AMD EPYC 9354 vs Intel Xeon 6745P Comparison

Head-to-Head Benchmarks

The benchmark database records 16 direct comparisons between the Intel Xeon 6745P and the AMD EPYC 9354, and the results are heavily skewed in favor of the Intel part. The Xeon 6745P claims 12 wins, while the EPYC 9354 takes 4. The overall average benchmark score for the Intel processor is 154,858, placing it in the 98th percentile of all CPUs, while the AMD processor averages 126,810, good for the 97th percentile.

The most decisive Intel victories come in floating-point math and extended instruction workloads. In the PassMark floating-point math test, the Xeon 6745P scores 267,438 against 188,894 for the EPYC 9354, a 41.6% advantage. That is the single largest delta in the entire head-to-head set. Similarly, in PassMark extended instructions, Intel leads 108,326 to 86,176, a 25.7% margin. These are not marginal differences; they indicate a substantial architectural advantage in compute-heavy, vectorized tasks.

Single-thread performance also favors Intel by a wide margin. The PassMark single-thread test shows the Xeon 6745P at 3,450 versus 2,601 for the EPYC 9354, a 32.6% lead. This pattern repeats across the Cinebench series: in Cinebench R15 single-core, Intel scores 1,018 against 878 (15.9% higher); in R20 single-core, 4,244 against 3,659 (16% higher). The Intel part's higher boost clock of 4.30 GHz, compared to 3.80 GHz for the AMD chip, aligns with this single-thread dominance.

Multi-threaded workloads tell a similar story, though with smaller margins. In Cinebench R23 multi-core, the Xeon 6745P scores 71,578 versus 61,722 for the EPYC 9354, a 16% lead. The R15 multi-core test shows 7,214 against 6,221 (16% higher), and R20 multi-core shows 30,062 against 25,923 (16% higher). PassMark multi-thread also records a 16% Intel advantage, with 84,210 versus 72,615. Integer math is closer, at 336,926 versus 304,828, a 10.5% Intel lead. Data compression favors Intel as well, 1,352,801 versus 1,168,626, a 15.8% margin.

The AMD EPYC 9354 does secure four wins, and they are worth examining closely. The largest AMD victory is in PassMark physics, where it scores 9,281 against 6,144 for Intel, a 33.8% advantage. Prime number finding also goes strongly to AMD: 934 versus 681, a 27.1% lead. Data encryption is another AMD win, 71,400 versus 66,665, a 6.6% margin. Finally, random string sorting goes to AMD by a modest 5.1%, with scores of 140,690 and 133,528. These wins suggest the AMD architecture retains an edge in specific integer-heavy, branch-predictive, or memory-latency-sensitive workloads, even though the overall multi-thread scores favor Intel.

Looking at the nearest rivals in the database provides context. The Xeon 6745P's closest competitor is the Intel Xeon 676X, which averages 158,540, meaning the 6745P trails by 2.3%. The AMD EPYC 9355P averages 160,358 (3.4% ahead of the 6745P), and the EPYC 7663 averages 161,973 (4.4% ahead). The EPYC 9375F sits 4.7% higher at 162,497. For the EPYC 9354, its nearest rivals include the Intel Xeon 6730P at 124,756, which the AMD part leads by 1.6%, and the AMD Ryzen Threadripper PRO 5975WX at 124,171, which it leads by 2.1%. The Intel Xeon 6710E sits at 129,930, putting the EPYC 9354 2.4% behind that part.

FAQ

Q: Which processor has the higher average benchmark score?

A: The Intel Xeon 6745P averages 154,858, while the AMD EPYC 9354 averages 126,810. The Intel part sits in the 98th percentile of all CPUs; the AMD part sits in the 97th.

Q: How large is the Intel lead in multi-threaded Cinebench tests?

A: Across R15, R20, and R23 multi-core tests, the Intel Xeon 6745P leads by exactly 16% in each: 7,214 vs 6,221, 30,062 vs 25,923, and 71,578 vs 61,722 respectively.

Q: In which test does the AMD EPYC 9354 have its largest winning margin?

A: The largest AMD win is in PassMark physics, where the EPYC 9354 scores 9,281 against 6,144 for Intel, a 33.8% advantage.

Q: What is the difference in single-thread performance?

A: In PassMark single-thread, the Intel Xeon 6745P scores 3,450 versus 2,601 for the EPYC 9354, a 32.6% lead. Cinebench R20 single-core shows a 16% Intel advantage, 4,244 versus 3,659.

Q: How does the Intel part compare to its closest rival, the Xeon 676X?

A: The Xeon 676X has an average score of 158,540, which is 2.3% higher than the Xeon 6745P's 154,858.

Q: Does the EPYC 9354 have any advantages in memory-related or encryption workloads?

