0
I have 3 x86 xeon machines [with RHEL 7.4], machines are using Haswell, Broadwell and Skylake based CPUs.
When I run prime number calculation using sysbench [using varying number of threads], I get better results with Skylake based machine. But when I ran the following benchmarks using Phoronix suite, I see 30-40% lesser stats on the Skylake machine, while Haswell and Broadwell perform as expected [i.e broadwell performs better than Haswell]:
- Fast Fourier Transform
- Successive Over Relaxation
- Monte Carlo Performance
- Bork File Encryption
- SunFlow Rendering
The same version of Phoronix test suite is running on all 3 machines.
Why is a Skylake machine performing better with prime numbers calculation while performing decisively slower in other comprehensive tests?
Do you have power saving profile set to Performance on Skylake machine? So it always runs at full speed. Are frequencies comparable? HT on/off on all nodes? – Jaroslav Kucera – 2018-03-29T14:14:36.807
All machines running with default BIOS/EFI settings. Default power profiles. Frequencies are different. HT is on all nodes. But the performance drop is too bad on Skylake, though it's a dual socket Platinum 8160 Xeon CPU – Junaid Shahid – 2018-03-29T16:11:47.227