VERY low physics benchmarks when running in WIN 10 VM


Recommended Posts

First off, THANK YOU to this incredible software, forum, videos guides and more. I am now able to be running my main windows 10 machine off the same hardware as my server, with great overall performance in gaming, photoshop, and more. 

 

I am having a strange cpu benchmark/performance issue that maybe someone know the fix for.

 

When running both Passmark software as well as 3D Mark Firestrike, I get VERY low physics calculation scores. So much so, that when viewing the test running in Firestrike, I can see how bad it is, around 7 FPS, with a physics score of about 2,000. (bare metal is 6000+ physics, and 25FPS)

 

Passmark CPUMark as a whole appears to be perfectly normal, minus just a bit probably due to unraid overhead. 7500 score overall. Passmark 3d Direct 12 score is only running at 10 FPS. 

 

Cinebench came in at CPU score of 420, not too bad for this particular processor.

 

Apps and Games run pretty well overall, but there are slight FPS drops here and there. Perhaps a 5-7% drop from bare metal. Very pleased overall, I just can't figure out why the insanely low caluclation/physics score.

 

I pulled unraid out of the equation, same hardware and overclock, and saw a Firestrike physics score of 6000+, and a FPS of 25 on the test, VM was 10. Passmark Direct X ran at 50FPS baremetal vs 10FPS with the VM.  HUGE difference here.  It just seems like something is off. Only variable is I used an 80GB SSD for the OS, instead of the Cache drive SSD for my VM.

 

There is something in a setting or passthrough or something else I missed that is causing the issue I'm sure. 

 

Since these are mostly all CPU tests I think, I feel it is not related to the GPU passthrough, but not sure. I have tried multiple different NVidia drivers, same result. Baremetal performance was flawless, and runs a bit faster overall I think than the VM.

 

The VM was a fresh Win10 install to begin with of course, as was the bare metal. I also have a few games that are not maxing out the GPU or CPU, even on ultra settings, it just doesn't give me a good FPS, so maybe it is some kind of issue related to this?

 

My Software:

 

Unraid 6.3

Dockers - DuckDNS, PlexServer

VM - Windows 10 Pro

 

Hardware:

 

ASROCK 970m PRO 3 motherboard

PNY GTX 1050TI for passthrough, GTX 6100S for unraid itself

AMD FX 6300 overclocked to 4.4ghz - temps very normal, max of 50C under load with evo 212x cooler

12gb DDR 1600 ram. 8gb of vengance, and 4gb of gskill (2 x 2gb modules)

antec 450 watt power supply

PCIe USB 3 passthrough to VM.

4x 3.5 WD Green 1tb HDD'S for the array and parity

240gb SSD for Cache

 

Any thoughts? Love to know 

 

Link to comment

I did just notice this nugget on the VM FAQ:

 

Further Analysis

Looking at the tests in close detail, you’ll find that the area of biggest variance between our physical and virtual tests was in physics.  This makes sense because we are running another entire operating system on top of the machine which does need some resources to operate, albeit a minimal set.  That said, when we look at the combined test, that same level of variance isn’t seen.  The reality is that while physics are important, they don’t impact overall frame rate as much as they impact how effects appear in game, and from personal play testing experience, I can tell you that I don’t notice any tangible differences.

 

Does that mean what I'm seeing is pretty normal, or am I still experiencing a huge impact on the physics?

Link to comment

Hey Kylel I don't think this is normal.  I am seeing the exact same thing.

I have a hunch this is unique to Nvidia GPUs or drivers.
 

I'm running a FX 8350 CPU on an ASUS Sabertooth 990fx motherboard.

I have two windows 10 VMs (called BLUE and RED)

RED is assigned cpu cores 4-7
and is passed 
1 USB controller (PCI-e card in the last motherboard slot)
as well as a Radeon RX 480

BLUE is assigned cores 0-3
and is passed
1 USB controller (built into the motherboard but not the one the unraid Flash drive is in obviously)
as well as an Nvidia GTX 980

I just recently put the RX 480 and GTX 980 GPUs in the system.  Prior to those cards being installed I was passing a Radeon R9 280 to each VM.

Running 3dmark Firestrike with an R9 280 in each VM I produced these results.

BLUE VM 3dMark score while the RED VM was completely shutdown - 5379
BLUE VM 3dMark score while the RED VM was on and idle - 5383

RED VM 3dmark score while the BLUE VM was completely shutdown - 5498
RED VM 3dmark score while the BLUE VM was on and idle - 5496

Scores while running 3d mark simultaneously on both.
BLUE - 5338
RED - 5437


Now unfortunately things get a bit hard to follow what may have changed because in troubleshooting getting the GTX 980 to work in a VM I did at least the following.

1) upgraded unraid from 6.1.something to 6.3.3
2) changed the port the unraid flash drive was in
3) passed a different usb controller to the BLUE vm than I was previously.

But now with the RX 480 installed the RED VM and the GTX 980 in the BLUE VM I'm getting the following results

BLUE VM 3dMark score while the RED VM was completely shutdown - 5901 
BLUE VM 3dMark score while the RED VM was on and idle - 5945

RED VM 3dmark score while the BLUE VM was completely shutdown - 6954
RED VM 3dmark score while the BLUE VM was on and idle - 6922

Scores while running 3d mark simultaneously on both.
BLUE - 5906
RED - 6787

Specifically in the phyics tests in Firestrike the RED VM with the RX 480 was getting 11-12 fps while the BLUE VM with the GTX 980 was getting ~6 fps

I thought maybe I was just completely bottle necking the GTX 980 with the CPU.  But I tried passing all 8 cores to the BLUE VM and running Firestrike again it produces about the same result.

So... something is odd here..  the RX 480 is a good card, but it certainly isn't faster than an RX 980 and certainly not by the amounts I'm seeing here.

Also of note I'm using the latest drives from AMD and Nvidia for both cards.  Don't have the versions handy but I'm happy post them on request.

Anyone have any thoughts on what could be going on here?

Link to comment

I came across this info here discussing possible performance loss with Nvidia cards when enabling the hyper-v option.

http://lime-technology.com/wiki/index.php/UnRAID_6/VM_Management

 

Quote

If you specify Windows as the guest operating system, you can toggle the exposure of Hyper-V extensions to the VM

  • This is disabled automatically if an NVIDIA-based graphics card is selected for assignment to the VM
  • See this post about 3D gaming performance with NVIDIA-based GPUs, Hyper-V settings, and various driver versions


I disabled that hyper-v option on both VMs and the scores did not change very much (within the margin of error I've seen while running all these tests)

Kind of out of ideas here I think.

Link to comment
  • 2 weeks later...
  • 1 month later...

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.