CPU Comparison

The AMD EPYC 9115 and AMD EPYC 7352 represent two distinct generations of AMD’s server strategy, with the former built on the modern Zen 5 architecture and the latter on the older Zen 2 design. Despite the EPYC 7352 offering 24 cores and 48 threads compared to the EPYC 9115’s 16 cores and 32 threads, the benchmark data reveals a decisive performance shift in favor of the newer chip. Across 17 head-to-head comparisons, the EPYC 9115 secures 14 wins, with the EPYC 7352 managing only 3 victories in specific workload niches. This analysis breaks down the benchmark results, architectural differences, and the specific use cases where each processor excels, based strictly on the provided data.

Head-to-Head Benchmarks

The EPYC 9115 dominates the multi-threaded rendering tests, establishing a consistent performance lead across all Cinebench versions. In Cinebench R15 multicore, the EPYC 9115 scores 4233 against the EPYC 7352’s 3458, a 22.4% advantage. This pattern repeats in Cinebench R20 multicore (17641 vs 14411, +22.4%) and Cinebench R23 multicore (42003 vs 34314, +22.4%). The consistency of the 22.4% delta across these tests indicates a fundamental throughput advantage for the Zen 5 chip, likely stemming from its superior per-core efficiency rather than raw core count.

Single-core performance shows an even more pronounced gap. In Cinebench R15 single-core, the EPYC 9115 scores 597 versus 488 for the EPYC 7352, a 22.3% lead. The delta remains at 22.4% in Cinebench R20 (2490 vs 2034) and R23 single-core (5929 vs 4844). However, the Passmark single-thread test reveals the largest single-core disparity: the EPYC 9115 posts 3360 against 1979 for the EPYC 7352, a massive 69.8% improvement. This suggests that the older Zen 2 architecture’s single-thread performance is severely outclassed, making the EPYC 9115 the clear choice for latency-sensitive or lightly threaded server workloads.

In Passmark’s math and physics tests, the EPYC 9115 continues its winning streak. Floating-point math scores 113853 for the EPYC 9115 versus 87969 for the EPYC 7352, a 29.4% advantage, while integer math shows a 22.3% lead (181807 vs 148605). The physics test is particularly lopsided: the EPYC 9115 scores 4188 against 2688, a 55.8% difference. Extended instructions also favor the newer chip, with a 13.1% lead (45477 vs 40203). Even in random string sorting, where the results are closer, the EPYC 9115 wins by 1.3% (70151 vs 69231). The overall Passmark multithread score confirms the trend: 48936 for the EPYC 9115 versus 40370 for the EPYC 7352, a 21.2% advantage despite the EPYC 7352 having 50% more cores.

The EPYC 7352’s three wins are concentrated in specific memory and encryption workloads. Its biggest victory comes in data encryption, where it scores 44426 against 33489 for the EPYC 9115, a 24.6% margin. Data compression also favors the older chip: 660712 versus 600099, a 9.2% advantage. Finally, the EPYC 7352 edges out the EPYC 9115 in the find prime numbers test, scoring 301 versus 289, a 4% difference. These results indicate that the EPYC 7352 retains some advantage in workloads that may benefit from its higher core count (24 vs 16) and its larger total L3 cache of 128 MB, compared to the EPYC 9115’s 64 MB shared L3.

Architecture Differences

The two processors are built on fundamentally different process nodes and architectures. The EPYC 9115 uses TSMC’s 4 nm process node and features the Zen 5 architecture under the codename "Turin," part of the EPYC 9005 series. In contrast, the EPYC 7352 is fabricated on TSMC’s 7 nm node with the Zen 2 architecture, codenamed "Rome," from the EPYC 7002 series. The transistor counts are similar, 16,630 million for the EPYC 9115 versus 15,200 million for the EPYC 7352, but the die layout differs significantly. The EPYC 9115 uses a 2x 70.6 mm² die configuration, while the EPYC 7352 employs a 4x 74 mm² setup.

Cache hierarchies also diverge sharply. The EPYC 9115 provides 80 KB of L1 cache per core and 1 MB of L2 per core, with a shared 64 MB L3. The EPYC 7352 offers 96 KB of L1 per core and 512 KB of L2 per core, but its L3 is organized as 32 MB per die, totaling 128 MB. This means the older chip has double the total L3 cache, which likely explains its wins in data compression and encryption, where large datasets may fit into the faster L3.

Memory support is another generational leap. The EPYC 9115 supports DDR5 memory across a twelve-channel bus, delivering a memory bandwidth of 576.0 GB/s. The EPYC 7352 is limited to DDR4 on an eight-channel bus, providing 204.8 GB/s. This nearly threefold increase in bandwidth for the EPYC 9115 is critical for memory-intensive workloads. PCIe connectivity also improves: the EPYC 9115 offers Gen 5 with 128 lanes, while the EPYC 7352 provides Gen 4 with the same 128 lanes. Both processors support ECC memory, and neither has integrated graphics.

Clock speeds favor the newer chip, with the EPYC 9115 running at a base clock of 2.60 GHz and a boost clock of 4.10 GHz, compared to the EPYC 7352’s 2.30 GHz base and 3.20 GHz boost. The EPYC 9115 also has a lower TDP of 125 watts versus 155 watts for the EPYC 7352, despite being the faster processor in most tests. The sockets differ as well: the EPYC 9115 uses AMD Socket SP5, while the EPYC 7352 uses the older AMD Socket SP3. The EPYC 9115 launched on 2024-10-09, while the EPYC 7352 is a much older design, released on 2019-08-06.

