CPU Comparison

The AMD EPYC 9965 and AMD EPYC 9754 are both flagship server processors, but the data places them in different tiers. The 9965 wins all 17 head-to-head benchmarks, with margins ranging from 36.4% in single-threaded tests to 112.7% in physics simulations. The 9965 also holds a 70.4% higher average benchmark score (620,487 vs 364,371). The 9754, however, remains a highly capable 128-core part that trades blows with Intel’s Xeon 6960P (0.2% difference in average score). For any workload where raw compute throughput is the priority, the 9965 is the clear choice; the 9754 makes sense only when its lower core count, lower TDP, and lower launch MSRP align with specific deployment constraints.

The Verdict

The benchmark data is unambiguous: the AMD EPYC 9965 outperforms the AMD EPYC 9754 in every single test recorded. The 9965’s average benchmark score of 620,487 is 70.4% higher than the 9754’s 364,371. In the nearest rival comparison, the 9965 sits 18.5% above the AMD EPYC 9845, while the 9754 is nearly tied with the Intel Xeon 6960P, trailing by just 0.2%. This means the 9754 is competitive with current top-end Intel parts, but the 9965 is in a class of its own, with a 100th percentile ranking among all CPUs.

The choice depends on whether you need the absolute maximum performance or a more balanced power profile. The 9965, with its 192 cores and 384 threads, delivers a 62.8% lead in all Cinebench tests, including a 136,661 score in Cinebench R23 multi-core. The 9754, with 128 cores and 256 threads, still posts a respectable 83,939 in the same test. For users with workloads that scale linearly with core count, like large-scale virtualization or heavy multi-threaded rendering, the 9965’s 64 additional cores are decisive. For those with tighter power budgets, the 9754’s 360W TDP versus the 9965’s 500W TDP is a significant operational difference, though the 9965’s performance per watt is not directly measured in the data.

Architecture Differences

The two processors represent two different generations of AMD’s dense-core strategy. The 9965 is built on the Zen 5 architecture, codenamed Turin, part of the EPYC 9005 series. It uses a 3nm process node from TSMC. The 9754 is based on Zen 4, codenamed Bergamo, part of the EPYC 9004 series, and uses a 5nm process node, also from TSMC. The 9965’s smaller process node is a key enabler for its higher core count and clock speeds.

Both CPUs feature the same AMD Socket SP5 and support DDR5 memory with a twelve-channel memory bus. The 9965 offers a memory bandwidth of 576.0 GB/s, while the 9754 provides 460.8 GB/s, a 25% difference that favors the newer part. The 9965 also has a larger L3 cache at 384 MB shared, versus 256 MB on the 9754. Per-core L1 cache is larger on the 9965 (80 KB) than on the 9754 (64 KB), while L2 cache is identical at 1 MB per core. The 9754’s transistor count is listed at 71,000 million, with a die size of 8x 73 mm²; no comparable data is provided for the 9965.

The 9965 boosts to 3.70 GHz, while the 9754 reaches 3.10 GHz. Base clocks are identical at 2.25 GHz. Both support PCIe Gen 5 with 128 lanes (CPU only) and have ECC memory support. The 9965 was released on 2024-10-09, over a year after the 9754’s release on 2023-06-12. The 9965 also carries a higher launch MSRP of $14,813, compared to $11,900 for the 9754.

Head-to-Head Benchmarks

The most striking pattern is the consistency of the 9965’s lead. In all six Cinebench tests, the delta is exactly 62.8%. This includes Cinebench R15 multi-core (13,775 vs 8,460), R15 single-core (1,944 vs 1,194), R20 multi-core (57,397 vs 35,254), R20 single-core (8,103 vs 4,977), R23 multi-core (136,661 vs 83,939), and R23 single-core (19,293 vs 11,850). The fact that the single-core and multi-core margins are identical suggests the architectural efficiency gain per clock is uniform across the board.

PassMark tests show more varied margins, revealing where the 9965 excels most. The largest gap is in physics simulations, where the 9965 scores 18,707 versus 8,793, a 112.7% advantage. Random string sorting follows with a 106.5% lead (633,030 vs 306,481). Floating point math shows a 96.1% delta (1,153,453 vs 588,187), and integer math is 87.6% higher (1,926,069 vs 1,026,896). Prime number finding shows a 100% improvement (1,208 vs 604).

