Post your Passmark results of your Windows VM

[powerhouse]

I'm starting this thread in the hope that you share your Windows Passmark results for your Windows virtual machine (VM), if possible compared to a bare-metal Windows installation on the same PC. In addition to the Passmark results, please specify your virtual machine manager (Xen, KVM, Virtualbox, etc.) and the hardware you are using.

Once this thread fills with data, it will give users and particularly Windows gamers an indication on what performance to expect from a virtual Windows installation on Linux.

Unfortunately I run Windows only in a VM, so I can only publish my Windows VM results.

Hardware:
CPU: Intel i7 3930K
Graphics (Linux): AMD Radeon HD 7770
Graphics (Windows): Nvidia Quadro 2000
Disk (Linux OS): 120 GB Sandisk SSD, LVM formatted (except /boot)
Disk (Windows VM): same as above, LVM formatted, using GPLPV drivers
Virtualization: Xen 4.1.3, "xm" toolstack
Remark: Identical (+/-0.2%) Passmark results with LM16 and Xen 4.3, again using xm toolstack

Passmark results - summary - for Windows VM on Xen 4.1.3 with VGA passthrough, 10 VCPUs, Nvidia Quadro 2000 GPU

Windows VM on Xen 4.1.3 with VGA passthrough, 10 VCPUs, Nvidia Quadro 2000 GPU

The 3D graphics aren't smashing, but still slightly better than the average for that graphics card .

[kgeorgeg7]

Hello powerhouse,

First of all I must thank you for all this guide you offered to us, which I partially translated to greek and posted it on a greek tech forum ADSLgr.com

I am running xen hypervisor for a lot of time. I remember I followed this guide here to make it work, and it did.

I have an Intel i7 3770, MSI Z77A-G43, 24GB of RAM (1600MHz @ 9-9-9-24), 3 gpus: the onboard Intel HD 4000, a Sapphire 7950 Toxic and a Sapphire 7770.

I run xen hypervisor with 2 guests: a Windows 7 system with 8GB of RAM and the 7950, and a Windows 8.0 system with 6GB of RAM and the 7770.
The thing is, I currently running Debian wheezy (stable) and not linux mint, is it ok to post my passmark results?

I must also say that when a friends come over we play a lot of games in the configuration I described above without any problems. (We usually play LoL Dota 2 and Counter Strike Source/GO)

Also, whatever method that I tried to make the Intel HD 4000 to work failed. The onboard gpu just don't want to play.

Regards,
George

[powerhouse]

Hello powerhouse,

First of all I must thank you for all this guide you offered to us, which I partially translated to greek and posted it on a greek tech forum ADSLgr.com

I am running xen hypervisor for a lot of time. I remember I followed this guide here to make it work, and it did.

I have an Intel i7 3770, MSI Z77A-G43, 24GB of RAM (1600MHz @ 9-9-9-24), 3 gpus: the onboard Intel HD 4000, a Sapphire 7950 Toxic and a Sapphire 7770.

I run xen hypervisor with 2 guests: a Windows 7 system with 8GB of RAM and the 7950, and a Windows 8.0 system with 6GB of RAM and the 7770.
The thing is, I currently running Debian wheezy (stable) and not linux mint, is it ok to post my passmark results?

I must also say that when a friends come over we play a lot of games in the configuration I described above without any problems. (We usually play LoL Dota 2 and Counter Strike Source/GO)

Also, whatever method that I tried to make the Intel HD 4000 to work failed. The onboard gpu just don't want to play.

Regards,
George

Efkaristo poli (sorry, I haven't got a Greek keyboard)! I'm very happy to hear you translated the how-to to Greek.

Please do post your Passmark results (and/or other benchmarks) here. It really doesn't matter if you use Debian, Ubuntu, Fedora, or Linux Mint.

About the HD4000, have you tried Xen 4.3 (for example with Linux Mint 16)? You should check the xen.org website - it might be supported in one of the newer Xen releases.

Yassu !

[Jess1981]

The Windows VM is still running fine. Reposted the benchmarks below:

[darthazad]

System:
VM: Xen-AMD64-4.3 on Linux mint 16 KDE DOM0
Guest: Win7
CPU: Intel i7 2600
Ram 16G (12G for the VM)
windows VGA: AMD R9 270x
Linux VGA: Intel HD Graphics 2000
Samsung SSD for Linux
2 Intel SSDs for Windows

--EDIT--
@powerhouse, I did use the xen pv widows drivers:
DriverVer=10/22/2012,0.11.0.372
Under device manager I have 2 disks that show up as Xen PV disk and seem to perform well
I also have a couple of exclamation marks next to (! SCSI controller) and (! Xen pci device #0),
but I am not worried about them. Cheers!

