AMD EPYC 9684X vs AMD EPYC 9754 Comparison

Head-to-Head Benchmarks

The benchmark data splits these two processors into clearly distinct profiles. The AMD EPYC 9684X dominates the Cinebench suite, winning all six recorded Cinebench tests by a uniform margin. In Cinebench R15 multicore, the 9684X scores 10418 against 8460 for the 9754, a delta of -18.8% (negative deltas favor the 9684X). The same -18.8% margin appears across Cinebench R15 singlecore, R20 multicore, R20 singlecore, R23 multicore, and R23 singlecore. This consistency suggests a fundamental per-core advantage rather than workload-specific behavior. The 9684X reaches 43409 in R20 multicore versus 35254, and 103355 in R23 multicore versus 83939.

Single-thread results reinforce the picture. The 9684X posts 1470 in Cinebench R15 singlecore and 14591 in R23 singlecore, while the 9754 manages 1194 and 11850 respectively. PassMark single-thread scoring shows the 9684X at 2891 versus 2328, a -19.5% delta. The 9684X also wins PassMark multithread at 121595 against 98752, again at -18.8%.

The 9754 fights back in specialized throughput workloads. PassMark data compression shows the 9754 at 3558043 versus 2698807, a 31.8% advantage. Data encryption goes to the 9754 at 231891 against 178453, a 29.9% margin. Extended instructions favor the 9754 by 26.1% (224322 versus 177956). Floating-point math and integer math also go to the 9754, with margins of 24.6% and 21.6% respectively.

The 9684X secures its largest wins in prime number finding and physics simulation. PassMark find prime numbers shows the 9684X at 2020 versus 604, a massive -70.1% delta. PassMark physics reaches 24686 versus 8793, a -64.4% margin. Random string sorting also favors the 9684X at 349126 versus 306481, a -12.2% delta.

The win tally stands at 5 for the 9754 and 12 for the 9684X. But raw wins do not tell the whole story. The 9684X wins every single-core test and every Cinebench test, while the 9754 wins a narrower set of data-intensive PassMark workloads. The 9754's wins are large, but they concentrate in areas where raw core count and memory bandwidth matter most. The 9684X's wins span a broader range of general-purpose compute.

FAQ

Q: Which processor has more cores?

A: The AMD EPYC 9754 has 128 cores and 256 threads, while the AMD EPYC 9684X has 96 cores and 192 threads.

Q: Which processor has the higher clock speed?

A: The AMD EPYC 9684X has a base clock of 2.55 GHz and a boost clock of 3.70 GHz. The AMD EPYC 9754 has a base clock of 2.25 GHz and a boost clock of 3.10 GHz.

Q: How do the Cinebench R23 multicore scores compare?

A: The AMD EPYC 9684X scores 103355 in Cinebench R23 multicore, while the AMD EPYC 9754 scores 83939. The 9684X leads by -18.8%.

Q: In which workloads does the EPYC 9754 take the lead?

A: The 9754 leads in PassMark data compression (3558043 versus 2698807, 31.8% ahead), data encryption (29.9% ahead), extended instructions (26.1% ahead), floating-point math (24.6% ahead), and integer math (21.6% ahead).

Q: What is the difference in L3 cache size?

A: The AMD EPYC 9684X has 1152 MB of shared L3 cache. The AMD EPYC 9754 has 256 MB of shared L3 cache.

Q: Which processor has the higher average benchmark score?

A: The AMD EPYC 9754 has an average benchmark score of 364371, while the AMD EPYC 9684X has an average benchmark score of 266914. The 9754 also ranks in the 100th percentile versus all CPUs, while the 9684X ranks in the 99th percentile.

Architecture Differences

Both processors belong to the EPYC 9004 series, share the Zen 4 architecture, and are built on TSMC's 5 nm process. The similarity ends there. The 9754 uses the Bergamo codename and is classified under the Zen 4c generation, while the 9684X uses the Genoa-X codename and falls under the standard Zen 4 generation. This distinction explains much of the performance divergence.

The transistor counts differ substantially. The 9754 packs 71,000 million transistors across 8 dies, each 73 mm². The 9684X contains 135,240 million transistors across 12 dies, each 72 mm². The 9684X crams nearly twice the transistors into a similar die area, which aligns with its larger L3 cache allocation. The 9754's cache hierarchy is comparatively modest: 64 KB of L1 per core, 1 MB of L2 per core, and 256 MB of shared L3. The 9684X matches the per-core L1 and L2 figures but jumps to 1152 MB of shared L3, a 4.5x increase in last-level cache.

The core count difference flows from these design choices. The 9754's 128 cores and 256 threads reflect a density-optimized approach, using more compact cores to maximize parallelism. The 9684X's 96 cores and 192 threads sacrifice 32 cores to make room for the massive L3 pool. The cache advantage appears to drive the 9684X's single-thread and Cinebench wins, while the 9754's additional cores drive its throughput advantages in data-oriented PassMark tests.

