AMD EPYC 9684X vs Intel Xeon 6960P Comparison

Head-to-Head Benchmarks

The Intel Xeon 6960P takes a decisive lead in the majority of head-to-head comparisons, winning 11 of the 14 recorded benchmarks against the AMD EPYC 9684X. The most striking margin comes in the PassMark single-thread test, where the Intel part scores 3287 against the AMD's 2891, a 13.7% advantage. This is a significant gap for single-threaded performance and suggests the Intel architecture holds a clear IPC advantage in lightly threaded workloads.

The Cinebench suite reinforces this pattern. Across all three versions, the Xeon 6960P wins by nearly identical margins: 7.4% in Cinebench R15 multicore (11194 vs 10418), 7.5% in R20 multicore (46645 vs 43409), and 7.5% in R23 multicore (111060 vs 103355). These consistent deltas indicate that the Intel chip scales its advantage evenly across both older and newer rendering workloads.

Floating-point math is another strong area for Intel. The Xeon 6960P posts 527473 in PassMark floating-point math versus the EPYC 9684X's 472174, an 11.7% win. Extended instruction workloads also favor Intel, with a 193404 score against 177956, an 8.7% edge. The PassMark multithread test follows the same trend: 130659 for Intel versus 121595 for AMD, a 7.5% victory. Even the physics test, often a mixed bag, goes to Intel by a slim 1% margin (24937 vs 24686).

Data compression is closer but still favors Intel, with the Xeon 6960P scoring 2797724 against 2698807, a 3.7% edge. Random string sorting also goes to Intel by 6.5% (371795 vs 349126).

The AMD EPYC 9684X takes only three wins, but they are notable. Integer math is a clear AMD strength: 844145 versus 727750, a 13.8% margin for the EPYC. Prime number finding is an even larger win, with AMD scoring 2020 versus Intel's 1484, a 26.5% advantage that is the single biggest delta in either direction across the entire benchmark set. Data encryption also favors AMD, with a 178453 score against Intel's 162013, a 9.2% edge.

Interestingly, the AMD part's wins map to workloads that leverage its massive L3 cache and higher core count. The Intel part's wins are more numerous and cover a broader range of tasks, from rendering to floating-point math to single-threaded responsiveness. The data suggests that for most general-purpose server workloads, the Xeon 6960P is the more balanced performer, while the EPYC 9684X has specific niches where it excels.

Where Each One Wins

The Intel Xeon 6960P is the clear choice for rendering and content creation workloads. Its Cinebench wins across R15, R20, and R23, all with margins around 7.5%, indicate that multi-core rendering pipelines will finish noticeably faster on the Intel platform. The 11.7% advantage in floating-point math further cements this position, as rendering engines and scientific simulations rely heavily on FPU throughput.

Single-threaded performance is another Intel stronghold. The 13.7% lead in the PassMark single-thread test means that database queries, web serving, and other latency-sensitive tasks that cannot easily parallelize will respond more quickly on the Xeon 6960P. The extended instructions win (8.7%) also suggests better throughput for vectorized and SIMD-heavy code.

The AMD EPYC 9684X wins where its architecture is specifically optimized. The 26.5% margin in prime number finding points to workloads that are heavily dependent on integer operations and cache residency. Integer math, with a 13.8% edge, reinforces this: financial modeling, cryptography, and certain database operations that are integer-heavy will favor the AMD part. The 9.2% win in data encryption is another clear niche, likely benefiting from the larger L3 cache that can hold more encryption keys and lookup tables.

The compression test is nearly a wash, with Intel winning by just 3.7%. Similarly, the physics test is effectively a tie at a 1% delta. For mixed workloads that combine compression, physics simulation, and general compute, neither chip has a decisive edge, and the choice would come down to other factors like platform integration or power delivery design.

In summary, the Xeon 6960P is the better all-rounder, suited for high-throughput compute, rendering, and single-threaded response. The EPYC 9684X is the specialist, dominating integer-heavy, cache-sensitive, and encryption workloads. A workload that is predominantly integer math or cryptography should favor AMD; everything else leans toward Intel based on the recorded data.

FAQ

Q: Which processor has a higher average benchmark score?

A: The Intel Xeon 6960P records an average benchmark score of 365194, while the AMD EPYC 9684X averages 266914. However, the AMD part's percentile ranking is 99 versus the Intel's 100, so both sit at the very top of the database.

Q: What is the largest performance gap between the two in any single test?

A: The biggest delta is in PassMark find prime numbers, where the AMD EPYC 9684X outperforms the Intel Xeon 6960P by 26.5%, scoring 2020 versus 1484.

