AMD EPYC 9565 vs Intel Xeon 6960P Comparison

The Intel Xeon 6960P and AMD EPYC 9565 are two server flagships that meet in an unusual spot: identical core counts of 72 cores and 144 threads, both on DDR5 platforms with twelve-channel memory buses, both launched within weeks of each other. Yet the recorded data shows they are not the same machine. The EPYC 9565 wins more benchmarks outright (9 wins to 5), but the Xeon 6960P posts several of the largest margins on the board, including a 38.3 percent win in PassMark Physics and a 27.4 percent win in Random String Sorting. This is a matchup where the workload, not the overall win count, decides the better buy.

Head-to-Head Benchmarks

The rendering suite belongs to AMD, and by a consistent margin. The EPYC 9565 takes Cinebench R15 Multi with 11585 points to 11194, Cinebench R20 Multi with 48273 to 46645, and Cinebench R23 Multi with 114937 to 111060. That is a 3.4 percent advantage in all three tests, a remarkably stable gap that suggests a genuine throughput edge in heavily threaded rendering rather than run-to-run noise. PassMark Multithread follows the same pattern: 135221 for the EPYC against 130659 for the Xeon, again a 3.4 percent AMD win.

Single-threaded performance tilts further toward AMD. The EPYC 9565 scores 3696 in PassMark Single Thread against 3287 for the Xeon 6960P, an 11.1 percent gap. The EPYC also records Cinebench single-core results of 1635 in R15, 6814 in R20, and 16226 in R23, tests where the database holds no comparable Xeon entries. That single-thread lead tracks directly with the clock specifications: the EPYC boosts to 4.30 GHz against 3.90 GHz for the Xeon, from a higher 3.15 GHz base versus 2.70 GHz.

The Intel side of the ledger is where the big swings live. PassMark Physics is the largest margin in the entire dataset: 24937 for the Xeon 6960P against 18036 for the EPYC, a 38.3 percent Intel victory. Random String Sorting is nearly as dramatic, 371795 to 291941, a 27.4 percent Intel win. Data Encryption goes to Intel 162013 to 141936 (14.1 percent), and Data Compression goes to Intel 2797724 to 2579631 (8.5 percent). Integer Math is a narrow Intel win, 727750 to 717948, just 1.4 percent apart.

AMD takes the remaining tests. Extended Instructions favors the EPYC 209595 to 193404 (7.7 percent), Floating Point Math favors it 549422 to 527473 (4 percent), and Find Prime Numbers is a blowout: 2422 for the EPYC against 1484 for the Xeon, a 38.7 percent AMD margin, the single biggest spread in the dataset.

Context from the wider database sharpens the picture. The Xeon 6960P averages 365194 across its benchmark suite, sitting in the 100th percentile of all recorded CPUs, and rates within 0.2 percent of the AMD EPYC 9754 (364371) while trailing the EPYC 9655 by 2.2 percent and the EPYC 9535 by 3.7 percent, and leading the EPYC 9475F by 4.1 percent. The EPYC 9565 averages 285471 in the 99th percentile, closest to the Intel Xeon 696X (286102, within 0.2 percent), the EPYC 9555P (287066, within 0.6 percent), the Xeon 6780E (1.8 percent ahead of it), and the Threadripper 9970X (2 percent ahead of it). Note the two averages come from different benchmark mixes, so treat the head-to-head results above as the real comparison.

FAQ

Q: Which CPU is faster in multi-threaded rendering?

A: The AMD EPYC 9565, by a consistent 3.4 percent across Cinebench R15, R20, and R23 Multi, plus a matching 3.4 percent win in PassMark Multithread (135221 vs 130659).

Q: Which one is better for single-threaded work?

A: The EPYC 9565. It scores 3696 in PassMark Single Thread versus 3287 for the Xeon 6960P, an 11.1 percent lead, consistent with its higher 3.15 GHz base and 4.30 GHz boost clocks.

Q: Where does the Xeon 6960P win big?

A: PassMark Physics (24937 vs 18036, 38.3 percent), Random String Sorting (371795 vs 291941, 27.4 percent), Data Encryption (162013 vs 141936, 14.1 percent), Data Compression (2797724 vs 2579631, 8.5 percent), and Integer Math by 1.4 percent.

