AMD EPYC 9575F vs AMD EPYC 9745 Comparison

The Verdict

The benchmark data splits these two AMD EPYC 9005 series processors into clearly different roles. The AMD EPYC 9575F is the outright performance leader in the majority of recorded tests, taking 11 of 17 head-to-head benchmarks, including every Cinebench iteration and the PassMark multithread, physics, prime number, and single-thread tests. The AMD EPYC 9745 wins 6 benchmarks, all of which are data-heavy PassMark workloads: compression, encryption, extended instructions, floating point math, integer math, and random string sorting. For workloads that rely on raw core count and parallel data processing, the 9745 is the pick. For workloads that benefit from higher clock speeds and stronger per-thread execution, the 9575F dominates.

The 9575F leads by 13% in all six Cinebench tests, a consistent margin across both single-core and multi-core versions. It also posts a 32.8% advantage in PassMark single-thread scores. The 9745 counters with a 44.7% lead in floating point math and a 41.7% lead in data compression. The choice comes down to workload type: the 9575F for latency-sensitive, frequency-driven tasks, the 9745 for throughput-oriented batch processing. The 9745 sits at the 100th percentile of all CPUs in the database, while the 9575F sits at the 99th percentile, so both are elite parts, but they are elite in different ways.

FAQ

Q: Which processor has more cores and threads?

A: The AMD EPYC 9745 has 128 cores and 256 threads. The AMD EPYC 9575F has 64 cores and 128 threads. The 9745 doubles the core and thread counts of the 9575F.

Q: Why does the 9575F win the Cinebench multi-core tests despite having half the cores?

A: The 9575F runs at a 3.30 GHz base clock and 5.00 GHz boost clock, while the 9745 runs at 2.40 GHz base and 3.70 GHz boost. The 9575F's higher clocks allow each of its 64 cores to complete work faster, and in the Cinebench workloads the clock advantage overcomes the core deficit. The 9575F scores 127739 in Cinebench R23 multi-core versus 111093 for the 9745, a 13% win.

Q: What is the process node difference between the two chips?

A: The 9745 is built on a 3 nm process node, while the 9575F uses a 4 nm node. Both are fabricated by TSMC. The 9575F has a transistor count of 66,520 million and a die size of 8x 70.6 mm²; the 9745 does not have recorded transistor or die size data in the database.

Q: Do both processors support the same memory and PCIe configurations?

A: Yes. Both support DDR5 memory with a twelve-channel memory bus and 576.0 GB/s memory bandwidth. Both have ECC memory support. Both provide PCIe Gen 5 with 128 lanes (CPU only). Neither has integrated graphics.

Q: Which processor has the higher average benchmark score?

A: The 9745 has a higher average benchmark score of 425973, compared to 311774 for the 9575F. The 9745's nearest rival is the AMD Ryzen Threadripper PRO 9995WX at 412068, a 3.4% delta. The 9575F's nearest rival is the AMD EPYC 9734 at 310619, a 0.4% delta.

Q: What are the launch MSRP values?

A: The 9745 has a launch MSRP of $12141. The 9575F has a launch MSRP of $11791.

Architecture Differences

Both processors belong to the EPYC 9005 series and share the Turin codename, but they use different Zen 5 variants. The 9745 is built on Zen 5c architecture, while the 9575F uses full Zen 5 architecture. This distinction is central to understanding their behavior. Zen 5c is a denser implementation designed for higher core counts at lower clock speeds, which explains the 9745's 128 cores. The 9575F's Zen 5 design prioritizes clock speed and per-core performance, which explains its 64-core layout with much higher frequencies.

The process nodes differ as well. The 9745 uses a 3 nm process from TSMC, while the 9575F uses a 4 nm process from the same foundry. The smaller node on the 9745 allows more transistors to fit in a given area, enabling the higher core count. The 9575F compensates with a larger node but higher clocks. The database records the 9575F's transistor count at 66,520 million and its die size at 8x 70.6 mm²; no equivalent figures are recorded for the 9745.

Cache configurations are identical across both parts. Each has 80 KB of L1 cache per core, 1 MB of L2 cache per core, and 256 MB of shared L3 cache. Neither has 3D V-Cache. The identical cache hierarchy means the performance differences in benchmarks come from core count, clock speed, and the architectural variant, not from cache capacity.

Both processors use the AMD Socket SP5, have a 400 TDP, and are not multiplier unlocked. Both are active production parts released on the same date. The 9745 has part number 100-000001460, while the 9575F has part number 100-000001554. Both target the Server/Workstation market segment.

Specification Differences

The core and thread counts are the most obvious difference. The 9745 offers 128 cores and 256 threads, exactly double the 9575F's 64 cores and 128 threads. Clock speeds move in the opposite direction. The 9575F has a base clock of 3.30 GHz and a boost clock of 5.00 GHz, while the 9745 has a base clock of 2.40 GHz and a boost clock of 3.70 GHz. The 9575F runs 0.90 GHz higher at base and 1.30 GHz higher at boost.