A: Yes, the EPYC 9354 wins in data encryption, scoring 71,400 versus 66,665 (a 6.6% margin), and in random string sorting, 140,690 versus 133,528 (a 5.1% margin).

The Verdict

The data points to a clear winner for most workloads. The Intel Xeon 6745P is the stronger processor in 12 of 16 benchmark comparisons, and its victories include the most important categories for general server and workstation use: multi-threaded rendering, single-thread responsiveness, integer math, floating-point math, and data compression. The 16% lead across all three Cinebench multi-core tests is consistent and significant. The 41.6% margin in floating-point math is particularly compelling for scientific, engineering, and financial workloads that rely heavily on vectorized computation. The 32.6% single-thread lead will also matter for database queries, application latency, and lightly threaded workloads.

The AMD EPYC 9354 is not without merit. Its 33.8% win in PassMark physics and 27.1% win in prime number finding indicate that specific integer-heavy or branch-predictive workloads may run faster on the AMD part. The 6.6% encryption advantage is also notable for security-focused applications. However, these wins are concentrated in a narrow set of tests, and the overall average score is 22% lower than the Intel part (126,810 vs 154,858).

For buyers choosing between these two, the decision should be driven by workload composition. If the dominant tasks involve physics simulation, prime number computation, encryption, or random string sorting, the EPYC 9354 may be the better fit. For essentially everything else, from rendering to general server throughput to single-threaded latency, the Xeon 6745P is the data-backed choice. The Intel part also carries a launch MSRP of $5250; the AMD part has a launch MSRP of $3420.

Specification Differences

The two processors share the same core and thread counts, both offering 32 cores and 64 threads. Both support DDR5 memory, both have ECC memory enabled, and both are listed as active server/workstation parts with no unlocked multipliers. The differences begin with clock speeds. The Intel Xeon 6745P has a base clock of 3.10 GHz and a boost clock of 4.30 GHz. The AMD EPYC 9354 has a base clock of 3.25 GHz and a boost clock of 3.80 GHz. The Intel part has a higher boost clock by 0.50 GHz, while the AMD part has a slightly higher base clock by 0.15 GHz.

Thermal design power differs, with the Intel part rated at 300 W and the AMD part at 280 W. The sockets are not interchangeable: Intel uses Socket 4710, while AMD uses Socket SP5. Memory channels also differ; the Intel part has an eight-channel memory bus with 409.6 GB/s bandwidth, while the AMD part has a twelve-channel bus with 460.8 GB/s bandwidth. PCIe lane counts differ as well, with the Intel part providing Gen 5 at 88 lanes (CPU only) and the AMD part providing Gen 5 at 128 lanes (CPU only).

The cache hierarchy differs in capacity and organization. The Intel Xeon 6745P has 112 KB of L1 per core, 2 MB of L2 per core, and 336 MB of shared L3. The AMD EPYC 9354 has 64 KB of L1 per core, 1 MB of L2 per core, and 256 MB of shared L3. The Intel part has more cache at every level. The release dates are also distinct: Intel launched on 2025-02-23, while AMD launched on 2022-11-09.

Architecture Differences

The architectural divide is significant. The Intel Xeon 6745P is built on the Granite Rapids architecture, specifically the Xeon 6 (Granite Rapids-SP) generation, and manufactured on a 5 nm process at Intel's own foundry. The die size is listed as 2x 598 mm². The AMD EPYC 9354 uses the Zen 4 architecture under the Genoa codename, part of the EPYC 9004 series, and is manufactured on a 5 nm process at TSMC. The AMD part has 52,560 million transistors and a die size of 8x 72 mm².

The cache architecture reflects the differing design philosophies. Intel's per-core L1 of 112 KB and L2 of 2 MB are double the AMD part's 64 KB L1 and 1 MB L2. The shared L3 also favors Intel, at 336 MB versus 256 MB. This larger cache hierarchy likely contributes to the Intel part's strong performance in data compression and floating-point workloads.

The Intel part has no integrated graphics, and the AMD part has a null entry for integrated graphics. The memory bandwidth difference is notable: AMD's twelve-channel memory bus delivers 460.8 GB/s, which is 51.2 GB/s higher than Intel's eight-channel 409.6 GB/s. This memory bandwidth advantage for AMD, combined with its wins in random string sorting and encryption, suggests that the EPYC 9354 may handle memory-latency-sensitive or pointer-chasing workloads more effectively. However, the benchmark data shows that Intel's larger cache and higher boost clock more than compensate in most measured tests.

The transistor counts tell a story of manufacturing scale. The AMD part integrates 52,560 million transistors across 8 chiplets of 72 mm² each, a multi-die approach. The Intel part uses a 2-die configuration with a total die area of 1,196 mm². These are fundamentally different physical designs, and the benchmark results show that Intel's monolithic-ish approach with larger per-core caches yields a higher average score despite the AMD part's higher memory bandwidth and more memory channels.