AMD EPYC 9455P vs AMD EPYC 9684X Comparison

The AMD EPYC 9684X and AMD EPYC 9455P are both server processors in the EPYC family, but they occupy very different positions in the design. The head-to-head record shows 15 wins for the 9684X and 2 for the 9455P, yet the 9455P wins its two tests by a wide margin of 22.8%, while the 9684X wins its tests by margins from 4% to 82.5%. The following sections walk through the benchmark data, the architectural differences, and the use cases that fit each part.

Head-to-Head Benchmarks

The Cinebench results show a uniform 4.2% advantage for the 9684X across all six tests. In Cinebench R15 multi-core, the 9684X scores 10418 against the 9455P's 9999. In R15 single-core, it is 1470 vs 1411. The R20 tests follow the same pattern: 43409 versus 41666 for multi-core, and 6128 versus 5882 for single-core. R23 multi-core is 103355 versus 99206, and R23 single-core is 14591 versus 14005. Every Cinebench delta is exactly 4.2%, which points to a consistent per-instruction advantage rather than a scaling effect.

The Passmark suite reveals a more divergent spread. The 9684X wins every data-heavy Passmark test. The largest margin is in Passmark find prime numbers, where the 9684X scores 2020 against 1107, a 82.5% delta. Data encryption follows at 54.6% (178453 versus 115403), physics at 42.6% (24686 versus 17315), random string sorting at 43.9% (349126 versus 242701), data compression at 39.9% (2698807 versus 1928897), integer math at 39.2% (844145 versus 606239), floating point math at 32% (472174 versus 357783), and extended instructions at 29.4% (177956 versus 137485). The smallest multi-thread margin is in Passmark multithread, at 4% (121595 versus 116927).

The 9455P wins only the two single-thread Passmark tests, which are passmark_single_thread and passmark_singlethread. Both show the same scores: 3745 for the 9455P versus 2891 for the 9684X, a 22.8% delta. This is the only set of tests in the entire head-to-head where the 9455P leads. Note that the 9455P loses the Cinebench single-thread tests by 4.2%, so the single-thread advantage is specific to the Passmark methodology.

Looking at the broader database context, the 9684X sits at the 99th percentile of all CPUs, with an average benchmark score of 266914. Its nearest rivals include the AMD Ryzen Threadripper 9970X, which is 4.6% faster, the Intel Xeon 6780E at 4.8% faster, the Intel Xeon 6980P at 6.1% slower, and the AMD EPYC 9565 at 6.5% faster. The 9455P also sits at the 99th percentile, with an average score of 217854. Its nearest rivals include the Intel Xeon w9-3595X, which is 3.8% slower, the Intel Xeon 6747P at 8.6% faster, the Intel Xeon 6741P at 11.8% slower, and the AMD EPYC 9335 at 12.2% slower.

The Verdict

The data makes a clear separation. For workloads that use many threads, the 9684X is the stronger part. It wins every multi-threaded benchmark in the head-to-head set, with the narrowest margin at 4% and the widest at 82.5%. It also has twice as many cores (96 versus 48) and threads (192 versus 96), which explains its consistent multi-thread lead. The 9684X has an average benchmark score of 266914, while the 9455P has 217854. The 9684X is 4.6% behind the Ryzen Threadripper 9970X and 4.8% behind the Xeon 6780E, but it is 6.1% ahead of the Xeon 6980P, so it is not the strongest in the database.

The 9455P is the part to pick when single-thread performance is the priority. Its Passmark single-thread score is 22.8% higher, and it has higher base and boost clocks (3.15/4.40 GHz versus 2.55/3.70 GHz) and a lower TDP (300 W versus 400 W). It also uses a 4 nm process node versus 5 nm. But the 9455P loses every multi-thread test, so it is not a general-purpose alternative. The verdict from the data is straightforward: choose the 9684X for throughput-bound workloads, and the 9455P for single-thread latency.

