Why you can trust Tom's Hardware Our expert reviewers spend hours testing and comparing products and services so you can choose the best for you. Find out more about how we test.
All systems are tested in a dual socket (2P) configuration.
| Test Platform | Memory | Tested Processors |
|---|---|---|
| Intel Software Development Platform — Quanta Rack Mount 2S 2U | 16x 64GB (1TB) DDR5-5600 ECC (EMR) - SPR 16x DDR5-4800 (1TB) | Intel Xeon 8592+, 8480+, 8490H, 8462Y+ |
| Supermicro SYS-621C-TN12R | 16x 32GB (512GB) Supermicro ECC DDR5-4800 | Intel Xeon Platinum 8468 (SPR) |
| 2P AMD Titanite Reference Platform | 24x 64GB (1.5TB) Samsung ECC DDR5-4800 | AMD EPYC Genoa 9654, 9554, 9374F, Bergamo 9754 |
AI Workloads
Generative AI is grabbing the headlines, not to mention the hearts and minds of marketing departments, but there's an entire universe of different types of AI models employed in the data center. For now, training large AI models remains the purview of GPUs, and the same can be said for running most large generative AI models. However, most AI inference continues to run on data center CPUs, and we expect that trend to not only continue but also intensify.
The ever-changing landscape of the explosively expanding AI universe makes it exceedingly challenging to characterize performance in such a way that it is meaningful to the average data center application. Additionally, batch sizes and other tested parameters will vary in real deployments. As such, take these benchmarks as a mere guide — these tests are not optimized to the level we would expect in actual deployments. Conversely, some data centers and enterprises will employ off-the-shelf AI models with a little tuning, so while the litmus of general performance is applicable, the models employed, and thus the relative positioning of the contenders, will vary accordingly.
It's clear that Intel's approach of enabling AI-boosting features like AMX, AVX-512, VNNI, and Bfloat16 has yielded a strong foundation for AI users to build upon. The Emerald Rapids 8592+ is 18% faster than the 64-core EPYC Genoa 9554 in the TensorFlow ResNet-50 test, but the chips effectively tie in the AlexNet and GoogLeNet models. The 96-core EPYC Genoa 9654 surprisingly falls to the bottom of the chart in all three of the TensorFlow workloads, implying that its incredible array of chiplets might not offer the best latency and scalability for this type of model.
MLpack finds the 8592+ taking a strong lead over both of the comparable AMD processors as it completes the task 40% faster than the competition. All three of the ONNX inference benchmarks also highlight the advantage of Emerald Rapids' in-built acceleration. The Emerald Rapids 8592+ is impressive in many of these benchmarks, but the competing 62-core EPYC 9554 is roughly 5% faster in the TSCP AI chess benchmark.
Critically, we notice a big gain for Emerald Rapids over the prior-gen 60-core 8490H in nearly every workload except the GoogLeNet model. Overall, this is an impressive show from Emerald Rapids.
As impressive as these benchmarks are, we didn't have time to fully test some of the models that experience the most uplift from AMX acceleration. We're working to address those challenges, but from other third-party benchmarks that we've seen, it's clear that AMX gives Intel an incredible lead in models that leverage the instruction set.
Molecular Dynamics and Parallel Compute Benchmarks
NAMD is a parallel molecular dynamics code designed to scale well with additional compute resources; it scales up to 500,000 cores and is one of the premier benchmarks used to quantify performance with simulation code. The 96-core Genoa 9654 is the hands-down winner here, but the 64-core 9554 is also impressive, with a strong win over Emerald Rapids.
You'll notice we also included the 128-core Bergamo 9754 in the test pool, but be aware that many of the benchmarks in our suite don't specifically target the types of workloads this chip is designed for. It does provide for interesting viewing, though. We'll be able to add Intel's competing Sierra Forest early next year, but this does highlight that Intel lags AMD in the density-optimized chip department.
The award-winning Stockfish chess engine is designed for the utmost scalability across core counts — it can scale up to 512 threads. The Genoa 9654 takes the lead with an effective tie with the Bergamo 9754. This massively parallel code scales well with EPYC's core counts — the 64-core EPYC 9554 is 39% faster than the 64-core 8592+.
The Gromacs water benchmark simulates Newtonian equations of motion with hundreds of millions of particles. This workload scales well, but the fact is that many workloads will hit other bottlenecks, like memory throughput, NoC, or power constraints before they can fully leverage the highest core counts. Here, the 8592+ is 19% faster than AMD's 96-core and right around twice as fast as the 64-core 9554.
