AMD EPYC 9845 vs Intel Xeon 6960P Comparison

The AMD EPYC 9845 and Intel Xeon 6960P represent two distinct approaches to extreme server computing. The data shows a clear split: the EPYC 9845 dominates in nearly every multithreaded and throughput-oriented task, while the Xeon 6960P counters with wins in specific single-thread and physics workloads. The head-to-head benchmark table records 10 wins for the AMD part and 4 for the Intel, but the margins tell a more nuanced story than the raw tally suggests.

Where Each One Wins

The AMD EPYC 9845 is the undisputed champion of parallel compute. Its victories span the entire Cinebench suite, with a consistent 17.1% lead over the Intel Xeon 6960P in R15, R20, and R23 multicore tests. The PassMark data reinforces this pattern: integer math shows the largest gap at 131.9%, floating-point math follows at 85.5%, and data encryption comes in at 83.2%. For workloads that scale with core count and thread count, the EPYC 9845's 160 cores versus the Xeon's 72 cores creates an insurmountable advantage. Data compression, extended instructions, and random string sorting also fall firmly in AMD's column, with leads of 67.3%, 62.8%, and 44.7% respectively.

The Intel Xeon 6960P claims a different territory: latency-sensitive and single-threaded tasks. It wins PassMark single-thread by 4.4%, scoring 3287 versus 3144. The physics test shows its strongest result, a 21.3% advantage (24937 versus 19631). Prime number finding also favors Intel, with a 15.4% edge (1484 versus 1255). These wins suggest the Xeon's higher base clock of 2.70 GHz and boost clock of 3.90 GHz, combined with its larger per-core L1 cache of 112 KB, deliver tangible benefits for workloads that cannot parallelize effectively. The Xeon also posts a higher PassMark multithread score in relative terms, but at 130659 versus 152985, it still trails the AMD part by 17.1%.

FAQ

Q: Which processor has more cores and threads?

A: The AMD EPYC 9845 has 160 cores and 320 threads. The Intel Xeon 6960P has 72 cores and 144 threads, which is less than half the core count of the AMD part.

Q: Does the Intel Xeon 6960P win any benchmark categories?

A: Yes, the Xeon wins 4 of the 14 head-to-head benchmarks: PassMark single-thread, PassMark singlethread, PassMark physics, and PassMark find prime numbers. Its largest victory is in physics with a 21.3% lead.

Q: How large is the AMD EPYC 9845's lead in Cinebench multicore tests?

A: The AMD part leads by exactly 17.1% in all three Cinebench multicore tests: R15 (13107 vs 11194), R20 (54615 vs 46645), and R23 (130037 vs 111060).

Q: What is the biggest single benchmark margin between the two?

A: The largest margin is in PassMark integer math, where the AMD EPYC 9845 scores 1687531 versus the Xeon's 727750, a delta of 131.9%.

Q: How do the two processors compare in memory bandwidth?

A: The Intel Xeon 6960P has a higher memory bandwidth at 614.4 GB/s, while the AMD EPYC 9845 offers 576.0 GB/s. Both use twelve-channel DDR5 memory.

Q: What are the process nodes for each chip?

A: The AMD EPYC 9845 is built on a 3 nm process by TSMC. The Intel Xeon 6960P uses a 5 nm process from Intel's own foundry.

Head-to-Head Benchmarks

The Cinebench results establish a baseline pattern. Across R15, R20, and R23 multicore, the AMD EPYC 9845 scores 13107, 54615, and 130037 respectively, versus the Intel Xeon 6960P's 11194, 46645, and 111060. The consistent 17.1% delta across all three generations of Cinebench indicates a stable architectural advantage rather than a workload-specific anomaly. This uniformity suggests the EPYC 9845's core count advantage translates linearly into rendering performance.

PassMark integer math delivers the most dramatic divergence. The AMD part's 1687531 score dwarfs the Xeon's 727750, representing a 131.9% advantage. This result likely reflects the EPYC 9845's 320 threads engaging fully in arithmetic operations, while the Xeon's 144 threads cannot compensate for the core deficit. Floating-point math tells a similar story: 978377 versus 527473, an 85.5% lead for AMD. Data encryption shows an 83.2% margin (296808 vs 162013), and extended instructions follow at 62.8% (314798 vs 193404).