Both chips use the same socket (AMD Socket SP5), support DDR5 memory over a twelve-channel bus, and offer 460.8 GB/s of memory bandwidth. Both enable ECC memory. PCIe support is identical: Gen 5 with 128 lanes (CPU only). Neither includes integrated graphics. Both are server/workstation parts, both are currently active in production, and both released on the same date.

Specification Differences

The core and thread counts are the first major divergence: 128 cores and 256 threads for the 9754 versus 96 cores and 192 threads for the 9684X. Clock speeds follow the core count in inverse fashion. The 9684X runs at 2.55 GHz base and 3.70 GHz boost, while the 9754 runs at 2.25 GHz base and 3.10 GHz boost. The 9684X's higher clocks contribute directly to its single-thread victories.

Thermal design power differs by 40 watts. The 9684X draws 400 W TDP, the 9754 draws 360 W TDP. Both use the same socket and same memory configuration, so platform-level compatibility is unchanged. The L3 cache is the other headline difference: 1152 MB for the 9684X versus 256 MB for the 9754. Total transistor count and die configuration also differ, as noted above. The 9684X carries 135,240 million transistors across 12 dies; the 9754 carries 71,000 million across 8 dies.

The launch MSRP for the 9754 is $11900. The launch MSRP for the 9684X is $14756. The 9684X commands a higher launch price, consistent with its larger cache and higher clock speeds. Both processors are multiplier-unlocked false, meaning overclocking is not supported. Part numbers differ: 100-000001234 for the 9754 and 100-100000892 for the 9684X.

The Verdict

The data points to a clear division of labor. The AMD EPYC 9684X is the stronger general-purpose compute processor. It wins every single-thread benchmark in the database, every Cinebench test, and the PassMark multithread test. Its physics simulation score is dramatically higher (24686 versus 8793, a -64.4% delta), and its prime number finding result is exceptional (2020 versus 604, -70.1%). For workloads that reward high clock speeds, large caches, and strong per-core performance, the 9684X is the choice.

The AMD EPYC 9754 is the throughput specialist. It wins the data compression, encryption, extended instruction, floating-point math, and integer math benchmarks. Its 128 cores give it a raw parallelism advantage that the 9684X cannot match in these specific workloads. The 9754 also holds the higher average benchmark score (364371 versus 266914) and sits in the 100th percentile versus all CPUs, while the 9684X sits in the 99th. The 9754's nearest rivals include the Intel Xeon 6960P (-0.2% delta) and AMD EPYC 9655 (-2.4% delta), placing it in a tight competitive cluster. The 9684X's nearest rivals include the AMD Ryzen Threadripper 9970X (-4.6% delta) and Intel Xeon 6780E (-4.8% delta).

For a buyer deciding between these two, the question is workload shape. If the application is database-heavy, compression-heavy, or requires massive parallel throughput, the 9754's extra cores and higher average score matter. If the application is simulation-heavy, single-thread sensitive, or cache-hungry, the 9684X's L3 cache and clock advantage will show up in most benchmarks.

Where Each One Wins

The AMD EPYC 9754 wins in five recorded benchmark categories: PassMark data compression (31.8% ahead), data encryption (29.9% ahead), extended instructions (26.1% ahead), floating-point math (24.6% ahead), and integer math (21.6% ahead). These are workloads where the 9754's 128 cores can be fed continuously, and the results suggest that memory bandwidth and core count matter more than cache capacity. The 9754's 460.8 GB/s memory bandwidth matches the 9684X's exactly, so the advantage must come from the additional 32 cores. Its 256 MB L3 cache is smaller, but the workloads in question appear to tolerate that limitation.

The AMD EPYC 9684X wins in twelve recorded benchmark categories: all six Cinebench tests, PassMark multithread, PassMark physics, PassMark find prime numbers, PassMark random string sorting, and both PassMark single-thread entries. The Cinebench sweep is uniform at -18.8% across every test, indicating a consistent per-core performance edge. The physics result stands out: 24686 versus 8793, which is a 64.4% advantage. Prime number finding shows the 9684X at 2020 versus 604, a 70.1% advantage. These are cache-sensitive and latency-sensitive workloads where the 1152 MB L3 cache provides a decisive benefit.

The single-thread results deserve emphasis. The 9684X leads by -19.5% in PassMark single-thread and -18.8% in every Cinebench single-core test. The higher boost clock (3.70 GHz versus 3.10 GHz) and the larger L3 cache both contribute. The 9684X's 96 cores are fewer, but each core performs better on an individual basis.

The 9754's wins are larger in percentage terms than most of the 9684X's wins, but they cover fewer categories. The 9684X's wins include the entire Cinebench suite and all single-thread tests, which are the most commonly cited metrics for general compute performance. The 9754's wins are confined to PassMark's specialized throughput tests. The verdict from the recorded data: the 9684X for broad compute workloads, the 9754 for high-core-count data processing. Each processor has a clearly defined territory, and the benchmark data does not blur the line between them.