AMD EPYC 9684X vs Intel Xeon 6741P Comparison

The AMD EPYC 9684X and the Intel Xeon 6741P occupy the same server and workstation segment, both sitting in the 99th percentile of all CPUs in the database, but the recorded data shows they compete at very different points on the throughput curve. The EPYC 9684X, a 96-core Zen 4 "Genoa-X" part, wins 15 of the 17 head-to-head benchmarks, including every multi-core test in the suite. The Xeon 6741P, a 48-core Granite Rapids design, takes the two single-thread Passmark tests and offers a lower thermal envelope, a newer platform, and more PCI Express lanes. This analysis walks through where each chip wins, why the architecture produces those results, and which buyer profile each one fits.

Where Each One Wins

The EPYC 9684X dominates every workload that scales across threads. Its average benchmark score of 266914 sits well clear of the Xeon's 194901, a gap of roughly 37 percent in aggregate. The margin is not uniform, though, and that is where the use-case split becomes clear.

In the Cinebench suite, a proxy for sustained all-core rendering and compute, the EPYC leads by 20.8 percent across the board: R15 multi-core (10418 vs 8624), R20 multi-core (43409 vs 35935), and R23 multi-core (103355 vs 85561). Notably, it also wins every Cinebench single-core test by the same 20.8 percent, which means its per-core throughput under that workload is not a compromise.

The Passmark workloads reveal where the gap widens dramatically. Integer math (844145 vs 458058, an 84.3 percent lead), physics (24686 vs 13890, 77.7 percent ahead), encryption (178453 vs 89746, 98.8 percent ahead), and string sorting (349126 vs 177322, 96.9 percent ahead) all favor the EPYC by margins far exceeding its two-to-one core-count ratio would suggest on its own. Data compression follows the same pattern at 48.6 percent. These results indicate that the EPYC's large shared L3 cache, 1152 MB versus 288 MB, is doing real work keeping data close to the cores, amplifying throughput beyond raw core counts.

The Xeon 6741P wins exactly two tests, but they are the same measurement recorded twice: Passmark single-thread, where it scores 3195 against the EPYC's 2891, a 9.5 percent advantage. For latency-sensitive, serial workloads, licensed software that counts cores, or lightly threaded services, this is the one column where the Intel part leads. Its 3.80 GHz boost clock edges the EPYC's 3.70 GHz, and the recorded single-thread data reflects that.

Architecture Differences

Both CPUs are built on 5 nm-class processes, but by different fabs with different packaging strategies. The EPYC 9684X is a TSMC 5 nm chiplet design, codenamed Genoa-X, with a stated 135,240 million transistors spread across 12 dies of 72 mm² each. The Xeon 6741P is an Intel 5 nm product, codenamed Granite Rapids, using two dies of 598 mm² each. The two approaches split the same node class into very different silicon layouts: many small chiplets on one side, two large dies on the other.

Core topology is the biggest differentiator. The EPYC carries 96 cores and 192 threads at a 2.55 GHz base clock, 3.70 GHz boost, and a 400 W TDP on Socket SP5. The Xeon carries 48 cores and 96 threads at a 2.50 GHz base, 3.80 GHz boost, and a 300 W TDP on Socket 4710. The Xeon's per-core cache allocation is larger at the first two levels, 112 KB of L1 and 2 MB of L2 per core versus the EPYC's 64 KB and 1 MB, but the EPYC's shared L3 pool of 1152 MB dwarfs the Xeon's 288 MB. That cache disparity is the clearest architectural explanation for the outsized Passmark leads in memory-hungry integer and sorting workloads.

Memory and I/O also diverge. The EPYC uses a twelve-channel DDR5 bus delivering 460.8 GB/s of bandwidth, while the Xeon uses an eight-channel DDR5 bus at 409.6 GB/s. Both support ECC. On expansion, the situation reverses: the Xeon exposes 136 Gen 5 PCI Express lanes against the EPYC's 128, which matters for GPU-dense or accelerator-heavy deployments. Neither part ships with integrated graphics, both are multiplier-locked, and both remain in active production. Release timing differs by platform generation, with the EPYC arriving June 2023 and the Xeon in February 2025.

The Verdict

The data points to a straightforward split. Buyers whose workloads scale with threads, integer throughput, encryption, compression, physics simulation, or large in-memory sorting, should pick the EPYC 9684X. It wins 15 of 17 recorded tests, posts an average benchmark score 37 percent higher, and its four largest margins (98.8, 96.9, 84.3, and 77.7 percent) land in exactly the categories that define high-throughput server computing. Its 460.8 GB/s memory bandwidth reinforces that profile.

