AMD EPYC 9575F vs AMD EPYC 9655 Comparison

The AMD EPYC 9655 and AMD EPYC 9575F are both Zen 5 Turin parts on the same SP5 socket, but they are tuned for different workloads. The 9655 is a 96-core count monster, while the 9575F is a 64-core chip with higher clocks. The benchmark data shows a clear split: the 9655 dominates almost every multi-threaded task, while the 9575F takes the single-thread crown.

Head-to-Head Benchmarks

The most striking result is the consistency of the 9655’s wins. Across all three Cinebench versions (R15, R20, R23), the 9655 beats the 9575F by exactly 3.9% in both single-core and multi-core tests. For example, in Cinebench R23 multi-core, the 9655 scores 132,672 against 127,739 for the 9575F. Even in single-core R23, the 9655 leads with 18,730 versus 18,033. This is surprising given the 9575F’s higher rated boost clock, but the data is unambiguous.

The PassMark suite shows where the core count advantage really bites. The 9655 wins 15 of the 17 head-to-head comparisons. The biggest margins are in data encryption (210,555 vs 162,818, a 29.3% lead) and prime number finding (1,598 vs 1,215, a 31.5% lead). Floating point math also favors the 9655 heavily: 662,958 vs 526,003, a 26% gap. Integer math is similarly lopsided at 27.7% in favor of the 9655 (1,139,161 vs 891,817). Data compression shows an 18% advantage for the 9655 (3,271,896 vs 2,773,634), and random string sorting is 26.2% better on the 9655 (439,682 vs 348,500).

The 9575F’s only wins are in the two identical single-thread PassMark tests. It scores 4,173 versus 3,847 for the 9655, a 7.8% advantage. That single-thread margin is meaningful for lightly-threaded workloads, but it is the exception. Even in the PassMark multithread test, which is not as scaling-sensitive as Cinebench, the 9655 wins by 5.5% (156,110 vs 147,998). The physics test also goes to the 9655 with a 25.6% lead (25,947 vs 20,652), suggesting that even simulated physics workloads benefit from the extra cores.

Interpreting the deltaPct values, the 9655’s 15 wins include six at or above 25%, while the 9575F’s two wins are both at 7.8%. The average benchmark score reflects this: the 9655 posts 373,479 versus 311,774 for the 9575F. That is a 19.8% overall difference in the aggregate score, which is close to the 18% delta seen in data compression. The 9655 sits at the 100th percentile of all CPUs, while the 9575F is at the 99th percentile. Both are elite, but the 9655 is the top of the absolute heap.

FAQ

Q: Which CPU wins more benchmarks?

A: The AMD EPYC 9655 wins 15 of the 17 head-to-head tests. The AMD EPYC 9575F wins only the two PassMark single-thread tests.

Q: What is the single-thread performance gap?

A: In PassMark single-thread, the 9575F scores 4,173 against the 9655’s 3,847, a 7.8% lead. However, in Cinebench R23 single-core, the 9655 leads by 3.9% (18,730 vs 18,033).

Q: How much faster is the 9655 in multi-core rendering?

A: In Cinebench R23 multi-core, the 9655 scores 132,672 versus 127,739 for the 9575F, a 3.9% lead. The same 3.9% delta appears in Cinebench R15 and R20 multi-core results.

Q: Where is the 9575F competitive?

A: Only in the PassMark single-thread test. It also has a higher base clock (3.30 GHz vs 2.60 GHz) and boost clock (5.00 GHz vs 4.50 GHz) per the specs, but those clocks do not translate into wins outside that one test.

Q: How do these compare to other CPUs in their class?

A: The 9655’s nearest rival is the AMD EPYC 9535, which is 1.6% slower on average. The Intel Xeon 6960P is 2.3% slower, and the AMD EPYC 9754 is 2.5% slower. The 9575F is 0.4% faster than the AMD EPYC 9734, but 1.2% slower than the Intel Xeon 6781P and 2.8% slower than the AMD Ryzen Threadripper PRO 9985WX.

Q: What is the memory bandwidth for both?

A: Both CPUs support DDR5 with a twelve-channel memory bus and a memory bandwidth of 576.0 GB/s. They also both have 128 PCIe Gen 5 lanes (CPU only).

The Verdict

The data makes the choice simple. If your workload scales with core count, the AMD EPYC 9655 is the correct pick. It wins every multi-threaded benchmark in the comparison, often by double-digit margins. The 26% lead in floating point math and the 29.3% lead in encryption are the kind of differences that cut real time off batch jobs. The 9655’s 96 cores and 192 threads simply overwhelm the 9575F’s 64 cores and 128 threads in any throughput-oriented task.

