AMD EPYC 9565 vs AMD EPYC 9745 Comparison

The AMD EPYC 9745 and AMD EPYC 9565 sit at opposite ends of the same design philosophy within the EPYC 9005 series, and the recorded data shows a split that is unusual for two chips in the same family. The 9745 brings 128 cores at conservative clocks on a denser process node, while the 9565 trades core count for substantially higher frequencies. Both share a 400 W TDP, the same SP5 socket, the same twelve-channel DDR5 platform, and the same launch date of October 2024. Yet the benchmark record flips depending on the workload: the 9745 dominates throughput-oriented compute tests by margins as wide as 70.5%, while the 9565 wins the aggregate of head-to-head tests, 11 to 6, largely on the strength of single-threaded and frequency-sensitive results. Understanding which of those profiles matches a deployment is the entire decision here.

The Verdict

The data points to a clear division. Workloads that scale across many cores and hammer integer math, floating point, encryption, compression, or string sorting should go to the EPYC 9745. Its integer math score of 1,224,315 against 717,948 for the 9565, a 70.5% gap, is the single largest margin in the entire head-to-head set, and it is not an outlier: encryption leads by 61.7%, string sorting by 60.6%, compression by 52.3%, and floating point by 38.5%. These are server-style throughput tasks, and the 9745 wins them decisively despite having the lower average benchmark score overall (425,973 versus 285,471 for the 9565 is actually reversed: the 9745's average score is 425,973 and the 9565's is 285,471, which reflects the enormous scale of the 9745's compute-test wins).

For anything that depends on per-core speed, latency, or responsiveness, the 9565 is the correct pick. Its single-thread score of 3,696 beats the 9745's 2,806 by 24.1%, and it sweeps every Cinebench test, single- and multi-core alike, by roughly 3.3%. The 9565 also wins the head-to-head series count 11 tests to 6, which matters for buyers who weight general-purpose performance suites heavily. Both chips sit in the top percentile of the database, at the 100th and 99th percentile respectively, so neither is a compromise in absolute terms.

Architecture Differences

Both processors are Zen 5 parts codenamed Turin, launched on the same day in 2024, and both use TSMC fabrication with Socket SP5. The divergence begins with the silicon itself. The 9745 is built on a 3 nm process and uses Zen 5c cores, the compact variant, which is how it fits 128 cores and 256 threads. The 9565 uses standard Zen 5 cores on a 4 nm process, with 72 cores and 144 threads, and the database records its physical details: 99,780 million transistors across a 12-die configuration of 70.6 mm² per die. No transistor or die-size data is recorded for the 9745.

Clock behavior follows directly from those choices. The 9745 runs a 2.40 GHz base and 3.70 GHz boost; the 9565 runs 3.15 GHz base and 4.30 GHz boost. That frequency advantage is the engine behind every 9565 win in the record.

Cache is another split. Per-core L1 (80 KB) and L2 (1 MB) are identical, but shared L3 differs: the 9565 carries 384 MB, the 9745 carries 256 MB. The higher-clocking part also gets more L3 to feed fewer cores, which reinforces its latency profile.

Everything else matches. Both support DDR5 across twelve channels at 576.0 GB/s of bandwidth, both offer ECC, both expose 128 lanes of PCIe Gen 5 from the CPU, neither has integrated graphics, both are multiplier-locked, and both carry the same 400 W TDP and Active production status. Launch MSRP is recorded once here for reference: $12,141 for the 9745 and $10,486 for the 9565.

Head-to-Head Benchmarks

Walking through the Cinebench suite first, the 9565 wins everything. R15 multi-core: 11,585 to 11,198 (3.3%). R15 single-core: 1,635 to 1,580 (3.4%). R20 multi-core: 48,273 to 46,659 (3.3%). R20 single-core: 6,814 to 6,586 (3.3%). R23 multi-core: 114,937 to 111,093 (3.3%). R23 single-core: 16,226 to 15,683 (3.3%). This is a notable result: even in multi-core rendering, where the 9745 holds a 128-to-72 core advantage, the 9565's higher clocks carry it past. Rendering, in this dataset, is frequency-tolerant enough that raw per-core speed beats core count.

