Mac Pro 3,1 bad VM performance for Win10

[hex41]

Hello,

I've got a Mac Pro 3,1 running High Sierra with 48GB RAM (6/8 Slots) and 3,2Ghz CPU Upgrade running from SATA SSD. In Virtualbox I get a strange bahavior for Windows 10 VMs. The Guest CPU ist shown as 100% utilized, but on the host it is not. Other Windows VMs run fine.

Can someone confirm that on his/her Mac Pro 3,1? Maybe it's due to the lack of the needed CPU extensions? But on the other hand all other Windows VMs run fine.

Thanks for your help

 

[Soba]

I had the same problem on a Mac Pro 5,1 as well as on an older iMac. I was never able to track down the cause despite extensive troubleshooting.

Make sure 2D and 3D acceleration are enabled in the VM settings, but ultimately, I ended up using VMWare Fusion.

 

[flyingmanatee]

The Mac Pro 3.1s lack the VT-x/EPT instruction sets which greatly improves virtualization speeds which were introduced in Nehalem (and trickled down into the i series family). That said, sounds like you can still improve it a bit.

I don't know specifically about VirtualBox but previous editions of Parallels and VMware had the option to enable it. I think now both require it. Its been years since I've touched VirtualBox but you might check the virtualbox settings to enable multicore support and more access to memory. I'd just be googling at this point. I used Windows 7 frequently for IE11 testing on my Mac Pro 3.1 for testing and it ran tolerably.

 

[Ludacrisvp]

I too have 3,1 8x@3.2 w/ 32gb on 10.13.6
Doesn’t seem to matter how many cores I toss at win10 vm it just seems to peg all cores at 100% seemingly doing nothing. On fusion 8.5.x. The vmdk is located on ssd as well.
Drives me nuts.

 

[w1z]

One of the reasons I switched to a MP5,1... A world of difference!

 

[Ludacrisvp]

So I moved the Win10 1607 VM from the SSD over to a 3 drive RAID0 and CPU is now 10-15% for each of the currently allocated 6 cores, activity monitor on the mac shows it using about 200% CPU (i.e. 2 full cores)
This is very different from before, didn't matter if I had 1 or anywhere between 8 cores allocated ...

It may be too soon to tell, but maybe this is something.

Theory being that there is something (VMWare bug??) interacting with being on SSD vs spinning disk.

 

[AidenShaw]

So I moved the Win10 1607 VM from the SSD over to a 3 drive RAID0 and CPU is now 10-15% for each of the currently allocated 6 cores, activity monitor on the mac shows it using about 200% CPU (i.e. 2 full cores)

I notice that the 200% CPU chart also shows 50% disk utilization - could be a big clue.

Click the "Open Resource Monitor" at the bottom, and look at the disk stats. It will show you disk activity by file and process - and perhaps solve the mystery.

 

[Ludacrisvp]

I notice that the 200% CPU chart also shows 50% disk utilization - could be a big clue.

Click the "Open Resource Monitor" at the bottom, and look at the disk stats. It will show you disk activity by file and process - and perhaps solve the mystery.

Right i get that this one showed higher than idle disk... but that isn't always the case, and in fact is hardly the case, i'd call it more of a red herring than anything.

Here was an example where CPU is pegged, disk is idle. while sure it was only 3 cores at the time it was also OC'd to 3.32Ghz from the 3.19Ghz that it normally runs at (i know not a huge change).

 

[startergo]

Has anybody tried bare metal performance with esxi:

"vSphere 6.5 introduces a Non-Volatile Memory Express (NVMe) virtual storage adapter (virtual NVMe, or vNVMe). This allows recent guest operating systems that include a native NVMe driver to use that driver to access storage through ESXi, whether or not ESXi is itself using NVMe storage. Because the vNVMe virtual storage adapter has been designed for extreme low latency flash and non-volatile memory based storage, it isn’t best suited for highly parallel I/O workloads and slow disk storage. For workloads that primarily have low outstanding I/O, especially latency-sensitive workloads, vNVMe will typically perform quite well. Compared to virtual SATA devices, vNVMe virtual storage adapters access local PCIe SSD devices with significantly lower CPU cost per I/O and significantly higher IOPS."

