AMD EPYC 9575F vs AMD EPYC 9965 Comparison

Where Each One Wins

The benchmark data splits these two AMD EPYC parts into clearly different roles. The AMD EPYC 9965 wins 13 of the 17 recorded head-to-head comparisons, and it does so by margins that range from modest to overwhelming. Its strengths cluster around throughput-heavy workloads: data compression, encryption, extended instructions, floating-point math, integer math, and random string sorting all go decisively to the 192-core part.

The AMD EPYC 9575F takes the remaining four wins, and they are just as telling. It wins in single-thread performance, physics simulation, and prime number finding. Those are workloads where raw clock speed and per-core efficiency matter more than sheer core count. The 9575F also wins in the PassMark multithread test? No, that actually goes to the 9965 by 8.5%. The 9575F's wins are narrower in scope but significant for specific applications.

The pattern is consistent across the Cinebench suite. The 9965 wins every Cinebench R15, R20, and R23 test, both single-core and multi-core, by exactly 7% in each case. That is a uniform margin that suggests architectural parity with a slight advantage in clock behavior or memory latency for the 9965 in those specific tests. The 9575F, despite its higher boost clock, does not overcome that gap in Cinebench.

For server workloads, the database shows the 9965 as the clear choice for virtualization, batch processing, and any workload that can scale across 192 cores and 384 threads. The 9575F is the better pick for latency-sensitive single-threaded tasks, physics-based simulations, and applications that cannot effectively use more than 64 cores. Its 4 wins are all in areas where the 9965's core advantage becomes a liability due to lower per-core frequency.

Architecture Differences

Both processors sit on AMD Socket SP5 and belong to the EPYC 9005 series, but they are built from different dies. The 9965 uses Zen 5c architecture on a 3 nm process from TSMC, while the 9575F uses full Zen 5 architecture on a 4 nm process. This is a critical distinction: the "c" in Zen 5c denotes a denser, more power-efficient implementation, not a crippled one. The 9965 packs 192 cores and 384 threads into its package, while the 9575F offers 64 cores and 128 threads.

The 9575F compensates for its lower core count with significantly higher clocks. Its base clock is 3.30 GHz and boost clock is 5.00 GHz, compared to the 9965's 2.25 GHz base and 3.70 GHz boost. That 1.30 GHz boost advantage explains the 9575F's dominance in single-threaded tests. The 9965's 500 W TDP is higher than the 9575F's 400 W, but the 9965 also has to feed three times as many cores.

Cache allocation differs as well. Both have 80 KB L1 and 1 MB L2 per core. The 9965 has 384 MB of shared L3 cache, while the 9575F has 256 MB. That extra 128 MB on the 9965 helps with large working sets in multi-threaded workloads, though the 9575F's higher clocks can compensate in latency-bound scenarios.

Memory support is identical: DDR5, twelve-channel, 576.0 GB/s bandwidth, with ECC support. PCIe is also the same: Gen 5 with 128 lanes from the CPU. The 9575F lists 66,520 million transistors across 8 dies of 70.6 mm² each, while the 9965 does not provide transistor or die size data in the database. The 9575F's transistor count is notable because it uses a larger process node (4 nm vs 3 nm) yet still delivers high clock speeds, indicating a less dense but potentially more frequency-friendly design.

Head-to-Head Benchmarks

The largest win for the 9965 comes in floating-point math, where it scores 1,153,453 against the 9575F's 526,003, a 119.3% advantage. That is more than double the throughput. Integer math shows a similar trend: 1,926,069 versus 891,817, a 116% lead. These are the kinds of workloads that scale almost linearly with core count, and the 9965's 192 cores dominate.

Data compression is another blowout. The 9965 scores 5,679,990 versus 2,773,634, a 104.8% advantage. Data encryption follows at 348,449 versus 162,818, a 114% lead. Extended instructions go 383,298 versus 197,484, a 94.1% gap. Random string sorting shows 633,030 versus 348,500, an 81.6% lead. These are all cases where the 9965's core count translates directly into higher aggregate throughput.

The multi-threaded PassMark test is closer. The 9965 scores 160,542 versus 147,998, an 8.5% win. That is a surprisingly small margin given the core count difference, which suggests the 9575F's higher clocks keep it competitive in mixed workloads. The Cinebench results are also tight: 7% for the 9965 across R15, R20, and R23, both single and multi-core. For example, Cinebench R23 multi-core shows 136,661 versus 127,739, and single-core shows 19,293 versus 18,033.

The 9575F's wins are narrower but clear. Single-thread PassMark shows 4,173 versus 3,176, a 23.9% advantage. That is the largest single win for the 9575F. Physics simulation goes 20,652 versus 18,707, a 9.4% lead. Prime number finding is nearly a tie: 1,215 versus 1,208, a 0.6% edge for the 9575F. These results show that in latency-sensitive, single-threaded, or branch-heavy workloads, the 9575F's 5.00 GHz boost clock matters more than having three times the cores.

The Verdict

The data supports a clear separation of roles. The AMD EPYC 9965 is the throughput champion. Its 13 wins include the most demanding math, encryption, and compression workloads, with margins from 81.6% to 119.3% over the 9575F. It also holds a 7% edge across the entire Cinebench suite, which is notable because Cinebench often rewards both core count and clock speed. For any workload that can use more than 64 cores, the 9965 is the correct choice.

The AMD EPYC 9575F is the latency specialist. Its 4 wins are all in areas where clock speed trumps core count. The 23.9% single-thread advantage is substantial, and the 9.4% physics win indicates strong per-core performance under specific instruction patterns. For databases, financial modeling, or any single-threaded application, the 9575F will feel snappier. Its 400 W TDP is also lower, which may simplify cooling requirements in dense servers.

The database shows the 9965 with a 100th percentile ranking among all CPUs, while the 9575F sits at 99th. The 9965's average benchmark score is 620,487, compared to 311,774 for the 9575F. The nearest rivals tell a similar story: the 9965 leads the EPYC 9845 by 18.5%, the EPYC 9755 by 22.7%, the EPYC 9745 by 45.7%, and the Threadripper PRO 9995WX by 50.6%. The 9575F trades blows with its nearest rivals: it is 0.4% ahead of the EPYC 9734, 1.2% behind the Xeon 6781P, 2.8% behind the Threadripper PRO 9985WX, and 3.1% behind the Threadripper 9980X.

If you are building a system for bulk processing, virtualization, or scientific computation that scales, the 9965 is the answer. If you need the fastest possible single-thread performance in an EPYC package, the 9575F is the answer. There is no universal winner; the benchmark data points to two different tools for two different jobs.

FAQ

Q: Which processor has more cores and threads?

A: The AMD EPYC 9965 has 192 cores and 384 threads, while the AMD EPYC 9575F has 64 cores and 128 threads.

Q: Which processor has the higher boost clock?

A: The AMD EPYC 9575F has a boost clock of 5.00 GHz, compared to the AMD EPYC 9965's boost clock of 3.70 GHz.

Q: Which processor wins in single-thread performance?

A: The AMD EPYC 9575F wins the PassMark single-thread test with a score of 4,173 versus 3,176, a 23.9% advantage. However, the 9965 wins all Cinebench single-core tests by 7%.

Q: How large is the margin in floating-point math?

A: The AMD EPYC 9965 scores 1,153,453 in PassMark floating-point math, which is 119.3% higher than the 9575F's 526,003.

Q: Do both processors support the same memory bandwidth?

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

Q: What are the TDP values for each processor?

A: The AMD EPYC 9965 has a TDP of 500 W, while the AMD EPYC 9575F has a TDP of 400 W.