AMD EPYC 9535 vs AMD EPYC 9684X Comparison

The AMD EPYC 9535 and AMD EPYC 9684X are both server processors for the AMD Socket SP5 platform, but they come from different generations and are built for different workload profiles. The 9535 is a Zen 5 (Turin) part from the EPYC 9005 series with 64 cores and 128 threads, while the 9684X is a Zen 4 (Genoa-X) part from the EPYC 9004 series with 96 cores and 192 threads. Their recorded benchmark results show a clear split: the 9684X dominates multi-threaded and cache-heavy tests, while the 9535 takes single-thread performance by a wide margin.

Head-to-Head Benchmarks

The head-to-head data covers eleven PassMark tests. The 9684X wins nine of them, and the 9535 wins two (the two identical single-thread entries). The largest victory for the 9684X is in physics, where it scores 24,686 against the 9535's 3,834, an 84.5% lead. Prime number finding is another blowout: 2,020 versus 874, a 56.7% advantage. Random string sorting shows a 29.1% lead (349,126 vs 247,506), and data encryption is 28.6% ahead (178,453 vs 127,372). Data compression goes to the 9684X by 14.5% (2,698,807 vs 2,308,822), integer math by 13.5% (844,145 vs 730,281), multithread by 5.8% (121,595 vs 114,528), floating point math by 3.2% (472,174 vs 457,047), and extended instructions by a slim 1.2% (177,956 vs 175,784). The 9535's only wins are the single-thread tests, where it scores 3,720 against 2,891, a 28.7% margin. This pattern indicates that the 9684X is the stronger choice for heavily parallel and cache-sensitive workloads, while the 9535 offers substantially better per-core performance.

In the database's overall average benchmark score, the 9535 records 379,408, placing it 1.6% above the EPYC 9655, 3.9% above the Xeon 6960P, and 4.1% above the EPYC 9754, but 4.6% below the EPYC 9655P. The 9684X averages 266,914, which is 4.6% below the Threadripper 9970X, 4.8% below the Xeon 6780E, and 6.5% below the EPYC 9565, but 6.1% above the Xeon 6980P.

FAQ

Q: Which processor has more cores and threads?

A: The EPYC 9684X has 96 cores and 192 threads, while the EPYC 9535 has 64 cores and 128 threads.

Q: How does the L3 cache compare?

A: The 9684X has 1152 MB of shared L3 cache, while the 9535 has 256 MB. The 9535 has a larger L1 cache per core (80 KB vs 64 KB), but both have 1 MB L2 per core.

Q: What are the clock speeds?

A: The 9535 has a base clock of 2.40 GHz and a boost of 4.30 GHz. The 9684X has a base of 2.55 GHz and a boost of 3.70 GHz.

Q: Which has higher memory bandwidth?

A: The 9535 has a memory bandwidth of 576.0 GB/s, while the 9684X has 460.8 GB/s. Both support DDR5 with a twelve-channel memory bus.

Q: What are their average benchmark scores and percentiles?

A: The 9535 has an average benchmark score of 379,408 and sits in the 100th percentile of all CPUs. The 9684X averages 266,914 and is in the 99th percentile.

Q: Which processor has the higher TDP?

A: The 9684X has a TDP of 400 W, while the 9535 has a TDP of 300 W.

Architecture Differences

The two processors are built on different microarchitectures and process nodes. The 9535 uses Zen 5 (Turin) on a 4 nm TSMC process, while the 9684X uses Zen 4 (Genoa-X) on a 5 nm TSMC process. The 9684X contains 135,240 million transistors across 12 dies of 72 mm² each, whereas the 9535 has 66,520 million transistors on 8 dies of 70.6 mm². The cache hierarchy also differs significantly. The 9535 offers 80 KB of L1 per core and 256 MB of shared L3. The 9684X has 64 KB of L1 per core but a massive 1152 MB of shared L3, likely to benefit workloads that can exploit large resident datasets. Both have 1 MB of L2 per core. The 9684X's higher transistor count and larger L3 come at the cost of a higher TDP (400 W vs 300 W) and a lower boost clock (3.70 GHz vs 4.30 GHz). These architectural differences explain the benchmark split: the 9684X's extra cores and huge cache drive its wins in parallel and cache-sensitive tests, while the 9535's newer Zen 5 design and higher boost clock deliver superior single-thread performance.

Specification Differences

The following specifications differ between the two processors:

• Cores: 64 (9535) vs 96 (9684X)
• Threads: 128 vs 192
• Base clock: 2.40 GHz vs 2.55 GHz
• Boost clock: 4.30 GHz vs 3.70 GHz
• TDP: 300 W vs 400 W
• Process node: 4 nm vs 5 nm
• Architecture: Zen 5 vs Zen 4
• Codename: Turin vs Genoa-X
• Transistors: 66,520 million vs 135,240 million
• Die size: 8x 70.6 mm² vs 12x 72 mm²
• L1 cache per core: 80 KB vs 64 KB
• L3 cache: 256 MB vs 1152 MB
• Memory bandwidth: 576.0 GB/s vs 460.8 GB/s
• Release date: 2024-10-09 vs 2023-06-12
• Launch MSRP: $8992 for the 9535, $14756 for the 9684X
• Part number: 100-000001147 vs 100-100000892

Both processors share the same socket (AMD Socket SP5), twelve-channel DDR5 memory support, ECC memory, 128 PCIe Gen 5 lanes, and lack integrated graphics. They also have the same production status (Active) and market segment (Server/Workstation).

The Verdict

The data points to two distinct roles. The EPYC 9535 is the better choice for workloads that depend on single-thread performance, low latency, and higher clock speeds. It wins the single-thread tests by 28.7% and carries a higher average benchmark score (379,408 vs 266,914) as well as a higher percentile ranking (100th vs 99th). Its lower TDP (300 W vs 400 W) also makes it easier to cool and power in dense server environments. The EPYC 9684X, on the other hand, is the superior option for massively parallel, cache-hungry applications. It wins nine of the eleven head-to-head tests, including physics, prime number finding, data compression, and encryption, with margins ranging from 1.2% to 84.5%. Its 96 cores and 1152 MB of L3 cache give it a clear edge in multi-threaded throughput, even though its average benchmark score is lower because that average includes a different set of tests. For a mixed workload that values per-core speed and modern architecture, the 9535 is the pick. For a dedicated compute node that can use every core and every megabyte of cache, the 9684X is the pick.

Where Each One Wins

AMD EPYC 9535 wins: single-thread performance, as shown by its 28.7% lead in PassMark single-thread tests. This makes it suitable for lightly threaded server tasks, database queries that are latency-bound, and any application that cannot scale beyond a few cores. Its higher boost clock (4.30 GHz) and lower TDP (300 W) also make it attractive for environments where power and cooling are limited.

AMD EPYC 9684X wins: all other measured categories. It leads in data compression (14.5%), data encryption (28.6%), extended instructions (1.2%), prime number finding (56.7%), floating point math (3.2%), integer math (13.5%), multithread (5.8%), physics (84.5%), and random string sorting (29.1%). These results indicate strengths in scientific computing, data analytics, compression workloads, and any parallel workload that can take advantage of 96 cores and 1152 MB of L3 cache. The 9684X also posts Cinebench R23 scores of 103,355 multi-core and 14,591 single-core, along with Geekbench scores of 15,668 multi-core and 1,586 single-core, further confirming its multi-threaded and single-threaded capabilities in those specific tests.