AMD EPYC 9655P vs Intel Xeon 6781P Comparison

Where Each One Wins

The benchmark data splits decisively, with the AMD EPYC 9655P taking 13 of 14 head-to-head tests and the Intel Xeon 6781P claiming a single, razor-thin victory. That lone Intel win comes in `passmark_find_prime_numbers`, where the Xeon scores 1687 against 1686 for the EPYC — a 0.1% margin that is statistically negligible but technically a win. Every other workload category favors AMD, often by substantial margins.

The EPYC 9655P dominates in multi-threaded throughput. In Cinebench R15, R20, and R23 multicore tests, it leads by exactly 36% in each case, posting scores of 13744, 57268, and 136354 respectively against Intel’s 10105, 42106, and 100254. This consistency across three Cinebench versions suggests a fundamental performance advantage in heavily parallel rendering workloads rather than a test-specific anomaly.

The AMD processor also wins decisively in memory and data-intensive operations. `passmark_data_compression` shows a 42.8% lead (3486158 vs 2441690), while `passmark_random_string_sorting` favors AMD by 68.2% (451824 vs 268573). The largest single delta appears in `passmark_integer_math`, where the EPYC’s 1225251 score more than doubles Intel’s 584834 — a 109.5% advantage. This is not a marginal gap; it is a category annihilation.

Single-thread performance also favors AMD, with the EPYC scoring 3849 in both `passmark_single_thread` and `passmark_singlethread` versus 3152 for Intel, a 22.1% lead. The Cinebench R15 single-core score of 1940 for AMD (Intel has no corresponding entry in the head-to-head data) reinforces this trend, though the absence of an Intel R15 single-core score limits direct comparison. The EPYC also leads in `passmark_floating_point_math` (715866 vs 507406, up 41.1%), `passmark_physics` (25847 vs 17753, up 45.6%), and `passmark_extended_instructions` (230609 vs 199048, up 15.9%).

The practical takeaway: the EPYC 9655P is the choice for integer-heavy compute, compression workloads, encryption, and all multi-threaded rendering. The Xeon 6781P wins only in prime-number finding, and even then by a hair.

FAQ

Q: Which processor has a higher average benchmark score?

A: The AMD EPYC 9655P has an average benchmark score of 397773, compared to the Intel Xeon 6781P’s 315524. The AMD part also sits at the 100th percentile among all CPUs, while Intel is at the 99th.

Q: How large is the AMD advantage in the Cinebench multicore tests?

A: The EPYC 9655P leads by 36% in all three Cinebench multicore tests (R15, R20, R23). Scores are 13744 vs 10105, 57268 vs 42106, and 136354 vs 100254 respectively.

Q: Does Intel win any benchmark at all?

A: Yes, one: `passmark_find_prime_numbers`. Intel scores 1687 against AMD’s 1686, a 0.1% margin. Every other head-to-head test goes to AMD.

Q: What is the biggest performance gap between the two?

A: The largest delta is in `passmark_integer_math`, where AMD’s 1225251 beats Intel’s 584834 by 109.5%. The second-largest is `passmark_data_encryption` at 84% (220074 vs 119623).

Q: How do the processors compare on single-thread performance?

A: AMD leads by 22.1% in both `passmark_single_thread` and `passmark_singlethread`, with identical scores of 3849 vs 3152. The EPYC also posts a Cinebench R15 single-core score of 1940; no Intel R15 single-core score is listed.

Q: What are the nearest rivals for each processor?

A: For AMD, the closest rival is the AMD Ryzen Threadripper PRO 9995WX, which scores 412068 — 3.5% higher. The AMD EPYC 9535 trails by 4.8% (379408), and the EPYC 9655 trails by 6.5% (373479). For Intel, the AMD EPYC 9575F is 1.2% lower (311774), the AMD EPYC 9734 is 1.6% lower (310619), and the AMD Ryzen Threadripper PRO 9985WX is 1.6% higher (320749).

Head-to-Head Benchmarks

The 14-test head-to-head table reveals a pattern of consistent AMD dominance with escalating margins. Starting with the narrowest wins: `passmark_find_prime_numbers` goes to Intel by 0.1% (1687 vs 1686), and `passmark_extended_instructions` goes to AMD by 15.9% (230609 vs 199048). From there, the gaps widen steadily.

The three Cinebench multicore tests all show exactly 36% deltas. This uniformity is notable because Cinebench R15, R20, and R23 have different workload characteristics and scoring scales — yet the relative performance ratio remains constant. This suggests the AMD advantage is structural, tied to core count and architecture rather than any single optimization.

The `passmark_multithread` test shows a 36.1% lead (160490 vs 117946), nearly identical to the Cinebench multicore margin. `passmark_floating_point_math` shows 41.1% (715866 vs 507406), while `passmark_data_compression` shows 42.8% (3486158 vs 2441690). The `passmark_physics` test yields 45.6% (25847 vs 17753).

The two most extreme results deserve emphasis. `passmark_data_encryption` gives AMD an 84% advantage (220074 vs 119623) — this is a workload where AMD’s architecture provides a nearly 2x throughput. `passmark_integer_math` is even more lopsided at 109.5% (1225251 vs 584834), meaning the EPYC executes more than twice as many integer operations per unit time. `passmark_random_string_sorting` sits between these at 68.2% (451824 vs 268573).

