AMD EPYC 9845 vs Intel Xeon 6780E Comparison

The AMD EPYC 9845 and Intel Xeon 6780E are both flagship-class server processors, but the recorded data shows this is not a close contest. The EPYC 9845, AMD's Zen 5-based Turin part, swept all sixteen benchmarks in the head-to-head set, winning by margins ranging from roughly 54 percent to over 176 percent. The Xeon 6780E, built on Intel's efficiency-focused Sierra Forest architecture, simply plays a different game: it trades raw per-thread speed for density, and the numbers reflect that trade everywhere.

FAQ

Q: Which CPU wins more benchmarks, the EPYC 9845 or the Xeon 6780E?

A: The EPYC 9845 wins all sixteen recorded head-to-head tests. The Xeon 6780E wins none.

Q: How large is the performance gap?

A: It depends on the workload. The smallest recorded gap is in data encryption, where the EPYC 9845 leads by 53.8 percent. The largest is in extended instructions, where the EPYC 9845 more than doubles the Xeon 6780E's score, a 176.1 percent advantage.

Q: Which chip has more cores?

A: The EPYC 9845 has 160 cores and 320 threads, thanks to simultaneous multithreading. The Xeon 6780E has 144 cores and 144 threads, so it runs one thread per core.

Q: Which CPU is more power efficient on paper?

A: The Xeon 6780E carries a lower TDP at 330 watts versus 390 watts for the EPYC 9845, and it does so with fewer cores and no SMT.

Q: How does each chip rank against the full database?

A: The EPYC 9845 sits in the 100th percentile versus all recorded CPUs, with an average benchmark score of 523613. The Xeon 6780E sits in the 99th percentile with an average of 280438.

Q: What was each chip's launch MSRP?

A: The EPYC 9845 launched with an MSRP of $13564. The Xeon 6780E launched at $11350.

Architecture Differences

These two processors embody opposite server philosophies. The EPYC 9845 belongs to the EPYC 9005 series and uses AMD's Zen 5 architecture under the Turin codename, specifically a dense Zen 5c implementation. It is fabricated on TSMC's 3 nm process, currently the finer of the two nodes represented here, and it packs 160 cores with two threads each for 320 total threads. Base clock sits at 2.10 GHz with boosts reaching 3.70 GHz.

The Xeon 6780E is a Sierra Forest-SP part from Xeon 6. Intel manufactures it on its own 5 nm process across a 578 mm² die. Its defining architectural choice is the absence of simultaneous multithreading: 144 cores, 144 threads. Each core runs at a 2.20 GHz base, slightly higher than the EPYC's, but tops out at 3.00 GHz boost, well short of the EPYC's ceiling. That clock structure, combined with the missing second thread per core, explains a great deal of the single-threaded and multi-threaded deficits recorded in the benchmark section.

Cache design also diverges sharply. The EPYC 9845 gives each core 80 KB of L1 and 1 MB of L2, with a large 320 MB shared L3. The Xeon 6780E has 96 KB of L1 per core and 4 MB of L2 per module, but only 108 MB of shared L3, roughly a third of the EPYC's pool. For throughput workloads with large working sets, that L3 disparity compounds the core and clock deficits.

Platform connectivity is another clear split. Both support DDR5 with ECC, but the EPYC 9845 runs a twelve-channel memory bus delivering 576.0 GB/s of bandwidth, versus eight channels and 409.6 GB/s on the Xeon. PCIe is likewise lopsided: 128 Gen 5 lanes on the EPYC against 88 on the Xeon. The sockets differ too, AMD Socket SP5 versus Intel Socket 4710, so platform choice is a one-way door.

Head-to-Head Benchmarks

Every recorded test went to the EPYC 9845, and most were not close.

The Cinebench suite is the cleanest illustration. In Cinebench R15 multi-core, the EPYC 9845 scored 13107 against 7431, a 76.4 percent win. R20 multi-core repeated the pattern: 54615 versus 30963. R23 multi-core landed at 130037 against 73723, again 76.4 percent ahead. Notably, the single-core tests show the same 76.4 percent margin, 1850 to 1049 in R15 and 7710 to 4371 in R20, meaning the EPYC's advantage is not merely a product of extra threads. Even core for core, clock for clock, the Zen 5 part is substantially faster in this workload class.