The PassMark compute suite tells the opposite story. Integer math: 1,224,315 for the 9745 versus 717,948, a 70.5% lead. Encryption: 229,447 versus 141,936, 61.7%. String sorting: 468,975 versus 291,941, 60.6%. Compression: 3,929,890 versus 2,579,631, 52.3%. Floating point: 761,219 versus 549,422, 38.5%. Extended instructions: 280,477 versus 209,595, 33.8%. Here the 9745's core density overwhelms the frequency gap completely. A 38.5% floating-point win on a chip with a lower boost clock is a direct measure of how much parallel capacity 128 compact cores add.

The remaining tests favor the 9565. PassMark multi-thread: 135,221 to 130,698 (3.3%). Physics: 18,036 to 17,122 (5.1%). Single-thread: 3,696 to 2,806 (24.1%). The one anomaly is find-prime-numbers, where the 9565 scores 2,422 against 979, a 59.6% gap in its favor, consistent with a test that leans on per-core throughput rather than aggregate parallelism.

Against the wider field, the 9745's average score of 425,973 places it 3.4% ahead of the AMD Ryzen Threadripper PRO 9995WX, 7.1% ahead of the EPYC 9655P, 12.3% ahead of the EPYC 9535, and 14.1% ahead of the EPYC 9655. The 9565's 285,471 average puts it in a tight cluster: 0.2% behind the Intel Xeon 696X, 0.6% behind the EPYC 9555P, 1.8% ahead of the Intel Xeon 6780E, and 2% ahead of the Ryzen Threadripper 9970X.

Specification Differences

Only a handful of fields differ between the two, and each difference explains the benchmark splits:

• Cores/threads: 128/256 (9745) versus 72/144 (9565)
• Base clock: 2.40 GHz versus 3.15 GHz
• Boost clock: 3.70 GHz versus 4.30 GHz
• Process node: 3 nm versus 4 nm
• Core variant: Zen 5c versus standard Zen 5
• L3 cache: 256 MB shared versus 384 MB shared
• Transistors: not recorded for the 9745; 99,780 million for the 9565
• Die size: not recorded for the 9745; 12x 70.6 mm² for the 9565
• Part number: 100-000001460 versus 100-000001447
• Launch MSRP: $12,141 versus $10,486

Identical fields include TDP (400 W), socket (SP5), architecture (Zen 5, Turin), memory (DDR5, twelve-channel, 576.0 GB/s, ECC), PCIe (Gen 5, 128 CPU lanes), and release date.

FAQ

Q: Which CPU wins more benchmark tests overall? A: The EPYC 9565, with 11 wins against 6 for the EPYC 9745 in the recorded head-to-head set, though the 9745's wins are far larger in magnitude.

Q: Which is faster in single-threaded work? A: The 9565, by a wide margin: 3,696 versus 2,806 in PassMark single-thread, a 24.1% gap, plus roughly 3.3% leads in Cinebench R20 and R23 single-core.

Q: Which is faster for integer math and encryption? A: The 9745, decisively: 70.5% ahead in integer math and 61.7% ahead in encryption.

Q: Do both use the same socket and platform? A: Yes. Both use Socket SP5, twelve-channel DDR5 at 576.0 GB/s with ECC, and 128 lanes of PCIe Gen 5, so platform compatibility is identical.

Q: Does the 9745's extra core count translate to better Cinebench scores? A: No. The 9565 wins every Cinebench test by about 3.3%, including the multi-core runs, because its higher boost clock (4.30 GHz versus 3.70 GHz) offsets the core deficit in these rendering workloads.

Q: How do they rank against the rest of the database? A: The 9745 sits in the 100th percentile; the 9565 sits in the 99th percentile.

Where Each One Wins

The EPYC 9745 wins in massively parallel throughput computing: integer math, floating point math, encryption, data compression, string sorting, and SIMD-style extended instruction workloads. Its 33.8% to 70.5% leads in these tests make it the clear choice for dense multi-tenant compute where aggregate capacity matters more than any individual thread. It also ranks 3.4% above the Threadripper PRO 9995WX in average score, the strongest rival in its recorded set.

The EPYC 9565 wins wherever frequency, latency, or per-thread performance dominates: single-threaded applications, physics simulation, prime-number computation, rendering across Cinebench R15, R20, and R23 in both single- and multi-core modes, and general multi-threaded suites where its higher clocks keep all 144 threads moving faster. Its 24.1% single-thread advantage is the largest frequency-driven gap in the record, and its tighter, lower-core design pairs with more L3 cache per core, which suits latency-sensitive server duties. Both deliver identical memory bandwidth, PCIe connectivity, and TDP, so the platform decision is neutral: choose the 9745 for scale, the 9565 for speed.