My non-virtualized windows is gone but I had the same WEI scores (except for disk, which inexplicably went from 5.9 bare to 7.7 virtual, go LVM?)

I really appreciate the guide, I would not have been successful otherwise.

[powerhouse]

@darthazard: Thanks for sharing your benchmarks and hardware specs! I also have the feeling that disk performance in the Windows VM is better than bare metal, but I can't proof it. It's great to see others who quit dual-booting .

One question: Do you use the GPLPV drivers under Windows? Which version? (You can post the answer in your original post above.)

[NotYetRated]

Awesome guide! Could not have done this without you.

I am going to post hardware specs now, and passmark results when I get home to run them.

System:
VM: Xen-AMD64-4.3 on Linux mint 16 KDE DOM0
Guest1: Win7a - Media Center and Gaming
Guest2: Win7b - Purely Media Center
Guest3: WinServer 2011 - Media Server/Backup Server
Guest4: CentOS - Minecraft server
CPU: Intel Xeon 1245V3
Mobo: AsRock 226WS
Ram 32G: 16 to CentOS, 2 to Media Center, 4 to Media Server, 6 to Gaming
Win7a: AMD 7850
Win7b: AMD 6450
IBM Serveraid passed through to Media Server, with 8 drives.

The rig runs surprisingly well. I took the dive in to this not really expecting nearly the level of performance that I actually get out of it. Win7a runs any game I want, while Win7b has a media center passively cooled card not really meant for gaming, but it plays some low end games as well as bitstreams blurays perfectly.

The MC server typically has between 5 and 50 people on it, and has never faced performance issues when I have been gaming and when my girlfriend has been watching a movie.

I plan to sell my dedicated gaming rig, as well as the hardware to this rig and upgrade to a 2011 Intel socket, or whatever the new enthusiast boards socket ends up being. This way, I can get a couple more cores, and throw a bit more RAM in to it. It is a little nerve racking having all of my eggs in one basket, however with the quality of modern Intel SSD's, and other components, I feel comfortable enough. Plus, it was easy enough to set up to begin with, thanks to the guide.

[Nesousx]

Hello all,

here are my hardware info and results:

Hardware:
CPU: Intel i5-3570 (only 3 VCPUs passed to win7 domU)
MB: Asrock Z77-Pro4-M (will add Bios version later)
RAM: 32Gb (dom0) and 12Gb (win7 domU)
Graphics (Linux): intel integrated with CPU
Graphics (Windows): AMD Radeon 7950
Disk (Linux OS): Crucial M4 240GB SSD
Disk (Windows VM): OCZ Agility 3 250GB SSD
Virtualization: 4.1.3-3ubuntu1

LVM + GPLV drivers (will add version later), and XM toolstack.

Results:

[powerhouse]

Deleted

[Nesousx]

@Nesousx: Thanks for sharing the benchmarks!

I have some questions/requests:
1. Motherboard model and BIOS release?
2. I assume you use the "xm" toolstack.
3. 3 cores - do you mean 3 VCPUs ?
4. Disk performance - do you use the GPLPV drivers and LVM?

Thanks again.

You're welcome! I edited my post with information asked.

[powerhouse]

@Nesousx: Thanks for updating - impressive results for only 3 VCPUs! Like in my test, the virtual disk drive seems to outperform the published average results here http://www.harddrivebenchmark.net/hdd.php?hdd=OCZ-AGILITY3&id=47.

[flobo71]

Hi all,

here some passmark results on my hardware: Asus Z87I-Pro, i4570, radeon 7750 (passive), 16Gb
vm: Win7 on Xen with 4 VCPUs, 8 Gb, harddisk with lvm on one partition of a WD disk with 7200rpm
bare metal: Win8 on a Samsung 830 SSD, same hardware otherwise.

Code: Select all

          vm       bare m.
total     1942       3810
cpu       7117       7179
2d         898        873
3d        1913       1910
mem.      2239       2512
disk       425       3590

CPU and graphics on par, memory 10% slower under xen. Disk is really really slow but its no SSD.
Any ideas how to speed up the disk?

