Test Methodology
Server Processor Configurations |
|||||
|---|---|---|---|---|---|
| CPU |
AMD Epyc 7763 |
AMD Epyc 7742 |
AMD Epyc 7402 |
AMD Epyc 7F52 |
Intel Xeon Platinum 8280 |
| Codename | Milan |
Rome |
Rome |
Rome |
Cascade Lake Refresh |
| CPU Base Clock | 2.45GHz |
2.25GHz |
2.8GHz |
3.7GHz |
2.70GHz |
| CPU Turbo Clock | 3.50GHz |
3.4GHz |
3.35GHz |
4.6GHz |
4.0GHz |
| CPU L3 Cache | 256MB |
256MB |
128MB |
256MB |
38.5MB |
| CPU Cores / Threads | 64 / 128 |
64 / 128 |
24 / 48 |
16 / 32 |
28 / 56 |
| CPU TDP | 280W |
225W |
180W |
240W |
205W |
| Test Configuration | 2P |
2P |
2P |
2P |
2P |
| Socket | SP3 |
LGA3647 |
|||
| Lithography | 7nm |
14nm |
|||
| Motherboard | AMD Daytona Reference |
Intel Reference |
|||
| Memory | Micron 512GB RDIMM |
Micron 384GB RDIMM |
|||
| Memory Config | 16x32GB - 2DPC |
12x32GB 2DPC |
|||
| Memory Speed | 3,200MHz |
2,933MHz |
|||
| Disk Drive | Micron 256GB 1100 SATA SSD 6x Micron 9300 3.84GB NVMe |
||||
| Power Supply | 2x 1,200W Redundant Configuration |
1,200W |
|||
| Graphics Card | - |
||||
| Operating System | Ubuntu 19.04 |
||||
Benchmark Suite |
|
|---|---|
| STREAM Memory 1.3.1 | Used to evaluate bandwidth using optimised compilers for both platforms. |
| NAMD 1.0.1 | NAMD is a parallel molecular dynamics code designed for high-performance simulation of large biomolecular systems. |
| OpenSSL 1.11.0 | The test measures the RSA 4096-bit performance of OpenSSL. |
| Linux Kernel Compilation 1.9.1 | This test times how long it takes to build the Linux kernel in a default configuration. |
| Redis 1.10 | Redis is an open-source in-memory data structure store, used as a database, cache, and message broker. |
| POV-RAY 1.21 | POV-Ray is used to create 3D graphics using ray-tracing. |
| m-queens 1.10 | A solver for the N-queens problem with multi-threading support via the OpenMP library. |
| JTR Blowfish 1.70 | A password-cracking test using brute CPU strength. |
| C-Ray 1.2.0 | A simple raytracer designed to test the floating-point CPU multithreaded performance. |
Notes
All server combinations use two processors. We're comparing the range-topping Epyc 7763 to Epyc 7742 from a couple of years back. Both harness 64 cores and 128 threads each, although the newer chips will be faster due to the Zen 3 architecture, myriad other performance tweaks, and 280W per-socket TDP. We've further included testing from last-generation Epyc 7402 (24C48T) and Epyc 7F52 (16C32T).
Moving over to team Intel, the company provided us with a dual socket server housing a pair of Xeon Platinum 8280 chips whose 28C56T architecture is based on the Cascade Lake Refresh. Soon to be replaced by Ice Lake-based processors, 8280s still sell for around $10,000 each, which is more than the cost of each undoubtedly faster Epyc 7763 chip.
It's not as clear cut as that, however, because our preferred comparison would have been the Xeon Gold 6258R whose specifications are identical to 8280 save for having 2P support instead of 8P and consequently two UPI links instead of three. That's no issue at all as AMD Epyc chips are limited to 2P functionality in any case.
The massive TCO benefit of Xeon Gold 6258R is price. Costing $3,950 they are better prepared to battle AMD's mid-core offerings, and we're particularly keen to see how they shape up against the also-28C56T Epyc 7453 in upcoming editorial. Nevertheless, for our purposes, it is reasonable to substitute Intel Xeon Platinum 8280 performance for Gold 6258R.
For those interested, we noted a peak 853W power consumption during testing, compared to 682W for a dual Xeon Platinum 8280.
We use a subset of the excellent Phoronix Test Suite to measure server performance.