Q: Do they support the same memory?

A: Both use DDR5 with ECC on twelve-channel buses, but bandwidth differs: 614.4 GB/s for the Xeon versus 576.0 GB/s for the EPYC.

Q: What about power draw and PCIe?

A: The Xeon 6960P is rated at 500 W TDP with 96 Gen 5 lanes; the EPYC 9565 is rated at 400 W TDP with 128 Gen 5 lanes. The EPYC offers more lanes at a lower rated TDP.

Q: How do their launch MSRPs compare?

A: The Xeon 6960P carries a launch MSRP of $9625 and the EPYC 9565 a launch MSRP of $10486.

Where Each One Wins

Pick the Xeon 6960P when memory bandwidth and integer-heavy throughput dominate. Its 614.4 GB/s of memory bandwidth is the highest figure in this pairing, and its wins line up with workloads that chew through streams and integer operations: compression, encryption, string sorting, physics simulation, and general integer math. If your stack spends its day compressing data, encrypting traffic, or sorting, the recorded gaps of 8.5 to 38.3 percent are too large to ignore, and no clock deficit offsets them.

Pick the EPYC 9565 for anything latency-sensitive, per-core licensed, or floating-point heavy. The 11.1 percent single-thread win matters for applications that cannot use 144 threads, and the Find Prime Numbers result (38.7 percent ahead) plus wins in Extended Instructions and Floating Point Math point to strong SIMD and arithmetic throughput. Rendering farms should also land on the EPYC given the sweep of every Cinebench multi-core test. Its 400 W TDP rating versus 500 W also helps in density-limited racks, and its 128 PCIe lanes accommodate more devices than the Xeon's 96.

Specification Differences

The two chips differ in nearly every measurable specification despite matching core and thread counts. Base clock is 3.15 GHz on the EPYC versus 2.70 GHz on the Xeon; boost is 4.30 GHz versus 3.90 GHz. TDP is 400 W for the EPYC against 500 W for the Xeon. The sockets are incompatible: AMD Socket SP5 versus Intel Socket 7529. Memory bandwidth is 576.0 GB/s versus 614.4 GB/s, both on twelve-channel DDR5 with ECC. PCIe is 128 Gen 5 lanes for the EPYC versus 96 Gen 5 lanes for the Xeon. Launch MSRP is $10486 for the EPYC and $9625 for the Xeon. Release dates are close: the Xeon on September 23, 2024, and the EPYC on October 9, 2024. Both are active server/workstation parts with no integrated graphics and locked multipliers.

Architecture Differences

These are fundamentally different designs. The Xeon 6960P is Intel's Granite Rapids generation, fabricated on Intel's 5 nm process, and built from three large dies of 598 mm² each. Cache is generous: 112 KB of L1 per core, 2 MB of L2 per core, and 432 MB of shared L3. That larger L3 pool and higher memory bandwidth pair naturally with its wins in bandwidth-hungry PassMark workloads.

The EPYC 9565 is AMD's Zen 5 architecture under the Turin codename, part of the EPYC 9005 series, built on TSMC's 4 nm process with a chiplet layout of twelve small dies at 70.6 mm² each and 99,780 million transistors. Per-core cache is smaller, 80 KB of L1 and 1 MB of L2, with 384 MB of shared L3. The newer node and higher clocks explain its single-thread edge, while the chiplet approach delivers the extra PCIe lanes at the lower TDP rating.

The Verdict

The data splits cleanly by workload. If the job list is compression, encryption, sorting, physics, or integer math, the Xeon 6960P is the correct choice: it wins all five of those tests, four of them by double digits, and brings 614.4 GB/s of memory bandwidth to feed them. If the job list is rendering, floating point, vectorized code, or anything that depends on per-core speed, the EPYC 9565 wins every relevant test, leads single-thread performance by 11.1 percent, and does it at a lower 400 W TDP rating with 128 PCIe lanes.

For a general-purpose server where the workload mix is uncertain, the EPYC 9565's broader win count (9 to 5) and its sweep of every multi-core rendering test make it the safer default. The Xeon 6960P is the specialist pick, and in its specialties the margins are enormous. Choose based on what the servers actually run, because in this pairing the benchmarks rarely compromise.