Architecture Differences

The two processors belong to different generations and architectures. The 9684X is Zen 4, from the Genoa-X codename, in the EPYC 9004 series. The 9455P is Zen 5, from the Turin codename, in the EPYC 9005 series. The process node differs: the 9684X is 5 nm, the 9455P is 4 nm, both from TSMC.

The transistor counts and die sizes differ substantially. The 9684X has 135,240 million transistors on a 12x 72 mm² die; the 9455P has 66,520 million transistors on an 8x 70.6 mm² die. The L1 cache per core is 64 KB for the 9684X and 80 KB for the 9455P. L2 cache is identical at 1 MB per core. L3 cache is very different: 1152 MB shared for the 9684X, 256 MB shared for the 9455P. The 9684X's much larger L3 is a likely factor in the data-heavy workloads where it wins by large margins.

Both support DDR5 memory with a twelve-channel bus. The peak memory bandwidth is higher for the 9455P at 576.0 GB/s, versus 460.8 GB/s for the 9684X. Both have Gen 5 PCIe with 128 lanes. Both use AMD Socket SP5 and are ECC-capable. Neither has integrated graphics: the 9684X lists none, and the 9455P lists N/A. Both are marked as locked production parts.

Specification Differences

The specification differences between the two parts are: cores (96 versus 48), threads (192 versus 96), base clock (2.55 GHz versus 3.15 GHz), boost clock (3.70 GHz versus 4.40 GHz), TDP (400 W versus 300 W), process node (5 nm versus 4 nm), L1 cache per core (64 KB versus 80 KB), L3 cache shared (1152 MB versus 256 MB), memory bandwidth (460.8 GB/s versus 576.0 GB/s), integrated graphics (none versus N/A), release date (2023-06-12 versus 2024-10-09), launch MSRP ($14756 versus $4819), and part number (100-100000892 versus 100-000001563). The series also differs: EPYC 9004 versus EPYC 9005. The generation and codename are different as well: Zen 4 Genoa-X versus Zen 5 Turin.

FAQ

Q: Which processor wins more head-to-head benchmarks?

A: The AMD EPYC 9684X wins 15 of the 17 recorded tests; the AMD EPYC 9455P wins 2.

Q: How large is the single-thread advantage of the 9455P?

A: The 9455P scores 3745 on Passmark single-thread, versus 2891 for the 9684X, a 22.8% edge.

Q: Does the 9684X have any single-core advantage?

A: Yes, the 9684X leads in all three Cinebench single-core tests by 4.2% (R15, R20, and R23).

Q: Which part has more L3 cache?

A: The 9684X has 1152 MB shared L3, while the 9455P has 256 MB shared L3.

Q: How does the 9684X compare to the AMD EPYC 9565 in the database?

A: The 9684X is 6.5% behind the AMD EPYC 9565 in average benchmark score.

Q: What is the memory bandwidth difference?

A: The 9455P has 576.0 GB/s of memory bandwidth, while the 9684X has 460.8 GB/s.

Where Each One Wins

The 9684X wins every multi-threaded workload in the head-to-head set. The largest advantages are in prime-number searching (82.5%), data encryption (54.6%), physics (42.6%), random string sorting (43.9%), data compression (39.9%), integer math (39.2%), floating point math (32%), and extended instructions (29.4%). It also wins the Cinebench multi-core tests by 4.2% and Passmark multithread by 4%. This points to workloads that are heavy on integer arithmetic, data processing, and large caches, where the 9684X's 96 cores and 1152 MB L3 cache provide the winning margin.

The 9455P wins only the single-thread Passmark tests, with a 22.8% margin. That advantage comes from a higher base clock (3.15 versus 2.55 GHz), a higher boost clock (4.40 versus 3.70 GHz), a lower TDP (300 W versus 400 W), and a higher memory bandwidth (576.0 GB/s versus 460.8 GB/s). The 9455P also uses a 4 nm process node. For single-thread-bound tasks, such as certain database queries or low-occupancy request handling, the 9455P offers the higher score. For any workload that can use more than a handful of threads, the 9684X is the part with the higher result, and the database shows 15 wins in its favor.