AMD EPYC 9655 vs Intel Xeon 696X Comparison

Where Each One Wins

The benchmark data is unambiguous: the AMD EPYC 9655 wins every single head-to-head test recorded in the database, taking all 14 comparisons. The Intel Xeon 696X does not secure a single win across the entire suite. This is not a close contest where each chip has its preferred workloads; the AMD part dominates across the board.

That said, the degree of dominance varies enormously by workload type. The EPYC 9655 is at its absolute best in integer-heavy and physics-based tasks. The PassMark integer math test shows a 99.1% advantage, while the physics test shows a staggering 667.2% lead. These are the kinds of workloads where the EPYC 9655 simply runs away from the Xeon 696X. For database operations, scientific computing, and any workload that hammer on integer arithmetic, the AMD chip is the clear choice.

The Xeon 696X comes closest in single-threaded performance. The PassMark single-thread test shows the EPYC 9655 ahead by only 2.8%. This is the one area where the gap narrows to a rounding error. If a workload is purely single-threaded and lightly threaded, the two chips are effectively interchangeable. The Xeon 696X also holds up reasonably in extended instruction workloads, where the EPYC 9655 leads by 17.5%. Still a win for AMD, but not the blowout seen elsewhere.

For memory-intensive and compression tasks, the EPYC 9655 maintains a substantial but not extreme lead. Data compression shows a 44.5% advantage, floating point math a 47.3% lead. These are meaningful differences that will translate to real-world throughput gains in database compression, scientific simulations, and financial modeling.

In encryption and prime number finding, the EPYC 9655 nearly doubles the Xeon 696X. Data encryption shows an 87.1% lead, and prime numbers an 87.3% lead. These results point to strong cryptographic and computational number theory performance on the AMD side.

The overall average benchmark score tells the same story. The EPYC 9655 has an average score of 373,479, while the Xeon 696X sits at 286,102. The AMD chip also ranks at the 100th percentile of all CPUs in the database, versus the 99th percentile for the Intel part. The Xeon 696X does land near the AMD EPYC 9565 in its nearest rivals list, with a 0.2% delta, so it is competitive with the previous-generation AMD parts, just not this one.

Architecture Differences

The architectural gap between these two chips is significant and explains the benchmark results. The AMD EPYC 9655 uses the Zen 5 architecture under the Turin codename, part of the EPYC 9005 series. It is built on a 4 nm process at TSMC, with a reported transistor count of 99,780 million spread across 12 chiplets, each measuring 70.6 mm².

The Intel Xeon 696X uses the Granite Rapids architecture, part of the Xeon 600 generation. It is built on a 5 nm process at Intel, with a die size of 2x 598 mm². Intel does not report a transistor count in the database.

Core counts differ substantially. The EPYC 9655 has 96 cores and 192 threads. The Xeon 696X has 64 cores and 128 threads. That 50% core advantage for AMD is the primary driver of the multicore benchmark results.

Cache configurations also differ. The EPYC 9655 has 80 KB of L1 cache per core, 1 MB of L2 per core, and 384 MB of shared L3 cache. The Xeon 696X has 112 KB of L1 per core, 2 MB of L2 per core, and 336 MB of shared L3. Intel gives each core more L1 and L2, but AMD has more total L3.

Memory architecture is another major split. The EPYC 9655 supports twelve-channel DDR5 memory with a peak bandwidth of 576.0 GB/s. The Xeon 696X supports eight-channel DDR5 with 409.6 GB/s bandwidth. That is a 40.6% bandwidth advantage for AMD, which directly feeds its strong showing in memory-sensitive workloads.

Both chips support PCIe Gen 5 with 128 lanes (CPU only), and both have no integrated graphics. Clock speeds are close: the EPYC 9655 bases at 2.60 GHz and boosts to 4.50 GHz, while the Xeon 696X bases at 2.40 GHz and boosts to 4.80 GHz. Intel has the higher boost clock by 0.30 GHz, but AMD has the higher base clock by 0.20 GHz.

The Xeon 696X has an unlocked multiplier, while the EPYC 9655 does not. Power targets differ, with the EPYC 9655 rated at 400 and the Xeon 696X at 350, but the database does not specify units beyond the raw number.

Head-to-Head Benchmarks

The most lopsided result in the entire comparison is the PassMark physics test. The EPYC 9655 scores 25,947 against 3,382 for the Xeon 696X, a 667.2% advantage. This is the single largest gap in the dataset and suggests the AMD chip handles physics simulation workloads at a fundamentally different level.

Integer math is nearly as one-sided. The EPYC 9655 scores 1,139,161 versus 572,072, a 99.1% lead. This essentially doubles the Intel part's integer throughput and explains why database and transactional workloads would favor the AMD chip.

Random string sorting shows a 143.7% advantage for AMD, with scores of 439,682 versus 180,392. This is a memory-latency-sensitive workload, and the twelve-channel memory controller likely plays a major role here.

The encryption and prime number tests both show roughly 87% leads. Encryption scores 210,555 versus 112,529, while prime numbers score 1,598 versus 853. These workloads benefit from the higher core count and the wider memory bus.

The Cinebench results are consistent across versions. R15, R20, and R23 multicore tests all show the EPYC 9655 ahead by exactly 48.7%. The scores are 13,373 versus 8,994 for R15, 55,722 versus 37,475 for R20, and 132,672 versus 89,227 for R23. This consistency across Cinebench versions indicates the multicore advantage is structural, not test-specific.