The Intel Xeon 6960P's wins are narrower but meaningful. In PassMark physics, it scores 24937 against the AMD's 19631, a 21.3% advantage. This benchmark often responds to clock speed and per-core efficiency, where the Xeon's 3.90 GHz boost clock and 2 MB per-core L2 cache provide an edge. Prime number finding shows a 15.4% lead (1484 vs 1255), and single-thread performance comes in at 3287 versus 3144, a 4.4% margin. Notably, the Xeon's single-thread score of 3287 is higher than the AMD part's 3144, but the absolute numbers are close enough that the EPYC 9845 remains competitive even in Intel's strongest category.

Specification Differences

The core and thread disparity defines the matchup: 160 cores and 320 threads for AMD versus 72 cores and 144 threads for Intel. Clock speeds favor the Xeon, with a base clock of 2.70 GHz and boost clock of 3.90 GHz against the EPYC 9845's 2.10 GHz base and 3.70 GHz boost. Thermal design power also differs substantially, with the Intel part rated at 500 TDP and the AMD part at 390 TDP. The sockets are incompatible: AMD uses Socket SP5 while Intel uses Socket 7529. Process nodes differ as well, with AMD on a 3 nm process from TSMC and Intel on a 5 nm process from its own foundry. The Intel die size is listed as 3x 598 mm², while no die size is provided for the AMD part.

Memory bandwidth favors Intel slightly at 614.4 GB/s versus 576.0 GB/s for AMD, though both use twelve-channel DDR5. PCIe lane counts differ: AMD offers 128 Gen 5 lanes while Intel provides 96 Gen 5 lanes. Cache hierarchies diverge significantly: AMD has 80 KB L1 per core, 1 MB L2 per core, and 320 MB shared L3, while Intel has 112 KB L1 per core, 2 MB L2 per core, and 432 MB shared L3. The launch MSRP for the AMD EPYC 9845 is $13564, while the Intel Xeon 6960P has a launch MSRP of $9625.

Architecture Differences

The AMD EPYC 9845 belongs to the EPYC 9005 series based on Zen 5 architecture, codenamed Turin, and manufactured on TSMC's 3 nm node. The Intel Xeon 6960P uses Granite Rapids architecture from the Xeon 6 generation, built on Intel's 5 nm process with a die size of 3x 598 mm². These architectural choices explain the performance patterns: AMD's smaller 3 nm node allows for 160 cores within a 390 TDP envelope, while Intel's larger 5 nm process with 72 cores still consumes 500 TDP.

The cache designs reflect different philosophies. AMD allocates 80 KB L1 and 1 MB L2 per core, with a massive 320 MB shared L3. Intel provides more per-core cache at 112 KB L1 and 2 MB L2, plus a larger 432 MB shared L3. The Intel approach of larger per-core caches likely contributes to its single-thread and physics benchmark wins, while AMD's greater core count and shared L3 benefit throughput workloads. Both processors support DDR5 memory with ECC, and neither includes integrated graphics. Both are active production parts targeting the server/workstation segment, with AMD releasing on 2024-10-09 and Intel on 2024-09-23.

The Verdict

The data points to a clear conclusion for most server workloads: the AMD EPYC 9845 is the superior choice when maximum parallel throughput matters. Its 17.1% lead across all Cinebench multicore tests, combined with margins exceeding 60% in integer math, floating-point math, encryption, and compression, makes it the default option for rendering, scientific computing, data processing, and virtualization environments. The 100th percentile ranking among all CPUs, with an average benchmark score of 523613, places it at the top of the performance hierarchy. Its nearest rival, the AMD EPYC 9965, scores 620487 on average, which is 15.6% higher, indicating headroom above even this flagship within the same family.

The Intel Xeon 6960P serves a narrower but valid purpose. Its wins in physics, single-thread, and prime number tasks suggest suitability for workloads with heavy sequential components or those sensitive to clock speed. The 4.4% single-thread advantage and 21.3% physics lead are not trivial, and the higher memory bandwidth of 614.4 GB/s could benefit memory-bound applications. However, its average benchmark score of 365194 places it far behind the AMD part, and its nearest rivals are all AMD EPYC parts within 4.1%—the EPYC 9754, EPYC 9655, EPYC 9535, and EPYC 9475F. The Xeon's role is best understood as a specialist for latency-critical tasks rather than a general-purpose throughput champion. For organizations running mixed workloads where the majority of threads scale well, the EPYC 9845's 10 out of 14 head-to-head wins, including every Cinebench multicore test, make it the data-driven recommendation.