"DirectPath I/O vSphere DirectPath I/O leverages Intel VT-d and AMD-Vi hardware support (described in “Hardware-Assisted I/O MMU Virtualization (VT-d and AMD-Vi)” on page12) to allow guest operating systems to directly access hardware devices."

Also check your processor for:
example for X5690 (versin 6.03 officially supports cMP 5.1 with x56XX series):

• Intel® Virtualization Technology (VT-x) ‡ Yes
  
• Intel® Virtualization Technology for Directed I/O (VT-d) ‡ Yes
  
• Intel® VT-x with Extended Page Tables (EPT) ‡ Yes
  https://ark.intel.com/content/www/u...0-12m-cache-3-46-ghz-6-40-gt-s-intel-qpi.html
  
  https://neckercube.com/index.php/2016/04/11/bringing-an-old-mac-pro-back-to-life-with-esxi-6-0/
  
  
  

 

[AidenShaw]

Right i get that this one showed higher than idle disk... but that isn't always the case, and in fact is hardly the case, i'd call it more of a red herring than anything.

Here was an example where CPU is pegged, disk is idle. while sure it was only 3 cores at the time it was also OC'd to 3.32Ghz from the 3.19Ghz that it normally runs at (i know not a huge change).
View attachment 825390

Let me repeat my recommendation to 'Click the "Open Resource Monitor" at the bottom'. Resource monitor will break out CPU/Disk/Network/Memory usage by process/file/socket.

You're using the wrong tool.

Task manager will tell you that the cores are 100% busy.

Resource monitor will tell you which processes and services and files and disks and sockets are busy. Much more useful.

 

[Ludacrisvp]

Has anybody tried bare metal performance with esxi

I've not tried loading ESXi on the 3,1.

Let me repeat my recommendation to 'Click the "Open Resource Monitor" at the bottom'. Resource monitor will break out CPU/Disk/Network/Memory usage by process/file/socket.

You're using the wrong tool.

Task manager will tell you that the cores are 100% busy.

Resource monitor will tell you which processes and services and files and disks and sockets are busy. Much more useful.

I'm fully aware of "Resource Monitor" and have dug into it many times hoping for an answer to this previously, it had yielded no info that was valuable. When it shows 1 process that is using 19% CPU (usually one of the handful of svchost.exe processes or on occasion the mcafee on access scanner, even though the only application that was open was taskman) with the rest floating between 0% and maybe 3% and yet all cores are pegged at 100%, it doesn't seem to explain anything.

In Task Manager, generally speaking, the process that should be 'using' the most CPU is 'System Idle Process', prior to this change of VM storage from SSD to HDD HW-RAID0 it was nowhere to be seen. Now it is there as it should be when the machine is idle it is the 'top' process.

I'm not a novice, and don't appreciate your tone Aiden, while I am more at home in macOS and other FreeBSD variants, I do have significant Windows experience... this is not my first rodeo.
The goal of my post was to say, "hey I saw this odd behavior, not sure if correlation or causation but booting the VM from spinning disk RAID didn't have the same CPU usage as it did when booted from SSD" Perhaps it is unique to my instance, perhaps not.

 

[stampedeadam]

I'm having the same problem with my 1,1 upgraded to 2,1. It runs W10 natively very well indeed - in fact, almost well enough for me to wonder why I bothered putting together my 4790k PC. It's even good at gaming which is the real shocker given the IPC deficit. But trying to run W10 in a VM is slow to the point of completely unusable, we're talking nearly an hour just to boot to the desktop.

 

[Project Alice]

I also run a lot of VMs on my 3,1. I don't have Windows 10 in a VM however, I dualboot that. But I do have Windows 7 and a couple Linux VMs.

So, obviously we're stuck with VMware workstation 8.5.x because of our CPUs. However, Parallels 13 and 14 work. Yes, VMware is usually better. But wow the performance difference in VMware 8 to Parallels 13 is amazing. Parallels is probably 50% to 70% faster, and has better 3D. This is tested on Windows 7 VMs I made. I didn't do any benchmarks as I find those pretty useless anyways. But the real world usage feel of the two VMs is no contest. Parallels 13 is WAY faster. VMware workstation actually feels closer to virtualbox. Just my two cents, download the free trial and see what you think.