Hello to everyone,

 

In the last few days i started to do some testing, comparing vm performance by running Cinebench R15 between unRAID, XCP-NG and ESXi.

 

I have 2 almost identical system, the first one (my main unraid server) has a Xeon E5-2698v3, 64 GB RDIMM, Asus x99 WS/IPMI and a Quadro P2000. This server has also 2x1TB SSD for cache and 4 HDD for the Array (28TB total with single parity).

The second one (test server) has only a different CPU, a Xeon E5 2695v3 and no GPUs.

 

On my main server i have a Win10 VM with 16 cpu assigned (8 cores + 8 matching HT) and a Quadro P2000 passed through. In cinebench i was normally getting 1000 points on the multicore test, and between 75 and 80 for the single core on this VM.

 

I tried to replicate the same VM on my test server with a brand new unraid installation and i got almost the exact same score on multi, but a slightly higher score on single (between 100 and 110 points), i think that the higher single score is due to the fact that the test server was only running this VM and nothing else.

When i booted up the same sized vm on XCP-NG and ESXi on my test server i was a little bit shocked, at the first try i got 1500 points on multi and 120-125 in single, in cinebench. I wasn't expecting this huge difference.

 

Since then i tried some configuration of the CPU in unraid and i discovered that if i assign to the VM only "true" cores (i tried with 12) i get a higher score (1150 on multi, but the same on single). i got this result on both my main and my test server.

I tried also to isolate the cpu cores, but the score got worse on multi, i was down to 850 points.

I tried also to add the "emulatorpin cpuset" in the XML of the VM but the performance didn't change that much.

 

Do you have any advice on what to try or what could be the cause of this difference in performance?

 

Thank you in advance

Gianluca

Interesting stuff.

 

I have to test a VM for CAD application soon that is running in unraid. It has also 16 cores (AMD Ryzen 9 3900x), 24Gb of RAM and a Quadro P2000 passed through and the VM runs off a Samsung 980 Pro SSD.

We'll see what I can get out of it.

I would be interested in optimisations you can think of to improve performance if there are any things that can help.

On 12/24/2021 at 10:39 AM, cam217 said:

Interesting stuff.

 

I have to test a VM for CAD application soon that is running in unraid. It has also 16 cores (AMD Ryzen 9 3900x), 24Gb of RAM and a Quadro P2000 passed through and the VM runs off a Samsung 980 Pro SSD.

We'll see what I can get out of it.

I would be interested in optimisations you can think of to improve performance if there are any things that can help.

So, what result did you get from your tests?

Unfortunately, I didn't have any optimization ideas and in the end I wasn't able to get better performance from unraid. I'm living with the degraded performance.

It would be nice if someone of the unraid developers could look into it.