The AMD EPYC 9575F is for a narrow niche. Its 7.8% single-thread PassMark advantage matters for software that is serialized and cannot use many cores. But even then, the Cinebench single-core results show the 9655 ahead by 3.9%, so the 9575F’s single-thread edge is not universal across all test suites. If your application is purely single-threaded and you can verify it responds to the 9575F’s higher clocks, that is the only scenario where the 9575F makes sense.

For a mixed workload server, the 9655 is the safer choice. It wins the aggregate benchmark score by 61,705 points (373,479 vs 311,774). The 9575F only wins the two single-thread tests, which are a minor part of overall performance. The 9655 also has a larger L3 cache (384 MB vs 256 MB), which helps in data-heavy workloads. In short, the 9655 is the general-purpose champion, and the 9575F is a specialist tool for single-threaded latency-sensitive tasks.

Specification Differences

The two CPUs share the same series, manufacturer, socket, architecture, codename, generation, process node, foundry, memory support, memory bus, memory bandwidth, ECC support, PCIe lanes, integrated graphics, market segment, production status, release date, multiplier status, and TDP of 400 W. They differ in the following fields:

• Cores: 96 (9655) vs 64 (9575F)
• Threads: 192 (9655) vs 128 (9575F)
• Base Clock: 2.60 GHz (9655) vs 3.30 GHz (9575F)
• Boost Clock: 4.50 GHz (9655) vs 5.00 GHz (9575F)
• Transistors: 99,780 million (9655) vs 66,520 million (9575F)
• Die Size: 12x 70.6 mm² (9655) vs 8x 70.6 mm² (9575F)
• L3 Cache: 384 MB shared (9655) vs 256 MB shared (9575F)
• Part Number: 100-000000674 (9655) vs 100-000001554 (9575F)
• Launch MSRP: $11852 (9655) vs $11791 (9575F)

The L1 and L2 cache are identical at 80 KB per core and 1 MB per core respectively. The 9655 uses 12 compute dies versus 8 for the 9575F, which explains the higher transistor count.

Architecture Differences

Both CPUs are built on TSMC’s 4 nm process node using the Zen 5 architecture with the Turin codename. The fundamental per-core design is the same: each core has 80 KB of L1 and 1 MB of L2 cache. The difference is in the number of active compute dies. The 9655 uses 12 dies of 70.6 mm² each, totaling 99,780 million transistors. The 9575F uses 8 dies of the same size, totaling 66,520 million transistors.

This die configuration directly drives the L3 cache difference. The 9655 has 384 MB of shared L3, while the 9575F has 256 MB. Since each die contributes its portion of L3, the 9655’s extra dies provide 128 MB more cache. The memory subsystem is identical in both: twelve-channel DDR5 with 576.0 GB/s bandwidth and ECC support. Both also expose the same 128 PCIe Gen 5 lanes from the CPU.

The clock strategy is the key architectural divergence. The 9575F runs at 3.30 GHz base and 5.00 GHz boost, while the 9655 runs at 2.60 GHz base and 4.50 GHz boost. Both are rated at 400 W TDP, so the 9575F concentrates that power budget into fewer cores to achieve higher clocks. The 9655 spreads the same power across more cores, accepting lower clocks for higher aggregate throughput.

Where Each One Wins

The AMD EPYC 9655 wins in every scenario that benefits from parallel execution. The Cinebench suite shows a 3.9% lead across all versions, both single and multi-core. The PassMark tests show larger wins in encryption (29.3%), prime number finding (31.5%), floating point math (26%), integer math (27.7%), and random string sorting (26.2%). Data compression is 18% better, multithread is 5.5% better, and physics is 25.6% better. Extended instructions are 2.9% better. If the workload can use more than a handful of cores, the 9655 is the answer.

The AMD EPYC 9575F wins only in PassMark single-thread performance, scoring 4,173 versus 3,847 for the 9655. That 7.8% advantage is real but isolated. It does not show up in Cinebench single-core, where the 9655 actually leads by 3.9%. The 9575F’s higher base and boost clocks do not translate into a sweeping single-thread victory across all benchmarks. Therefore, the 9575F is only the pick for applications that are strictly single-threaded and specifically responsive to the PassMark single-thread workload. For everything else, the 9655’s core count and cache size win the day.