search

utmapp / UTM

Virtual machines for iOS and macOS
https://getutm.app
Apache License 2.0
26.7k stars 1.34k forks source link

UTM seems to run slower than Parallels Desktop on Apple Silicon #3330

Open mjonsson1986 opened 2 years ago

mjonsson1986 commented 2 years ago

if you compare e.g parallels speed with utm parallels run with near native speed utm are litle slower iven if you run with virtualization=on why? i think both uses mac os virtualization engine?

conath commented 2 years ago

@mjonsson1986 It would be interesting to compare the speed with some tests. Are you able to test e. g. starting a Windows ARM guest in Parallels and UTM and measure the time it takes? Maybe a CPU benchmark?

mjonsson1986 commented 2 years ago

i dont own parallels anymore

24 nov. 2021 kl. 21:17 skrev CONATH @.***>:

@mjonsson1986 https://github.com/mjonsson1986 It would be interesting to compare the speed with some tests. Are you able to test e. g. starting a Windows ARM guest in Parallels and UTM and measure the time it takes? Maybe a CPU benchmark?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/utmapp/UTM/issues/3330#issuecomment-978191524, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGEUP5IKLKRIAFCC2K5RAK3UNVB6PANCNFSM5IROGO5A. Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

brunocastello commented 2 years ago

Both don't emulate the exact same configuration, so it is hard to know what exactly is different...

mjonsson1986 commented 2 years ago

emulation? qemu dont emulate arm if you run a arm device

26 nov. 2021 kl. 15:09 skrev Si13ntV0ice @.***>:

virtualization VS emulation .... real hardware multi core vs soft-single-emulated-cpu ?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/utmapp/UTM/issues/3330#issuecomment-980003517, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGEUP5JMRQSGVC3SAQZQ5C3UN6IJFANCNFSM5IROGO5A. Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

brunocastello commented 2 years ago

correct, but the hardware being virtualized/emulated is not the same between qemu and parallels, hence why there may be a performance difference, as well as methods used by them to use the hypervisor must be different as well for that difference to happen.

AwlsomeAlex commented 2 years ago

@conath I own Parallels Standard (4 vcore, 8GB RAM), what type of benchmarks do you want to see?

conath commented 2 years ago

Any CPU and I/O related tests would be interesting. If possible, a few tests that can be also run natively on your Mac, too.

For example CineBench, CrystalDiskMark (not sure what a Mac equivalent would be).

jjangga0214 commented 2 years ago

@AwlsomeAlex May I ping?

adespoton commented 2 years ago

Geekbench scores from both would be interesting.

Because it's different hardware configurations, drivers etc, I'd expect some things to be faster and others slower.

osy commented 2 years ago

Parallels also has GPU acceleration…

jjangga0214 commented 2 years ago

@osy utm provides experimental hardware GPU acceleration by OpenGL.

osy commented 2 years ago

Not on windows which is what I presume we are talking about

jjangga0214 commented 2 years ago

@osy Oh, you're right. Thanks. I've just read it's only for Linux.

brunocastello commented 2 years ago

Not on Windows? really? So why I was able to run Grand Prix 4 on it (albeit with a slower performance) ?

DirectX setup shows I have some D3D support under Windows 11 Pro (ARM64) after installing SPICE Tools. Probably only for 2D games. Still quite impressive tho.

conath commented 2 years ago

@brunocastello Windows would be falling back to Microsoft's basic WARP software renderer in that case, so basically you're running 3D in software. I would expect modern games would refuse to launch with a GPU unsupported error even though tools like dxdiag show 3D acceleration is available.

brunocastello commented 2 years ago

@brunocastello Windows would be falling back to Microsoft's basic WARP software renderer in that case, so basically you're running 3D in software. I would expect modern games would refuse to launch with a GPU unsupported error even though tools like dxdiag show 3D acceleration is available.

Luckily I don’t play modern games, only late 90s to early 2000s

aisobe commented 2 years ago

Soaring Flight Simulator Condor2 runs very slow on UTM. On parallels, it is butter smooth. The diff that I see is only 1 CPU / 1 core for UTM. Can this be configured via settings? I hear UTM not support DirectX yet.

conath commented 2 years ago

@aisobe You can enable more cores in the VM’s System settings.

jhgorse commented 2 years ago

Also, ensure that the hypervisor is enabled. I had it unticked and this makes things slow.

aisobe commented 2 years ago

Also, ensure that the hypervisor is enabled. I had it unticked and this makes things slow.

already on, but runs slow.

LeonSolisPedro commented 2 years ago

Hi! I'm not an expert, but I run the benchmarks, I hope its useful, here are the specs: (for some reason crystaldiskmark shows strange scores on both UTM and Parallels, it could be a bug thought)

Windows 11 Parallels

Benchmark Scores
Geekbench 5 5037 multicore / 1534 singlecore
Geekbench 5 (Through Intel emulation) 2214 multicore / 640 singlecore
Cinebench r23 (Through Intel emulation) 1741 multicore / 543 singlecore
CrystaldiskMark 8487.83 MBs read 8731.60 MBs write / 68.75 MBs random read 56.19 MBs random write
unigine heaven basic dx11 (Through Intel emulation) 57.1 fps / 1437 score