[powerhouse]

Hi all,

here some passmark results on my hardware: Asus Z87I-Pro, i4570, radeon 7750 (passive), 16Gb
vm: Win7 on Xen with 4 VCPUs, 8 Gb, harddisk with lvm on one partition of a WD disk with 7200rpm
bare metal: Win8 on a Samsung 830 SSD, same hardware otherwise.

Code: Select all

          vm       bare m.
total     1942       3810
cpu       7117       7179
2d         898        873
3d        1913       1910
mem.      2239       2512
disk       425       3590

CPU and graphics on par, memory 10% slower under xen. Disk is really really slow but its no SSD.
Any ideas how to speed up the disk?

Yep, disk performance sucks. Did you format LVM and use the GPLPV driver (take the latest non-beta)?

[flobo71]

Yes, LVM and GPLPV were installed as in your guide.

Today, I moved the guest vg to a new Samsung 840 SSD and did the benchmark again.
HD score is now at 4819 with xen! So it was really the difference between SSD and HDD.

[powerhouse]

Yes, LVM and GPLPV were installed as in your guide.

Today, I moved the guest vg to a new Samsung 840 SSD and did the benchmark again.
HD score is now at 4819 with xen! So it was really the difference between SSD and HDD.

Thanks for the update. I didn't notice that your bare metal Windows install was using an SSD, which explains the difference.

Now how is your Xen Passmark result (total) versus the Windows bare metal installation?

EDIT: I just checked some benchmark results regarding regular HDD performance. HDD performance depends a lot on the model, size, and age of the disk, but in any case it is far slower than SSD.

I did some benchmarks on VM disk performance a year ago, comparing SSD, HDDs in stripe mode (LVM, similar to RAID0) and regular LVM formatted disks. Here are the results: http://forums.linuxmint.com/viewtopic.php?f=42&t=112013&start=40#p702361.

While I cannot compare VM disk performance with bare metal performance of Windows, the graphs shown in the link compare the benchmark results of my Windows VM with benchmark results obtained with other/similar disks running Windows on bare metal. The most telling benchmark result is the last, which shows a Windows VM disk performance result when using LVM disks. One of the disks in the comparison chart is a WD 5000AAKS model, which is one of the disks used in the LVM volume (and probably the one that was used by the benchmark program, as the other disk in the LVM volume is already full). You can see a minor performance penalty when using Xen, where the native (bare metal) performance of the disk is 580 versus 570 points achieved by the Xen Windows VM. Using these figures, Windows VM disk performance under Xen reaches 98% of the performance of a Windows bare metal installation, or 2% disk performance penalty when using Xen (this result is based on at least 2 assumptions and may be wrong, that is, there may be a bigger performance penalty when using Xen).

Having used the Windows VM to copy several terrabytes of data between different LVM volumes (disks), I didn't see any difference in disk performance between a Windows bare metal installation (or Linux for that matter) and Windows running under Xen. Some time ago Phoronix published benchmarks comparing Xen and KVM performance using an Ubuntu 12.10 guest, but I think those results may be flawed - see http://www.phoronix.com/forums/showthread.php?72332-Ubuntu-12-10-Linux-KVM-vs-Xen-Virtualization-Preview and my comments at posts #18 and #19.

There are a number of benchmarks floating on the web, plus user comments praising one and condemning the other virtualization technology, that I was hoping that this thread here could provide some hard facts, like the Passmark results that allow easy comparison.

I'm sure limitations or flaws can be found with the Passmark benchmarks, but for the purpose of comparing a single Windows VM with Windows bare metal performance in a desktop setup, the Passmark results should be quite indicative.

[Trojal]

Excellent guide, thank you powerhouse!

I haven't run it yet on barebones, but here is my virtualization performance:

Win 8.1 pro x64 DomU:
i7 3770: 6 cores of 8
11GB RAM
LVM on SAMSUNG 840 Pro SSD
Radeon HD 7970 GHz GPU

Installed GPLPV drivers.
Also passed through secondary onboard network adapter directly to DomU, not using a bridge.

I'm getting consistent windows lockup when trying to pass 2x Radeon HD 7970 through to DomU, but using either one independently works. So for now giving up on getting Crossfire working.

[powerhouse]

@Trojal: Thanks for posting your Passmark results - they look fine.

I can't help much with the crossfire configuration except that some users reported success. You may want to search the lengthy HOW-TO thread, but I can't recall anything out of the ordinary.