Single-thread performance, often considered a potential Intel strength, also favors AMD by 22.1% (3849 vs 3152) in both PassMark single-thread tests. This is significant because it indicates the EPYC’s Zen 5 cores are individually faster, not merely more numerous.

Specification Differences

The two processors differ on nearly every core specification. The AMD EPYC 9655P has 96 cores and 192 threads, while the Intel Xeon 6781P has 80 cores and 160 threads — a 16-core and 32-thread deficit for Intel. Base clocks are 2.60 GHz for AMD versus 2.00 GHz for Intel, and boost clocks are 4.50 GHz versus 3.80 GHz. AMD’s TDP is 400 W against Intel’s 350 W.

Cache configurations differ substantially. AMD allocates 80 KB L1 per core (totaling 7.68 MB across 96 cores), while Intel uses 112 KB L1 per core (totaling 8.96 MB across 80 cores). L2 is 1 MB per core for AMD (96 MB total) versus 2 MB per core for Intel (160 MB total). L3 cache is 384 MB shared for AMD versus 336 MB shared for Intel. In aggregate, Intel has more L1 and L2 cache, but AMD has more L3.

Memory subsystems diverge sharply. AMD uses a twelve-channel memory bus with 576.0 GB/s bandwidth; Intel uses eight channels with 409.6 GB/s. Both support DDR5 and ECC memory. PCIe lanes are Gen 5 for both, but Intel offers 136 CPU-only lanes versus AMD’s 128.

The sockets are incompatible: AMD uses Socket SP5, Intel uses Socket 4710. Process nodes differ — AMD is on TSMC’s 4 nm process, Intel on its own 5 nm. The die configuration also differs: AMD uses 12 chiplets of 70.6 mm² each (total transistor count 99,780 million), while Intel uses 2 dies of 598 mm² each with no transistor count listed.

Release dates and launch MSRP also differ. AMD launched on 2024-10-09 with a launch MSRP of $10811; Intel launched on 2025-02-23 with a launch MSRP of $8960.

Architecture Differences

The EPYC 9655P is built on AMD’s Zen 5 architecture, codenamed Turin, part of the EPYC 9005 series. It uses a chiplet design with 12 separate 70.6 mm² dies fabricated by TSMC on a 4 nm process, totaling 99,780 million transistors. This is a mature multi-die approach that allows AMD to scale core counts and cache simultaneously.

The Xeon 6781P uses Intel’s Granite Rapids architecture, part of the Xeon 6 family (Granite Rapids-SP). It is built on Intel’s 5 nm process with 2 dies of 598 mm² each — a much larger die size per chip, which suggests a monolithic or dual-die design rather than AMD’s multi-chiplet strategy. No transistor count is provided for Intel.

Cache distribution reflects architectural philosophy. AMD’s per-core L1 and L2 are smaller (80 KB and 1 MB), but the shared L3 is larger (384 MB). Intel’s per-core L1 and L2 are larger (112 KB and 2 MB), but the shared L3 is smaller (336 MB). This suggests AMD optimizes for shared data access across many cores, while Intel emphasizes per-core locality.

Memory architecture differences are stark: AMD’s twelve-channel bus provides 576.0 GB/s, which is 40.6% more bandwidth than Intel’s eight-channel 409.6 GB/s. In memory-bandwidth-bound workloads like data compression and random string sorting, this explains part of AMD’s large leads. The PCIe lane count also favors Intel (136 vs 128), which may matter for systems with many expansion cards.

Both processors have no integrated graphics, target the server/workstation segment, and are production-active. Neither has an unlocked multiplier.

The Verdict

The data directs different workloads to different processors, but the direction is heavily skewed toward AMD. For multi-threaded rendering, the EPYC 9655P wins all three Cinebench multicore tests by 36%. For integer-heavy computation, it more than doubles the Xeon’s throughput in `passmark_integer_math`. For encryption workloads, it leads by 84%. For data compression, by 42.8%. For floating-point math, by 41.1%. For physics simulations, by 45.6%. Even in single-thread performance, where Intel might traditionally compete, AMD leads by 22.1%.

The Intel Xeon 6781P holds exactly one advantage: prime number finding, by 0.1%. That is a statistical tie. If a workload is dominated by primality testing, the Intel part is technically the winner, but the margin is within noise.

The EPYC 9655P’s higher core count (96 vs 80), higher clocks (2.60/4.50 GHz vs 2.00/3.80 GHz), larger L3 cache (384 MB vs 336 MB), and higher memory bandwidth (576.0 GB/s vs 409.6 GB/s) all contribute to its benchmark dominance. The Xeon’s advantages — larger per-core L1/L2, more PCIe lanes (136 vs 128), and a lower 350 W TDP — do not translate into any meaningful benchmark win.

For buyers, the choice depends on whether the Xeon’s single 0.1% victory in prime numbers matters. For every other measured workload, the EPYC 9655P is faster, often dramatically so. The EPYC also holds a higher average benchmark score (397773 vs 315524) and sits at the 100th percentile of all CPUs. The Xeon, while at the 99th percentile, is in a different performance class in the data.

The launch MSRP difference ($10811 for AMD vs $8960 for Intel) is notable, but the benchmark data shows AMD delivering substantially higher performance per dollar in most categories. The verdict is unambiguous: choose the AMD EPYC 9655P for any workload represented in these benchmarks, unless prime-number finding at a 0.1% margin is the sole criterion.