Q: How do the two chips compare in single-threaded performance?

A: The Intel Xeon 6960P leads by 13.7% in the PassMark single-thread test, scoring 3287 against the AMD's 2891. This is one of the largest margins in favor of Intel across all benchmarks.

Q: Does the AMD EPYC 9684X win any rendering benchmarks?

A: No. The Intel Xeon 6960P wins all three Cinebench multicore tests: R15 by 7.4%, R20 by 7.5%, and R23 by 7.5%.

Q: Which processor has a higher core count?

A: The AMD EPYC 9684X has 96 cores and 192 threads, while the Intel Xeon 6960P has 72 cores and 144 threads.

Q: What are the nearest rivals to each chip according to the database?

A: For the Intel Xeon 6960P, the closest competitors are the AMD EPYC 9754 (0.2% behind), AMD EPYC 9655 (2.2% ahead), AMD EPYC 9535 (3.7% ahead), and AMD EPYC 9475F (4.1% behind). For the AMD EPYC 9684X, the nearest rivals are the AMD Ryzen Threadripper 9970X (4.6% ahead), Intel Xeon 6780E (4.8% ahead), Intel Xeon 6980P (6.1% behind), and AMD EPYC 9565 (6.5% ahead).

Specification Differences

The two processors differ in almost every core specification. The Intel Xeon 6960P has 72 cores and 144 threads, while the AMD EPYC 9684X has 96 cores and 192 threads, a 24-core advantage for AMD. Base clocks are similar, with Intel at 2.70 GHz and AMD at 2.55 GHz, but the Intel part boosts higher at 3.90 GHz versus AMD's 3.70 GHz.

Thermal design power is a major differentiator: the Xeon 6960P is rated at 500 watts, while the EPYC 9684X draws 400 watts. The sockets are incompatible, with Intel using Socket 7529 and AMD using Socket SP5. The Intel part is built on the Granite Rapids architecture, while the AMD uses Zen 4 (Genoa-X).

Cache configurations are dramatically different. The Intel Xeon 6960P has 112 KB of L1 cache per core and 2 MB of L2 per core, with a shared 432 MB L3. The AMD EPYC 9684X has 64 KB L1 per core, 1 MB L2 per core, and a massive 1152 MB shared L3, which is more than double the Intel's L3 capacity. The die size also differs: Intel uses three dies totaling 598 mm² each (3x 598 mm²), while AMD uses twelve smaller dies at 72 mm² each (12x 72 mm²). Transistor count is listed only for AMD at 135,240 million.

Memory bandwidth is another split: Intel supports 614.4 GB/s, while AMD supports 460.8 GB/s, both over twelve-channel DDR5. PCIe lanes differ, with Intel offering Gen 5, 96 lanes (CPU only) and AMD offering Gen 5, 128 lanes (CPU only). The launch MSRP for the Intel Xeon 6960P is $9625, and for the AMD EPYC 9684X it is $14756.

Architecture Differences

The Intel Xeon 6960P is built on the Granite Rapids architecture, which is Intel's Xeon 6 generation (Granite Rapids-AP). It uses a 5 nm process node fabricated by Intel itself. The chip is physically composed of three dies, each 598 mm², which is a large multi-die design. The Intel architecture emphasizes high clock speeds and a balanced cache hierarchy, with a 432 MB shared L3 that is substantial but far smaller than AMD's offering.

The AMD EPYC 9684X is part of the EPYC 9004 series, using the Zen 4 architecture with the Genoa-X codename. It is also on a 5 nm process, but fabricated by TSMC. The physical layout is twelve chiplets, each 72 mm², which is a chiplet-based design that allows AMD to scale core counts and cache more flexibly. The standout architectural feature is the 1152 MB shared L3 cache, enabled by 3D V-Cache technology, which explains the EPYC's dominance in cache-sensitive workloads like prime number finding and integer math.

Both support DDR5 memory with ECC, and both use twelve-channel memory buses. However, Intel's memory bandwidth is 614.4 GB/s versus AMD's 460.8 GB/s, giving Intel a 33% bandwidth advantage that likely contributes to its wins in floating-point and compression tasks.

The PCIe configurations differ as well: Intel provides 96 Gen 5 lanes, while AMD provides 128 Gen 5 lanes, giving AMD more headroom for high-speed I/O expansion. Neither chip has integrated graphics. The Intel part has a part number of SRPKX, while the AMD part is 100-100000892. Both are currently in active production, with the Intel released on 2024-09-23 and the AMD released earlier on 2023-06-12. Neither has an unlocked multiplier.