensitive to per-core integer performance and clock speed under sustained load. The 9565's base clock of 3.15 GHz, compared to 2.60 GHz on the 9655, gives it a meaningful advantage in lightly threaded or latency-sensitive integer tasks. The 9565 also holds up well in single-threaded tests, trailing by only 4.1% in PassMark single-thread, which is acceptable given its lower core count.

The 9565 also has a lower launch MSRP of $10486 compared to the 9655's $11852. The data does not show any performance-per-dollar metrics, but the price difference is recorded. For workloads that are primarily single-threaded or lightly threaded, the 9565 is competitive despite the core deficit. For workloads that can use 96 cores, the 9655 is the stronger part.

The Verdict

The AMD EPYC 9655 is the higher-performing processor in the database's measurements. It wins 15 of 17 head-to-head tests, including every Cinebench test, every major multithreaded PassMark test, and both data compression and encryption tests. Its average benchmark score of 373479 is significantly higher than the 9565's 285471, and it is the only one of the two that reaches the 100th percentile of all CPUs. The 9655 is the part to choose for heavily multithreaded server workloads, rendering farms, and any application that can scale across 96 cores and 192 threads. The 15.4% advantage in Cinebench and multithread tests is consistent and repeatable across the entire benchmark suite.

The AMD EPYC 9565 is not a weak processor. It sits at the 99th percentile, and its average score of 285471 is close to the Intel Xeon 696X at 286102, trailing by just 0.2%. It wins find prime numbers by 34% and extended instructions by 3%, and it trails by only 4.1% in single-threaded performance. For workloads that are latency-sensitive, lightly threaded, or dominated by prime number or extended instruction execution, the 9565 is the better choice. Its higher base clock of 3.15 GHz gives it an edge in sustained single-thread and low-thread workloads. The 9565 is also the lower-priced part, with a launch MSRP of $10486 versus $11852.

The decision is straightforward. Choose the 9655 for maximum multithreaded throughput and for workloads that can use 96 cores. Choose the 9565 for lightly threaded integer workloads, extended instruction workloads, and scenarios where the higher base clock matters more than core count. The data does not support a single universal winner, it supports two different tools for two different jobs.

FAQ

Q: How much faster is the AMD EPYC 9655 in Cinebench R23 multicore?

A: The 9655 scores 132672 versus 114937 for the 9565, a 15.4% advantage.

Q: In which benchmark does the AMD EPYC 9565 show the largest lead?

A: The 9565 wins passmark find prime numbers with a score of 2422 versus 1598, a 34% advantage.

Q: What is the single-threaded performance difference between the two?

A: The 9655 scores 3847 in PassMark single-thread versus 3696 for the 9565, a 4.1% edge for the 9655.

Q: How do these processors compare to their nearest rivals?

A: The 9655 averages 373479, which is 2.3% ahead of the Intel Xeon 6960P and 2.5% ahead of the AMD EPYC 9754. The 9565 averages 285471, which is 0.2% behind the Intel Xeon 696X and 0.6% behind the AMD EPYC 9555P.

Q: Which processor has better data encryption performance?

A: The 9655 scores 210555 in PassMark data encryption versus 141936 for the 9565, a 48.3% advantage.

Q: Do both processors support the same memory configuration?

A: Yes, both support DDR5 with a twelve-channel memory bus and 576.0 GB/s bandwidth, and both support ECC memory.

Architecture Differences

Both the AMD EPYC 9655 and AMD EPYC 9565 are built on the Zen 5 architecture with the Turin codename, belong to the EPYC 9005 series, and are manufactured by TSMC on a 4 nm process node. Both use the AMD Socket SP5 and have a TDP of 400. The transistor count is identical at 99,780 million, and both use 12 chiplets, each with a die size of 70.6 mm². Neither has integrated graphics, and both are unlocked for overclocking, meaning they are fixed at their rated clocks.

The core counts differ significantly. The 9655 has 96 cores and 192 threads, while the 9565 has 72 cores and 144 threads. The 9655 has a base clock of 2.60 GHz and a boost clock of 4.50 GHz. The 9565 has a base clock of 3.15 GHz and a boost clock of 4.30 GHz. The 9565 runs a higher base clock, which explains its wins in lightly threaded workloads, while the 9655 has a higher boost clock and 24 more cores.

Cache configurations are identical. Both have 80 KB of L1 cache per core, 1 MB of L2 cache per core, and 384 MB of shared L3 cache. Neither has 3D V-Cache. Memory support is also identical: DDR5 with a twelve-channel bus and 576.0 GB/s bandwidth, with ECC support. PCIe is Gen 5 with 128 lanes on the CPU. Both parts were released on the same date, 2024-10-09, and both are active in production.

The key architectural difference is purely the core count and clock speeds. The 9655 trades a lower base clock for more cores, while the 9565 trades cores for a higher base clock. Both have the same L3 cache, memory bandwidth, PCIe lanes, and process node, so the performance differences in the benchmark data can be attributed almost entirely to the core count and clock speed configuration. The 9565's higher base clock gives it a 34% win in find prime numbers and a 3% win in extended instructions, while the 9655's additional 24 cores give it margins of 58.7% in integer math, 48.3% in encryption, and 50.6% in random string sorting. The architecture is the same, but the configuration creates two distinctly different performance profiles.