![]() ![]() I have another baremetal machine running a 3700x that doesn't even clock as high as my 3960x and it gets better single core results! I'm obsessing over single-core performance because I game with a lot of unoptimized titles that are single-threaded and I know my 3960x on a VM is bottlenecking my 3070 ti, that would otherwise not in baremetal. Yeah that's the consensus I noticed with a lot of people after researching a LOT. And those seem mitigable through passed-through devices if possible.Įasy way to dump xml is to use: virsh dumpxml (your_VM_name) > (name_of_xml_file.xml)Īnd it'll dump to whatever directory you are in currently My belief now (and it was likely known by others already) based on these tests is that CPU performance is the least impacted resource for a VM, likely graphics and storage (maybe networking?) are far more impacted resources. I mention this just in case you struggle with it under Ubuntu. But I could never get it to run under an F35 VM for some reason (something seems off with my wine install there). Other than that I think the config is fairly standard (based on my memory).Ĭinebench R23 ran fine under wine on my bare metal F35 and my F36 VM after installing a lot of wine packages. I am passing through a secondary graphics card and the bare SSD as you said and have the CPU configured as host-passthrough (I would think this is the only item that might have an impact on a single core CPU benchmark). I see others posting such info but I am currently clueless as to how. The only way I can copy/paste is to switch to Markdown Mode and that seems to cram all my pasted text together. I would happily share the xml with you if I knew a way other than manually typing it in. The Ubuntu VM's "see" an Epyc for some reason. I have all VM's set to "Copy host CPU configuration". Not sure why U20 doesn't work like the others. The U21 sees the native resolution and looks nice. Since I don't use Ubuntu much.the U20 VM, for some reason, refuses to "see" the native monitor resolution so it has to run at a slightly lower res. I also re-ran it for my bare host and got 1681 for single core/14301 for multi. ![]() I then re-ran the Geekbench tests as well - the Ubunut 20 VM did 1697 for single core/9252 for multi, the Ubuntu for single core/9322 for muiti. So very similar, if not identical, to the Fedora tests above. Each ran a little over 4M in 1m for the fast test. The cpu-x benches were almost idential - each ran 80K in 1m for the slow test. I re-ran these same tests, this time using an Ubuntu 20 VM, an Ubuntu 21 VM and my bare host (F35). ![]() So, to answer your question, to me it looks like I am getting 1:1 single core performance. no manual cpu pinning or anything like that. But those benches were - 8936 for F35 VM, 12739 for F35 bare host, and 8852 for F36 VM.Īs an fyi, i have done nothing other than basic virt-manager creation of these VM's, i.e. I assume the multi-core benches would not be comparable since I was only assigning 4 logical cores to each VM (this is on a 5950x). Again, the single core benches were incredibly (to me) similar - 1622 for F35 VM, 1641 for F35 bare host, and 1664 for F36 VM. I then installed Geekbench on bare host and both VM's. But in the bare host, the cores used for its bench as well as the cores used for the VM's benches migrated to different cores several times. I was also running bpytop on each while the bench ran and found it interesting that, in the VM's, the bench process stayed on a particular core (from the VM's viewpoint of what the core was). The "Prime number (fast)" bench was still 90+% identical - all a little over 4M in 1m. The "Prime numbers (slow)" bench on each ran almost identically, 80K in 1 m. I ran cpu-x on my bare host Fedora 35, a Fedora 35 VM and a Fedora 36 VM all at the same time. I wasn't sure how you were running Cinebench or CPUz since they both appear to be Windows only apps? Your question struck me as interesting so I thought I'd try to test it. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |