- Joined
- Apr 15, 2011
- Messages
- 20
- Motherboard
- Z690M AORUS ELITE DDR4
- CPU
- i9-12900KF
- Graphics
- RX570
- Mac
- Mobile Phone
Cinebench score differences are interesting because I thought full-load scores will be almost same.Benchmark with all 3 situation. Left without CpuTopologySync.kext, middle with CpuTopologySync.kext, right with CpuTopologySync.kext and -ctssmt.
Until new findings I will stay without.
I found another interesting performance diffrences.
I ran Windows 11 VM on this Hackintosh with VMware Fusion 12 Player in different host core/thread configuraions and guest vCPU configurations. I think limited vCPU count is good indicator for thread allocation.
I executed the Windows version of Geekbench 5 on the VM. I attached the result table.
8P+HT means E-Cores disabled, 8P+8E means HT disabled, and 8P+8E+HT means all features are enabled. And score texts indicated by bold are provided by the CpuTopologySync.kext. 8c24t is provided by -ctssmt boot-arg.
VM is run in different guest vCPU configurations, 8v/12v/16v.
10,689 is scored in 8vcpu with 8P+HT configuration, probably it's ideal 8vcpu performance. 24c24t and 16c24t configurations got lower scores than this, but 8c24t got equivalent score, 10,641.
In 12vcpu, 16c24t configuration got lower score than 24c24t, but 8c24t still got almost same score as 8P+HT. I think 3-way SMT spoofing working well as I expected.
This spoofing setting is strange topology, but I hope to make it working steadily.
( I forgot to take some screenshots, so the table and other Geekbench 5 screenshots show little different score. )
Attachments
Last edited: