CPU Comparison

The AMD EPYC 9845 and AMD EPYC 9754 are both flagship-class server processors built for the same SP5 socket, but they represent two different generations of AMD’s dense-core strategy. The 9845, part of the EPYC 9005 series, leverages the Zen 5 architecture on a 3 nm process, while the 9754 from the EPYC 9004 series uses Zen 4 on a 5 nm node. Benchmark data shows a decisive performance gap favoring the newer 9845, which wins all 17 recorded head-to-head tests. This analysis breaks down exactly where those wins come from, what the architectural shifts mean, and which processor suits which workload profile.

Head-to-Head Benchmarks

The most striking pattern in the data is the consistency of the 9845’s lead. Across every single benchmark, from Cinebench multi-threaded renders to PassMark’s integer math, the 9845 posts higher scores, with deltas ranging from 28% to over 123%. The margin in Cinebench R23 multi-core is representative: the 9845 scores 130,037 against the 9754’s 83,939, a 54.9% advantage. That same 54.9% delta repeats across all six Cinebench tests (R15, R20, and R23, both single and multi-core), indicating a uniform architectural uplift rather than a workload-specific quirk.

Single-threaded performance tells a similar story. In Cinebench R23 single-core, the 9845 hits 18,358 versus 11,850 for the 9754, again a 54.9% gap. PassMark’s single-thread test shows a slightly smaller but still substantial 35.1% lead, with scores of 3,144 and 2,328 respectively. This is notable for a server chip: the 9845’s higher boost clock of 3.70 GHz versus 3.10 GHz on the 9754, combined with the newer Zen 5 core, delivers meaningful gains even in lightly threaded tasks.

The largest single deltas appear in PassMark’s physics simulation and prime number finding tests. The 9845 scores 19,631 in physics, a 123.3% improvement over the 9754’s 8,793. Prime number finding shows a 107.8% jump, from 604 to 1,255. These are extreme margins that point to the 9845’s superior per-core efficiency and larger cache hierarchy handling branch-heavy, memory-latency-sensitive workloads. Floating-point math also shows a big gap: 978,377 versus 588,187, a 66.3% lead. Integer math follows at 64.3% (1,687,531 vs 1,026,896), while random string sorting comes in at 75.6% (538,060 vs 306,481).

Even in areas where the 9754 was expected to be competitive, the 9845 pulls ahead. Data compression shows a 31.5% lead (4,680,013 vs 3,558,043), and data encryption shows a 28% advantage (296,808 vs 231,891). Extended instruction set performance is 40.3% higher (314,798 vs 224,322). The smallest deltas are still substantial, with single-threaded PassMark at 35.1% and encryption at 28%. There is no benchmark where the 9754 wins, the closest it gets is data encryption, where the 9845’s lead is still nearly a third of the 9754’s score.

Where Each One Wins

Given that the 9845 wins every recorded test, the use-case split is less about one CPU beating the other and more about whether the 9754’s lower power draw and lower launch price justify its adoption for less demanding roles. The 9845 is the clear choice for any workload that scales with core count or benefits from raw compute throughput. Its 160 cores and 320 threads, combined with a 320 MB shared L3 cache, make it ideal for large-scale virtualization, database processing, and multi-tenant cloud workloads where the 54.9% multi-core advantage translates directly to higher throughput per socket.

The 9754, with 128 cores and 256 threads, still holds value for dense compute environments that are power-constrained. Its TDP of 360 W is lower than the 9845’s 390 W, and its launch MSRP of $11900 is less than the 9845’s $13564. For organizations with existing SP5 infrastructure that doesn’t require the absolute maximum core count, the 9754 can serve as a capable workhorse for container orchestration or web serving, but the data shows it will take roughly 55% longer to complete the same multi-threaded render or simulation task compared to the 9845.

Single-threaded workloads are also a win for the 9845. The 54.9% delta in Cinebench single-core and 35.1% in PassMark single-thread indicate that the 9845 is not just a core-count monster but also a faster core. This matters for mixed workloads that include legacy software with limited threading. The 9754’s only theoretical advantage is its lower TDP, which could enable higher density in power-capped data centers, but the benchmark results do not show any performance scenario where the 9754 leads.

Architecture Differences

The two processors are separated by a full generation of AMD’s dense-core design. The 9845 uses the Zen 5 architecture, codenamed Turin, built on a 3 nm process at TSMC. The 9754 uses Zen 4, codenamed Bergamo, on a 5 nm process at the same foundry. This node shrink is a primary driver of the 9845’s efficiency and clock speed advantages. The 9845’s base clock is 2.10 GHz and boost clock 3.70 GHz, while the 9754 runs at 2.25 GHz base and 3.10 GHz boost. The 9754 actually has a higher base clock, but the 9845’s boost clock is 19.4% higher, which contributes to the single-threaded wins.

Core and cache configurations differ significantly. The 9845 packs 160 cores and 320 threads, a 25% increase over the 9754’s 128 cores and 256 threads. L1 cache per core is 80 KB on the 9845 versus 64 KB on the 9754, a 25% increase per core. L2 cache is identical at 1 MB per core. L3 cache jumps from 256 MB shared on the 9754 to 320 MB shared on the 9845, a 25% increase. The 9845 also benefits from a higher memory bandwidth rating: 576.0 GB/s versus 460.8 GB/s on the 9754, despite both using twelve-channel DDR5 memory.

The 9754’s transistor count is listed at 71,000 million across a die size of 8x 73 mm². No transistor or die size data is provided for the 9845, but the 3 nm process and higher core count suggest a denser design. Both CPUs support ECC memory and PCIe Gen 5 with 128 lanes on the CPU only, so platform-level I/O capabilities are identical. Neither has integrated graphics, and both have locked multipliers. The 9845 was released on 2024-10-09, while the 9754 came earlier on 2023-06-12. The part numbers differ (100-000001458 for the 9845, 100-000001234 for the 9754), confirming distinct silicon.

FAQ

Q: Is the AMD EPYC 9845 faster in every benchmark?

A: Yes. The data shows the 9845 wins all 17 head-to-head tests, with deltas ranging from 28% in data encryption to 123.3% in physics simulation.

Q: How much better is the 9845 in multi-threaded workloads?

A: In Cinebench R23 multi-core, the 9845 scores 130,037 versus 83,939 for the 9754, a 54.9% advantage. PassMark multithread shows a similar 54.9% lead (152,985 vs 98,752).

Q: Does the 9754 have any advantage in single-threaded performance?

A: No. The 9845 leads by 54.9% in Cinebench R23 single-core (18,358 vs 11,850) and by 35.1% in PassMark single-thread (3,144 vs 2,328).

Q: What is the difference in core count and cache?

A: The 9845 has 160 cores and 320 threads, with 320 MB of shared L3 cache. The 9754 has 128 cores and 256 threads, with 256 MB of shared L3 cache. L1 cache per core is 80 KB on the 9845 versus 64 KB on the 9754.

Q: Are the memory bandwidth specifications different?

A: Yes. The 9845 is rated for 576.0 GB/s, while the 9754 is rated for 460.8 GB/s. Both use twelve-channel DDR5 memory.

Q: What are the launch prices of these CPUs?

A: The 9845 has a launch MSRP of $13564, and the 9754 has a launch MSRP of $11900.

Specification Differences

• Series: EPYC 9005 series (9845) vs EPYC 9004 series (9754)
• Architecture: Zen 5 (9845) vs Zen 4 (9754)
• Codename: Turin (9845) vs Bergamo (9754)
• Process Node: 3 nm (9845) vs 5 nm (9754)
• Cores: 160 (9845) vs 128 (9754)
• Threads: 320 (9845) vs 256 (9754)
• Base Clock: 2.10 GHz (9845) vs 2.25 GHz (9754)
• Boost Clock: 3.70 GHz (9845) vs 3.10 GHz (9754)
• TDP: 390 W (9845) vs 360 W (9754)
• L1 Cache: 80 KB per core (9845) vs 64 KB per core (9754)
• L3 Cache: 320 MB shared (9845) vs 256 MB shared (9754)
• Memory Bandwidth: 576.0 GB/s (9845) vs 460.8 GB/s (9754)
• Transistors: Not specified (9845) vs 71,000 million (9754)
• Die Size: Not specified (9845) vs 8x 73 mm² (9754)
• Release Date: 2024-10-09 (9845) vs 2023-06-12 (9754)
• Launch MSRP: $13564 (9845) vs $11900 (9754)
• Part Number: 100-000001458 (9845) vs 100-000001234 (9754)

The Verdict

The benchmark data is unambiguous: the AMD EPYC 9845 outperforms the 9754 in every measurable category, often by wide margins. The 54.9% lead in all Cinebench tests and the 123.3% lead in physics simulation demonstrate a generational leap that goes beyond mere core count. The 9845’s 160 cores, larger caches, higher boost clock, and faster memory bandwidth combine to deliver a processor that is uniformly faster for compute-heavy, memory-intensive, and even single-threaded tasks. For any new deployment where performance is the primary criterion, the 9845 is the only choice based on this data.

The 9754 is not without merit. Its lower TDP of 360 W and launch MSRP of $11900 make it a more economical option for power-sensitive or cost-conscious environments. But the data shows that choosing the 9754 means accepting a 28-55% performance deficit across the board, depending on the workload. For mixed workloads that include legacy single-threaded applications, the 9845’s 35.1% single-thread lead further erodes the 9754’s appeal. The 9845’s 3 nm process and Zen 5 architecture represent the current state of the art, and the benchmark results confirm that the extra cost of the 9845 buys substantial, measurable performance across every test. If the goal is maximum throughput per socket, the 9845 is the definitive winner.