Sysbench, Stream, C-Ray, 7-Zip
Sysbench is a widely used suite that characterizes CPU, memory, file I/O, mutex, and MySQL performance. We focus on the CPU test, which measures the amount of time required to verify prime numbers in both single- and multi-threaded workloads.
The EPYC 7601's stellar performance in the single-threaded test is impressive given its relatively conservative clock rates. But the chip loses steam during the multi-threaded test, possibly due to architectural influences from the multi-chip module (MCM) design.
The Cascade Lake line-up aligns neatly based on clock rates during the single-threaded test. Then it falls into the expected hierarchy based on core count during the multi-threaded benchmark.
Stream is a relatively simple test developed by Dr. John D. McCalpin. It measures the sustainable memory throughput of a given system in MB/s.
AMD's EPYC 7601 leads the pack during the copy and scale tests, but it isn't uncommon to see the chip deliver more throughput thanks to its eight-channel memory interface.
The improvement from Intel's Broadwell (quad-channel) to Skylake (six-channel) architectures is impressive. The increased throughput from Skylake to Cascade Lake isn't as significant, though we do see some gains from support for faster DDR4-2966 memory.
C-Ray is a ray tracing benchmark designed to reside entirely inside of a CPU's cache, thus eliminating RAM and disk I/O overhead from the measurement. This test focuses purely on floating-point performance during rendering, and it runs across multiple threads.
The ray tracing workload scales nicely based on core count, and AMD's EPYC 7601 shows its chops with 32 cores that carve out a lead. Clock rate clearly plays a role too, as we see the 24-core Xeon 8268 eke out a win over the previous-gen 28-core Xeon 8176. We also notice the Xeon 6154, which has a 900 MHz advantage over the 22-core Xeon 6152, beat its counterpart.
7-Zip is open source software that measures compression and decompression performance, which can be a key capability for storage and networking applications.
As service contracts expire, most enterprises will move up from older Haswell, Broadwell, and Skylake platforms to the Platinum series. This convincing set of benchmarks highlights yet another aspect of day-to-day operations that will benefit from boosted performance.
AMD's EPYC 7601 showcases the Zen architecture's fondness for decompression workloads, while delivering substantially less compression performance.
HardInfo provides granular system information and includes a suite of benchmarks for measuring CPU performance. It's easily accessible and comes as a standard component in many Ubuntu desktop systems. We include these tests because they allow our Linux brethren to easily run comparison data.
Intel's Platinum 8280 outperforms our expectations in the Cryptohash test. However, we notice reduced performance from the company's LGA 3647 processors during the FPU ray tracing benchmark. Intel informs us that some applications suffer mesh-imposed performance penalties until the software ecosystem adjusts, though we've only found the altered architecture to affect a few enterprise-class benchmarks. We suspect we can add this specific title to the list of applications that might see better performance after optimizations.
MORE: Best CPUs
MORE: All CPU Content