The Xeon 6741P is the pick where the constraint is not throughput. It delivers the higher single-thread score (3195 vs 2891), does so within a 300 W envelope versus the EPYC's 400 W, and offers eight more PCI Express lanes for expansion. Its competitive position is also solid within its own bracket: the database shows it essentially tied with the AMD EPYC 9335 (0.3 percent) and Intel Xeon 678X (0.7 percent), and ahead of the Xeon 6740E by 3.8 percent and the EPYC 8534P by 5.3 percent. For core-limited licensing or per-core software scenarios, half the cores with the faster serial thread is a rational trade.

One positioning note the data makes unavoidable: the EPYC 9684X benchmarks within a tight band of far newer flagship parts, trailing the Threadripper 9970X by only 4.6 percent and the Xeon 6780E by 4.8 percent, while beating the Xeon 6980P by 6.1 percent. The Xeon 6741P, by contrast, sits in a lower performance tier against its own rivals. Its launch MSRP was $4421; the EPYC 9684X launched at $14756.

FAQ

Q: Which CPU is faster overall? A: The AMD EPYC 9684X. It wins 15 of 17 head-to-head benchmarks and posts an average score of 266914 versus 194901 for the Xeon 6741P, a gap of roughly 37 percent.

Q: Is the Xeon 6741P faster in any workload? A: Yes. It wins Passmark single-thread with 3195 points against 2891, a 9.5 percent advantage, and it holds the higher boost clock at 3.80 GHz versus 3.70 GHz.

Q: How much cache does each CPU have? A: The EPYC 9684X has 64 KB of L1 and 1 MB of L2 per core plus 1152 MB of shared L3. The Xeon 6741P has 112 KB of L1 and 2 MB of L2 per core plus 288 MB of shared L3.

Q: Which platform has more memory bandwidth? A: The EPYC. Its twelve-channel DDR5 controller delivers 460.8 GB/s, while the Xeon's eight-channel controller delivers 409.6 GB/s.

Q: Which CPU has more PCI Express lanes? A: The Xeon 6741P, with 136 Gen 5 lanes versus 128 Gen 5 lanes on the EPYC 9684X.

Q: Do both CPUs support ECC memory? A: Yes, both support DDR5 with ECC.

Head-to-Head Benchmarks

The Cinebench results are remarkably consistent: the EPYC 9684X wins every one of the six Cinebench tests by exactly 20.8 percent. R23 multi-core stands at 103355 versus 85561, R20 multi-core at 43409 versus 35935, and R15 multi-core at 10418 versus 8624. The single-core Cinebench results mirror that margin, 14591 vs 12079 in R23 and 6128 vs 5073 in R20, showing the EPYC does not sacrifice per-core rendering speed to get its throughput.

The Passmark suite is where the spread opens up. Encryption is the EPYC's biggest win at 98.8 percent (178453 vs 89746), followed closely by string sorting at 96.9 percent (349126 vs 177322) and integer math at 84.3 percent (844145 vs 458058). Physics lands at 77.7 percent in the EPYC's favor (24686 vs 13890), prime number finding at 62.6 percent (2020 vs 1242), compression at 48.6 percent (2698807 vs 1816408), floating point math at 31.7 percent (472174 vs 358423), and extended instructions at 24.7 percent (177956 vs 142682). The overall Passmark multithread score finishes at 121595 versus 100660, a 20.8 percent EPYC win.

The Xeon's lone victory arrives in Passmark single-thread: 3195 to 2891, a 9.5 percent edge. That result, combined with its lower TDP and higher lane count, defines the narrow but genuine use case where it is the better fit.

Specification Differences

• Cores/Threads: EPYC 9684X has 96 cores and 192 threads; Xeon 6741P has 48 cores and 96 threads.
• Clocks: EPYC base 2.55 GHz, boost 3.70 GHz; Xeon base 2.50 GHz, boost 3.80 GHz.
• TDP: 400 W for the EPYC, 300 W for the Xeon.
• Socket: AMD Socket SP5 versus Intel Socket 4710.
• Architecture: Zen 4 (Genoa-X) versus Granite Rapids (Granite Rapids-SP).
• Foundry and dies: TSMC 5 nm with 12 dies of 72 mm² and 135,240 million transistors versus Intel 5 nm with 2 dies of 598 mm².
• Cache: 64 KB L1 and 1 MB L2 per core with 1152 MB shared L3 (EPYC) versus 112 KB L1 and 2 MB L2 per core with 288 MB shared L3 (Xeon).
• Memory bus: Twelve-channel at 460.8 GB/s (EPYC) versus eight-channel at 409.6 GB/s (Xeon); both DDR5 with ECC.
• PCIe: 128 Gen 5 lanes (EPYC) versus 136 Gen 5 lanes (Xeon).
• Release date: June 2023 for the EPYC, February 2025 for the Xeon.
• Part numbers: 100-100000892 (EPYC) and SRVEY (Xeon).

Fields where the two are identical: process node class (5 nm), market segment (Server/Workstation), ECC support, lack of integrated graphics, locked multipliers, and active production status.