Floating point math shows a 47.3% lead for AMD, with scores of 662,958 versus 450,164. Data compression shows a 44.5% lead, with 3,271,896 versus 2,264,907. Multithreaded performance shows a 48.7% lead, with 156,110 versus 104,974.

The closest result is single-thread performance. The EPYC 9655 scores 3,847 versus 3,742, a 2.8% lead. This is the only test where the two chips are in the same ballpark. The extended instructions test is the next closest, with AMD ahead 17.5% at 203,285 versus 172,975.

FAQ

Q: Which CPU is better for heavily threaded workloads?

A: The AMD EPYC 9655. It wins every multicore benchmark in the database, with a 48.7% lead across all three Cinebench versions and the PassMark multithread test.

Q: Is the Intel Xeon 696X competitive in single-threaded tasks?

A: Yes, nearly so. The EPYC 9655 leads by only 2.8% in the PassMark single-thread test, with scores of 3,847 versus 3,742. For lightly threaded workloads, the two chips are effectively equivalent.

Q: How does memory bandwidth compare between the two?

A: The EPYC 9655 supports twelve-channel DDR5 with 576.0 GB/s bandwidth, while the Xeon 696X supports eight-channel DDR5 with 409.6 GB/s. That is a 40.6% bandwidth advantage for AMD.

Q: What is the core and thread count difference?

A: The EPYC 9655 has 96 cores and 192 threads. The Xeon 696X has 64 cores and 128 threads. AMD has 50% more cores and threads.

Q: Which chip has more L3 cache?

A: The EPYC 9655 has 384 MB of shared L3 cache. The Xeon 696X has 336 MB of shared L3 cache. AMD has 48 MB more L3.

Q: Are both CPUs unlocked for overclocking?

A: No. The Xeon 696X has an unlocked multiplier, while the EPYC 9655 does not. This is relevant for users who plan to manually adjust clock speeds.

The Verdict

The data is decisive. The AMD EPYC 9655 wins all 14 recorded head-to-head benchmarks, with an average benchmark score of 373,479 against 286,102 for the Intel Xeon 696X. The EPYC 9655 also holds the 100th percentile ranking among all CPUs in the database, while the Xeon 696X sits at the 99th percentile.

For anyone building a server or workstation where multicore throughput is the priority, the EPYC 9655 is the clear choice. It delivers roughly 49% better multicore performance across Cinebench and PassMark multithread tests, and it nearly doubles the Intel part in integer math, encryption, and prime number workloads. The twelve-channel memory interface provides 40.6% more bandwidth, which feeds its strong results in memory-sensitive tasks like compression and random string sorting.

The Xeon 696X is not without merit. Its single-thread performance is within 2.8% of the EPYC 9655, and it offers a higher boost clock at 4.80 GHz versus 4.50 GHz. It also has an unlocked multiplier for users who want manual tuning, and its 2x 598 mm² die size indicates a substantial physical implementation. But in every measured workload, it trails the AMD part, often by a wide margin.

The nearest rival data contextualizes the Xeon 696X further. It sits within 0.2% of the AMD EPYC 9565 and within 0.3% of the AMD EPYC 9555P, and it is 2% ahead of the Intel Xeon 6780E. This places the Xeon 696X as a solid mid-pack performer, competitive with the previous generation of EPYC parts but clearly behind the EPYC 9655.

The verdict is straightforward. Choose the AMD EPYC 9655 for maximum throughput, memory bandwidth, and overall benchmark dominance. Choose the Intel Xeon 696X only if single-threaded performance is the sole concern, if an unlocked multiplier is essential, or if the lower launch MSRP is a deciding factor. The Xeon 696X has a launch MSRP of $5,599, while the EPYC 9655 has a launch MSRP of $11,852. The AMD part costs more but delivers substantially more performance in nearly every measurable way.

Specification Differences

| Specification | AMD EPYC 9655 | Intel Xeon 696X |

|---|---|---|

| Cores | 96 | 64 |

| Threads | 192 | 128 |

| Base Clock | 2.60 GHz | 2.40 GHz |

| Boost Clock | 4.50 GHz | 4.80 GHz |

| TDP | 400 | 350 |

| Socket | AMD Socket SP5 | Intel Socket 4710 |

| Architecture | Zen 5 | Granite Rapids |

| Codename | Turin | Granite Rapids |

| Process Node | 4 nm | 5 nm |

| Foundry | TSMC | Intel |

| Die Size | 12x 70.6 mm² | 2x 598 mm² |

| Transistors | 99,780 million | Not reported |

| L1 Cache | 80 KB (per core) | 112 KB (per core) |

| L2 Cache | 1 MB (per core) | 2 MB (per core) |

| L3 Cache | 384 MB (shared) | 336 MB (shared) |

| Memory Bus | Twelve-channel | Eight-channel |

| Memory Bandwidth | 576.0 GB/s | 409.6 GB/s |

| PCIe | Gen 5, 128 Lanes (CPU only) | Gen 5, 128 Lanes (CPU only) |

| Unlocked Multiplier | No | Yes |

| Launch MSRP | $11,852 | $5,599 |

| Release Date | 2024-10-09 | 2026-02-01 |

| Part Number | 100-000000674 | SRWQ7 |