The process node differs: 3 nm for the 9745 versus 4 nm for the 9575F. The architecture differs: Zen 5c for the 9745 versus Zen 5 for the 9575F. The 9575F has recorded transistor and die size data (66,520 million transistors, 8x 70.6 mm² die), while the 9745 does not have these fields populated in the database.

Memory support, memory bus width, memory bandwidth, ECC support, PCIe lanes, socket, TDP, integrated graphics, market segment, production status, release date, and multiplier unlock status are all identical between the two. The launch MSRP differs slightly: $12141 for the 9745, $11791 for the 9575F. The part numbers differ as noted above.

The average benchmark score differs substantially. The 9745 averages 425973, which is 36.6% higher than the 9575F's 311774. The percentile rankings also differ: the 9745 is at the 100th percentile of all CPUs, while the 9575F is at the 99th percentile.

Head-to-Head Benchmarks

The Cinebench results are uniform. Across R15, R20, and R23, in both single-core and multi-core variants, the 9575F wins by exactly 13% in every case. Cinebench R15 multi-core shows 12876 for the 9575F against 11198 for the 9745. R15 single-core shows 1817 versus 1580. R20 multi-core shows 53650 versus 46659. R20 single-core shows 7573 versus 6586. R23 multi-core shows 127739 versus 111093. R23 single-core shows 18033 versus 15683. This consistent 13% delta across all six Cinebench tests indicates a systematic advantage for the 9575F in that benchmark family, likely driven by its higher clock speeds.

The PassMark suite tells a different story. The 9745 wins six of the eleven PassMark tests, and its margins are larger than the 9575F's Cinebench margins. The biggest win for the 9745 is floating point math at 761219 versus 526003, a 44.7% advantage. Data compression follows at 3929890 versus 2773634, a 41.7% lead. Extended instructions show a 42% advantage at 280477 versus 197484. Data encryption is close behind at 40.9%, with 229447 versus 162818. Integer math shows a 37.3% lead at 1224315 versus 891817. Random string sorting completes the set at 34.6%, with 468975 versus 348500.

The 9575F wins the remaining PassMark tests. Single-thread performance shows its largest margin: 4173 versus 2806, a 32.8% advantage. This is the biggest single-test win for either chip. Prime numbers show a 19.4% lead at 1215 versus 979. Physics shows a 17.1% lead at 20652 versus 17122. The multithread test shows an 11.7% lead at 147998 versus 130698.

The win count favors the 9575F at 11 wins against 6 for the 9745. However, the magnitude of the 9745's wins in the PassMark data workloads is substantial. The 9745's wins range from 34.6% to 44.7%, while the 9575F's wins range from 11.7% to 32.8%. The 9575F wins more tests, but the 9745 wins its tests by larger margins.

Where Each One Wins

The 9575F wins in every Cinebench test, making it the stronger choice for rendering workloads that use Cinebench-style multi-threaded rendering with significant per-thread instruction-level parallelism. Its 13% lead across all Cinebench versions is uniform, which suggests a predictable advantage. The 9575F also wins PassMark multithread, physics, prime numbers, and single-thread tests. The single-thread win at 32.8% is particularly notable. Workloads that are latency-sensitive, require fast response times per thread, or depend on high boost clocks will favor the 9575F. Its 5.00 GHz boost clock is the highest in this comparison and drives all of its single-thread and Cinebench victories.

The 9745 wins in data compression, data encryption, extended instructions, floating point math, integer math, and random string sorting. These are throughput-oriented workloads where the 128-core count can be fully utilized. The 44.7% lead in floating point math is the largest margin in the entire comparison, and the 41.7% lead in data compression is nearly as large. The 9745's wins all exceed 34%, meaning that in its winning workloads, it is not just marginally better but substantially ahead. The 9745's 3 nm process and Zen 5c architecture allow it to pack twice the cores into the same TDP envelope, and these PassMark data workloads scale well with core count.

The average benchmark score also favors the 9745 at 425973 versus 311774, a 36.6% gap. The 9745's nearest rivals include the AMD Ryzen Threadripper PRO 9995WX at 3.4% behind and the AMD EPYC 9655P at 7.1% behind. The 9575F's nearest rivals are much closer: the AMD EPYC 9734 is only 0.4% behind, and the Intel Xeon 6781P is 1.2% ahead. This suggests the 9575F sits in a more contested performance tier, while the 9745 sits at the top of the database at the 100th percentile.

For a server handling database encryption, file compression, scientific floating point calculations, or integer-heavy analytics, the 9745's 44.7% floating point and 37.3% integer math leads make it the clear choice. For a workstation running interactive rendering, physics simulations, or single-threaded legacy applications, the 9575F's 13% Cinebench leads and 32.8% single-thread lead make it the better fit. The data does not show one processor as universally superior; it shows two differently optimized designs for different segments of the server and workstation market.