Passmark's CPU Suite tells a similar story with wider variance. The tightest race of the entire comparison is data encryption: 296808 for the EPYC versus 193004 for the Xeon, still a 53.8 percent gap. Random string sorting (538060 versus 326954) and single-thread performance (3144 versus 1923) land at 64.6 and 63.5 percent respectively. Passmark multithread scored 152985 against 86734, a 76.4 percent margin, and physics reached 19631 versus 10951, 79.3 percent apart.

The heaviest blowouts come in compute-bound integer and floating point work. Integer math ran 1687531 on the EPYC against 641817 on the Xeon, a 162.9 percent difference. Floating point math was nearly as lopsided at 978377 versus 433862, 125.5 percent. Extended instructions was the single worst result for the Xeon 6780E: 114008 against 314798, a 176.1 percent deficit. Data compression (4680013 versus 2557582, 83 percent) and prime number computation (1255 versus 708, 77.3 percent) round out the sweep.

For context within the database, the EPYC 9845's average score of 523613 puts it 3.5 percent ahead of the EPYC 9755, 22.9 percent ahead of the EPYC 9745, and 27.1 percent ahead of the Ryzen Threadripper PRO 9995WX, while trailing the EPYC 9965 by 15.6 percent. The Xeon 6780E's average of 280438 places it in a tight cluster: within 0.2 percent of the Ryzen Threadripper 9970X, 1.8 percent behind the EPYC 9565, 2 percent behind the Xeon 696X, and 2.3 percent behind the EPYC 9555P. In short, the EPYC 9845 competes near the very top of the chart, while the Xeon 6780E competes in a crowded upper tier, and the two do not overlap.

The Verdict

Based strictly on the recorded data, the EPYC 9845 is the performance pick in every measurable category. It is faster in single-threaded work by 63.5 percent, faster in every multi-threaded test by 54 to 176 percent, and it sits in the 100th percentile of the database versus the Xeon's 99th. Workloads dominated by floating point math, integer math, or extended instruction throughput favor it most dramatically.

The case for the Xeon 6780E rests on platform characteristics rather than benchmark scores. It posts a lower TDP, 330 watts versus 390, while still delivering 144 physical cores, and its one-thread-per-core design avoids SMT contention in workloads that prefer dedicated cores. Buyers already standardized on Intel Socket 4710 platforms, or those whose software licensing counts threads, may find that profile relevant. But nothing in the benchmark data favors it: zero wins across sixteen tests.

If maximum throughput, memory bandwidth, and PCIe lanes are the criteria, the EPYC 9845's twelve-channel memory at 576.0 GB/s and 128 Gen 5 lanes make it the clear choice. If lower nominal power with a large physical core count on an Intel platform matters more than raw speed, the Xeon 6780E fits, with the explicit understanding that it measures slower in every recorded test.

Specification Differences

• Cores: 160 (EPYC 9845) vs 144 (Xeon 6780E)
• Threads: 320 vs 144 (SMT on the EPYC, none on the Xeon)
• Base clock: 2.10 GHz vs 2.20 GHz
• Boost clock: 3.70 GHz vs 3.00 GHz
• TDP: 390 W vs 330 W
• Architecture: Zen 5 (Turin) vs Sierra Forest
• Process node: 3 nm (TSMC) vs 5 nm (Intel)
• Die size: not recorded for the EPYC; 578 mm² for the Xeon
• L1 cache: 80 KB per core vs 96 KB per core
• L2 cache: 1 MB per core vs 4 MB per module
• L3 cache: 320 MB shared vs 108 MB shared
• Memory bus: Twelve-channel vs Eight-channel
• Memory bandwidth: 576.0 GB/s vs 409.6 GB/s
• PCIe: Gen 5, 128 lanes vs Gen 5, 88 lanes
• Socket: AMD Socket SP5 vs Intel Socket 4710
• Release date: October 2024 vs June 2024
• Launch MSRP: $13564 vs $11350
• Part number: 100-000001458 vs SRPG3