We can also say much the same about the LAAMPS molecular dynamics code. While this code is inherently scalable, we're obviously reaching other bottlenecks before the full might of the 9654's cores can be unleashed. The 8592+ carves out another surprising win in this benchmark.
y-cruncher is incredibly sensitive to memory throughput, and here we can see that the per-core memory throughput for the 96-core Genoa 9654 is stretched to its limits while its slimmer 64-core counterpart, the Genoa 9554, takes the top of the chart as it carves out a solid win against Emerald Rapids in this multi-threaded AVX test.
Finally, if you're foolhardy enough to run cryptomining on server CPUs that weigh in over $10,000 apiece and price-to-performance is no concern, the Bergamo 9754 is your chip.
Scientific and Computational Fluid Dynamics
The Weather Research and Forecasting model (WRF) is memory bandwidth sensitive, so it's no surprise to see the Genoa 9554 with its 12-channel memory controller outstripping the eight-channel Emerald Rapids 8592+. Again, the 9654's lackluster performance here is probably at least partially due to lower per-core memory throughput. We see a similar pattern with the NWChem computational chemistry workload, though the 9654 is more impressive in this benchmark.
The 8592+ is impressive in the SciMark benchmarks, but falls behind in the Rodinia LavaMD test. It makes up for that shortcoming with the Rodinia HotSpot3D and CFD Solver benchmarks, with the latter showing a large delta between the 8592+ and competing processors. The 64-core Genoa 9554 fires back in the OpenFOAM CFD workloads, though.
Rendering Benchmarks
Turning to the more standard fare, provided you can keep the cores fed with data, most modern rendering applications also take full advantage of the compute resources. This type of workload has long been AMD's forte, and here we see that trend continue.
Intel developed Embree, a ray tracing kernel for CPUs that leverages the AVX-512 instruction set. In either case, the EPYC Genoa processors dominate these Intel-designed benchmarks.
Moving on to compute-intensive ray tracing benchmarks, the Genoa chips stretch their legs in the Embree-based OSPray benchmark and Tachyon, taking impressive leads again.
Overall, AMD's Genoa wins this group of benchmarks, even when judged on a core-to-core basis.
Compilation Benchmarks
AMD's Genoa is brutal in multi-threaded workloads, so we shifted gears to compilation work to see how the chips handle more varied workloads. Intel has made considerable gains in these types of workloads, likely borne of the decreased latency associated with the pared-back dual-tile architecture. Of course, faster memory and larger L3 caches also float all boats in compilation workloads.
Compression, Security, Webserver, NumPy and Python Benchmarks
The open-source OpenSSL toolkit uses SSL and TLS protocols to measure RSA 4096-bit performance in verify and sign operations, along with SHA-256 throughput. AMD dominates in the RSA 4096-bit sign and verify operations, but Intel takes the lead in the SHA256 workload.
Take note of the ebizzy web server benchmark, as Emerald Rapids delivers strong performance in this test. The Pybench and Numpy benchmarks are used as a general litmus test of Python performance, and as we can see, these tests typically don't scale linearly with increased core counts, instead prizing per-core performance. Intel has made impressive gen-on-gen gains here, matching the 64-core Genoa processor in Numpy and taking an easy lead in Pybench.
Compression workloads come in many flavors. The 7-Zip (p7zip) benchmark exposes the heights of theoretical compression performance because it runs directly from main memory, allowing both memory throughput and core counts to impact performance heavily. As we can see, this benefits the core-heavy chips as they easily dispatch with the chips with lesser core counts, but AMD once again wins on a core-to-core basis when we compare the 64-core Intel and AMD chips. Gzip, which we use to show compression performance with a more frequency-sensitive engine, finds the 8592+ taking the slightest of leads.
Supercomputing Benchmarks
The EPYC 9654 again takes the lead in our supercomputer test suites, though the core-dense Bergamo 9754 tops all the charts. The 8592+ shows some improvements over the previous generation 8490H, particularly in the Graph 500 test, but on a core-for-core basis AMD still leads Intel.
Encoding Benchmarks
Encoders tend to present a different type of challenge. As we can see with the FLAC benchmark, they often don't scale well with increased core counts. Instead, they often benefit from per-core performance and other factors, like cache capacity.
- MORE: Best CPUs for Gaming
- MORE: CPU Benchmark Hierarchy
- MORE: AMD vs Intel