The Verdict

The data makes a compelling case for the EPYC 9115 in nearly every general-purpose server role. Its 22.4% lead across all Cinebench multicore tests and 69.8% advantage in single-thread performance make it the superior choice for virtualization, database serving, and high-frequency trading where per-core speed matters. The 55.8% delta in Passmark physics and 29.4% lead in floating-point math further cement its position for scientific computing and simulation workloads. The EPYC 9115 achieves all this with a lower TDP (125W vs 155W) and faster memory bandwidth (576.0 GB/s vs 204.8 GB/s), making it the more efficient and capable part.

However, the EPYC 7352 is not without a niche. Its 24.6% lead in data encryption and 9.2% lead in data compression suggest that workloads heavily dependent on cryptographic operations or lossless compression algorithms may still favor the older chip, likely due to its 128 MB total L3 cache. For servers dedicated solely to these tasks, the EPYC 7352 remains a viable option. Additionally, its higher core count (24 vs 16) could appeal to environments where raw parallel thread availability is more important than IPC improvements, though the benchmark results show the EPYC 9115 still wins the multithreaded Passmark test by 21.2%, indicating that core count alone does not guarantee superiority.

For most buyers, the EPYC 9115 is the clear winner based on performance metrics alone. It is faster in 14 of 17 benchmarks, offers newer DDR5 and PCIe Gen 5 support, and does so at a lower TDP. The EPYC 7352 should only be considered for specialized encryption or compression workloads where its cache advantage translates to measurable wins. The data does not support choosing the EPYC 7352 for general-purpose computing, as its single-thread and multi-thread performance deficits are too large to ignore.

Specification Differences

• Cores: 16 (EPYC 9115) vs 24 (EPYC 7352)
• Threads: 32 (EPYC 9115) vs 48 (EPYC 7352)
• Base Clock: 2.60 GHz (EPYC 9115) vs 2.30 GHz (EPYC 7352)
• Boost Clock: 4.10 GHz (EPYC 9115) vs 3.20 GHz (EPYC 7352)
• TDP: 125 W (EPYC 9115) vs 155 W (EPYC 7352)
• Socket: AMD Socket SP5 (EPYC 9115) vs AMD Socket SP3 (EPYC 7352)
• Architecture: Zen 5 (EPYC 9115) vs Zen 2 (EPYC 7352)
• Process Node: 4 nm (EPYC 9115) vs 7 nm (EPYC 7352)
• Transistors: 16,630 million (EPYC 9115) vs 15,200 million (EPYC 7352)
• Die Size: 2x 70.6 mm² (EPYC 9115) vs 4x 74 mm² (EPYC 7352)
• L1 Cache: 80 KB per core (EPYC 9115) vs 96 KB per core (EPYC 7352)
• L2 Cache: 1 MB per core (EPYC 9115) vs 512 KB per core (EPYC 7352)
• L3 Cache: 64 MB shared (EPYC 9115) vs 32 MB per die, totaling 128 MB (EPYC 7352)
• Memory Support: DDR5 (EPYC 9115) vs DDR4 (EPYC 7352)
• Memory Bus: Twelve-channel (EPYC 9115) vs Eight-channel (EPYC 7352)
• Memory Bandwidth: 576.0 GB/s (EPYC 9115) vs 204.8 GB/s (EPYC 7352)
• PCIe: Gen 5, 128 Lanes (EPYC 9115) vs Gen 4, 128 Lanes (EPYC 7352)
• Release Date: 2024-10-09 (EPYC 9115) vs 2019-08-06 (EPYC 7352)
• Launch MSRP: $726 (EPYC 9115) vs $1350 (EPYC 7352)

FAQ

Q: Which processor is faster in multi-threaded workloads?

A: The AMD EPYC 9115 wins all three Cinebench multicore tests by 22.4%, including a score of 42003 in R23 compared to 34314 for the EPYC 7352, and also leads Passmark multithread by 21.2% (48936 vs 40370).

Q: The EPYC 7352 has more cores. Why does it lose in most tests?

A: While the EPYC 7352 has 24 cores versus 16 for the EPYC 9115, the newer Zen 5 architecture and 4 nm process node deliver significantly higher per-core performance. The EPYC 9115’s single-thread score is 69.8% higher (3360 vs 1979), which compensates for the core deficit.

Q: In which workloads does the EPYC 7352 outperform the EPYC 9115?

A: The EPYC 7352 wins in Passmark data encryption (44426 vs 33489, a 24.6% lead), data compression (660712 vs 600099, a 9.2% lead), and find prime numbers (301 vs 289, a 4% lead). These wins likely relate to its larger 128 MB total L3 cache.

Q: How do memory specifications differ between the two?

A: The EPYC 9115 supports DDR5 on a twelve-channel bus with 576.0 GB/s bandwidth, while the EPYC 7352 uses DDR4 on an eight-channel bus with 204.8 GB/s. This gives the EPYC 9115 nearly three times the theoretical memory bandwidth.

Q: What is the difference in power consumption?

A: The EPYC 9115 has a TDP of 125 watts, which is lower than the EPYC 7352’s 155 watts, despite the EPYC 9115 being faster in most benchmarks.

Q: Are these processors from the same generation?

A: No. The EPYC 9115 is from the EPYC 9005 series (Zen 5, codename Turin), released on 2024-10-09, while the EPYC 7352 is from the EPYC 7002 series (Zen 2, codename Rome), released on 2019-08-06.