One thing that comes to mind is the AMD driver. Have you installed the latest driver? Also, the Catalyst Control Center is known to cause problems. You may need to uninstall it. See here for some hints.

It might be related to your motherboard and the PCIe configuration. Some boards use PCIe controllers or switches that can pose a problem with certain PCIe slots. So your first PCIe slot might be OK, but the second one not. You can test that by inserting your GPU in the corresponding PCIe slot and checking it (IIRC the PCI ID will change, so you need to update your initramfs with the new PCI IDs). There are some guest configuration parameters that might help (msi-translate, etc.). Check here for xl guest configuration options.

The xenusers mailing list is an excellent source for more information and help!

[powerhouse]

I did some initial performance comparisons between KVM and Xen. Unfortunately my regular Xen installation runs on an SSD, whereas I do my experiments using an external USB3 HDD drive. This means it's hard to tell which virtualisation solution performs better. Moreover, under KVM the Passmark PerformanceTest will run only in SAFEMODE, else Windows crashes (BSOD). Again under KVM, Passmark doesn't pick up SSE support, although CPU-Z detects SSE.

CPU-Z under KVM

Here the CPU performance comparison:

KVM versus Xen (click image for large view)

KVM versus Xen CPU performance

Since I'm not familiar with KVM, the above results and the incompatibility issue with Passmark may be my fault (i.e. misconfiguration), or some patches haven't made it into the kernel. But perhaps Xen is more mature and simply performs better. I can't tell yet.

Here the configuration I used:

Code: Select all

sudo qemu-system-x86_64 \
-bios /usr/share/qemu/bios.bin -vga none \
-name win7 \
-cpu host \
-smp 10,sockets=1,cores=5,threads=2 \
-enable-kvm \
-m 24G \
-rtc clock=host \
-vga none \
-serial null \
-parallel null \
-monitor none \
-display none \
-k en-us \
-machine type=q35,accel=kvm \
-boot order=cd \
-device ahci,id=ahci \
-device virtio-scsi-pci,id=scsi \
-drive file=/dev/external/win7amd,cache=none,if=virtio \
-drive file=/home/heiko/Win7.iso,id=isocd -device ide-cd,bus=ide.1,drive=isocd \
-drive file=/home/heiko/virtio-win-0.1-74.iso,id=virtiocd -device ide-cd,bus=ide.2,drive=virtiocd \
-device ioh3420,bus=pcie.0,addr=1c.0,multifunction=on,port=1,chassis=1,id=root.1 \
-device vfio-pci,host=01:00.0,bus=root.1,addr=00.0,multifunction=on,x-vga=on \
-device vfio-pci,host=01:00.1,bus=pcie.0 \
-device vfio-pci,host=00:1a.0,bus=pcie.0 \
-net nic,model=virtio,macaddr=00:16:3e:00:03:03 -net tap


[powerhouse]

Shocky_Han over at the Ubuntu forum posted his passmark results here. This is the direct link: http://www.passmark.com/baselines/V8/display.php?id=24178651031.

The results are, however, a little problematic as they are far lower than expected.

[kgeorgeg7]

Efkaristo poli (sorry, I haven't got a Greek keyboard)! I'm very happy to hear you translated the how-to to Greek.

Please do post your Passmark results (and/or other benchmarks) here. It really doesn't matter if you use Debian, Ubuntu, Fedora, or Linux Mint.

About the HD4000, have you tried Xen 4.3 (for example with Linux Mint 16)? You should check the xen.org website - it might be supported in one of the newer Xen releases.

Yassu !

I must apologise for that late response. Your greek is awsome !

Well I finally managed ( more like remembered ) to do the passmark test on one of my VMs that I run with the Xen.

Please note that I have 2 R9 280X both of them work nicely, and I have played some games with some friends when they come over. Usually League of Legends, Dota 2 and Battlefield 4. Yes i7 3770 manages to run 2 games at the same time !

I run a VM with my main Windows 7 VM ( I name it win7_primary ) and my second Windows 7 VM ( I name it win7_secondary ). Each of, has a R9 280X for itself. When I use both the VMs for Gaming I let the VMs 7GiB of RAM to each.

Also I must note that I never had a success with xl toolstack with ANY version of Xen with or without the famous ati-passthrough.patch (which needs some modification to patch correctly to Xen 4.4 version but despite that it doesn't work for me).


So I'll stick with Xen 4.3 and xm toolstack until something stable comes out.