The smallest margins are in single-threaded workloads. PassMark single-thread shows a 36.4% lead (3,176 vs 2,328), and data encryption is 50.3% higher (348,449 vs 231,891). Data compression is 59.6% ahead (5,679,990 vs 3,558,043), and extended instructions are 70.9% higher (383,298 vs 224,322). PassMark multi-thread shows a 62.6% delta (160,542 vs 98,752). These results indicate that while the 9965 wins everywhere, its advantage grows with thread count and complexity, particularly in floating-point and physics-heavy workloads.

Specification Differences

The core specifications that differ between the two CPUs are straightforward. The 9965 has 192 cores and 384 threads, while the 9754 has 128 cores and 256 threads. Boost clock differs: 3.70 GHz for the 9965 versus 3.10 GHz for the 9754. The TDP is 500W for the 9965 and 360W for the 9754. The process node is 3nm for the 9965 and 5nm for the 9754.

Cache configurations differ in L1 (80 KB per core vs 64 KB) and L3 (384 MB vs 256 MB), but L2 is the same at 1 MB per core. Memory bandwidth is 576.0 GB/s for the 9965 versus 460.8 GB/s for the 9754. The 9754 has listed transistor count (71,000 million) and die size (8x 73 mm²), while the 9965 has none listed. The release dates differ by over a year, and the launch MSRP is $14,813 for the 9965 versus $11,900 for the 9754. Both share the same socket, memory type, memory bus width, PCIe lanes, ECC support, and lack of integrated graphics.

FAQ

Q: Which CPU is faster in multi-core workloads?

A: The AMD EPYC 9965 is faster in every multi-core test. In Cinebench R23 multi-core, it scores 136,661 versus 83,939 for the 9754, a 62.8% difference. PassMark multi-thread shows a 62.6% lead (160,542 vs 98,752).

Q: Is the single-core performance gap as large as the multi-core gap?

A: No, the single-core gap is smaller but still significant. In Cinebench R23 single-core, the 9965 scores 19,293 versus 11,850 for the 9754, also a 62.8% delta. However, in PassMark single-thread, the delta is only 36.4% (3,176 vs 2,328).

Q: What is the difference in memory bandwidth?

A: The 9965 offers 576.0 GB/s of memory bandwidth, while the 9754 provides 460.8 GB/s. Both use DDR5 with a twelve-channel memory bus.

Q: How does the 9754 compare to its nearest rival, the Intel Xeon 6960P?

A: The 9754’s average benchmark score is 364,371, which is 0.2% lower than the Intel Xeon 6960P’s 365,194. This places the two processors in a statistical tie.

Q: What is the TDP difference between the two?

A: The 9965 has a TDP of 500W, while the 9754 has a TDP of 360W. This 140W difference is notable for power-constrained data centers.

Q: Which CPU has a higher boost clock?

A: The 9965 boosts to 3.70 GHz, which is higher than the 9754’s 3.10 GHz. Their base clocks are identical at 2.25 GHz.

Where Each One Wins

The 9965 wins in every single benchmark category recorded. It is the definitive choice for workloads that demand maximum core count and memory bandwidth. The 112.7% lead in PassMark physics and the 106.5% lead in random string sorting indicate that the 9965 is particularly strong in scientific computing, simulation, and data processing tasks that involve complex floating-point math and memory-intensive sorting algorithms. Its 384 MB of L3 cache and 576.0 GB/s memory bandwidth make it ideal for large in-memory databases and high-performance computing clusters.

The 9754, while losing all head-to-head comparisons, still represents a very capable 128-core option. Its 360W TDP is 28% lower than the 9965’s 500W, making it a more practical choice for dense deployments where power delivery and cooling are limiting factors. Its average score of 364,371 is within 0.2% of the Intel Xeon 6960P, meaning it is fully competitive with the leading Intel offering in its generation. The 9754’s lower launch MSRP of $11,900 also makes it a more approachable entry point for large-scale multi-socket systems, though pricing is not a factor in its performance profile.

For organizations with existing EPYC 9004 platform investments, the 9754 offers a straightforward upgrade path without changing sockets. The 9965, however, is the performance leader in this comparison, with a 70.4% higher average score overall. The data shows no workload category where the 9754 comes out ahead, so the only reason to select it over the 9965 is if the lower TDP or lower launch MSRP is a hard requirement. Otherwise, the 9965 is the superior part in every measurable way.