I have some issue on the multi-core performance of some server. The server are HPE DL360 Gen10, mounting 2x Xeon Gold 6154 (18 cores). When i refer to the performances they are slower then some older counterpart on HPC computation (CFD, in particular), and I confirmed it with Geekbench value.
On the single core Geekbench 5 gives a score of 1001, while on multicore 13292. From the topology Geekbecnch recognizes 2 Processor, 36 cores, thus should be using every core. But looking on other benchmark online I can expect at least 22000. Other than Geekbench scores. the server is running around 10% slower than an older 2x Xeon E5-2690 (14 cores).
The server is running Windows Server 2012 R2, RAM configuration is 8x16GB Single Rank x4 DDR4-2666 CAS-19-19-19, the RAM slot are correctly populated. During the HPC computations Task manager is steadily at 100% usage, 3.44 Ghz speed. Hyper threading is off.
What i did until now was looking to the RAM bandwidth, the server had only 4 RAM slot per processor and if I understood correctly, this Xeon family should have maximum bandwidth performance with 6 channel, thus I installed 2 other stick per CPU (same type), for a total of 12 RAM stick. The ram is populated according to the HPE guide printed on the server case, thus slot 1,3,5,8,10,12 are populated for each core. However, even with the RAM upgrade, the score were: 998 single core, 13331 multi-core, thus basically no difference.
Tomorrow I may be able to bring the server offline and do some other testing, but I'm out of my depth on this issue, thus even some pointer to what to look to into would be useful.
Thank you for any information in advance.
EDIT: Probably is something related to the OS.I tried to switch one of the server to Centos 7.3, score jumped up over 25k. So, same hardware, same BIOS configuration (probably, I compared them manually), different OS, different performance.
I tried to update Win Server 2012 to Win Server 2019, but no visible change on performance. But i need Win Server running. Any suggestion?