Windows 11 UTM

Benchmark Scores
Geekbench 5 4839 multicore / 1520 singlecore
Geekbench 5 (Through Intel emulation) 2146 multicore / 627 singlecore
Cinebench r23 (Through Intel emulation) 1620 multicore / 537 singlecore
CrystaldiskMark 7581.12 MBs read 1969.95 MBs write / 35.16 MBs random read 31.51 MBs random write
unigine heaven basic dx11 (Through Intel emulation) It simply Crashes

macOS 12 Native

Benchmark Scores
Geekbench 5 7728 multicore / 1755 singlecore
Cinebench r23 6737 multicore / 1491 singlecore
Amorphousdiskmark 3419.24 MBs read 2062.46 MBs write / 71.95 MBs random read 37.77 MBs random write
unigine heaven basic opengl N/A

macOS 12 rossetta 2

Benchmark Scores
Geekbench 5 5990 multicore / 1322 singlecore
Cinebench r23 4553 multicore / 982 singlecore
Amorphousdiskmark 3470.86 MBs read 2321.74 MBs write / 49.44 MBs random read 38.07 MBs random write
unigine heaven basic opengl 68.1 fps / 1716 score

Conclusion: I don't think it's slow, it's just that the lack of GPU acceleration makes it "feel" slow, parallels feels buttery smooth, and doesn't crash or freeze like UTM does.

adespoton commented 2 years ago

On your UTM test, are using a qcow2 or a raw volume? Those writes make me suspect qcow2, which, while the default in UTM come with a noticeable performance impact. If you don’t mind losing state saving and decreased image size, raw should result in improved disk I/O values.

-Em

On Mar 24, 2022, at 11:44 AM, Pedro León @.***> wrote:

 Hi! I'm not an expert, but I run the benchmarks, I hope its useful, here are the specs: (for some reason crystaldiskmark shows strange scores on both UTM and Parallels, it could be a bug thought)

M1 Macbook Air 16 GB RAM 8 CPU Cores / 7 GPU Cores Windows 11 8 GB RAM 4 CPU Cores on both UTM and Parallels macOS 12.3 and Windows 11 with latest updates UTM 3.1.5 with SPICE Guest Tools and drivers Windows 11 Parallels

Benchmark Scores Geekbench 5 5037 multicore / 1534 singlecore Geekbench 5 (Through Intel emulation) 2214 multicore / 640 singlecore Cinebench r23 (Through Intel emulation) 1741 multicore / 543 singlecore CrystaldiskMark 8487.83 MBs read 8731.60 MBs write / 68.75 MBs random read 56.19 MBs random write unigine heaven basic dx11 (Through Intel emulation) 57.1 fps / 1437 score Windows 11 UTM

Benchmark Scores Geekbench 5 4839 multicore / 1520 singlecore Geekbench 5 (Through Intel emulation) 2146 multicore / 627 singlecore Cinebench r23 (Through Intel emulation) 1620 multicore / 537 singlecore CrystaldiskMark 7581.12 MBs read 1969.95 MBs write / 35.16 MBs random read 31.51 MBs random write unigine heaven basic dx11 (Through Intel emulation) It simply Crashes macOS 12 Native

Benchmark Scores Geekbench 5 7728 multicore / 1755 singlecore Cinebench r23 6737 multicore / 1491 singlecore Amorphousdiskmark 3419.24 MBs read 2062.46 MBs write / 71.95 MBs random read 37.77 MBs random write unigine heaven basic opengl N/A macOS 12 rossetta 2

Benchmark Scores Geekbench 5 5990 multicore / 1322 singlecore Cinebench r23 4553 multicore / 982 singlecore Amorphousdiskmark 3470.86 MBs read 2321.74 MBs write / 49.44 MBs random read 38.07 MBs random write unigine heaven basic opengl 68.1 fps / 1716 score Conclusion: I don't think it's slow, it's just that the lack of GPU acceleration makes it "feel" slow, parallels feels buttery smooth, and doesn't crash or freeze like UTM does.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.

Saheber commented 2 years ago

Can someone tell me if the MacBook heats up while using apps like vscode and andriod studio as i am planning to shift on the m1 MacBook Air base model

brunocastello commented 2 years ago

If you have the money, get the Pro model, 16GB and 512GB minimum. Vscode will be fine with the native mac version. I dont know about Android Studio but chances are you'll have the performance throttled down to mantain the temperature cool, since the Air does not have a cooling system.

Saheber commented 2 years ago

If you have the money, get the Pro model, 16GB and 512GB minimum. Vscode will be fine with the native mac version. I dont know about Android Studio but chances are you'll have the performance throttled down to mantain the temperature cool, since the Air does not have a cooling system.

No bud i don't have the money for a 16 but will install a thermal pad in the air but only doubt bout the battery temps after thermal pad install

brunocastello commented 2 years ago

That thermal pad will not work in the long run and neither will improve the performance significantly. I saw the videos on youtube and I must say that apart of voiding the warranty, the gains from it are too small and will not be enough.

Android Studio is really, really, really demanding. A base model Air will not make the cut for it.

aisobe commented 2 years ago

@Saheber How about rent a Macbook Air and see what happens before buying the thing?

brunocastello commented 2 years ago

@Saheber How about rent a Macbook Air and see what happens before buying the thing?

https://www.youtube.com/watch?v=f_0M3BuIbJk

https://www.youtube.com/watch?v=-HOWvvOnhHQ

Both testers recommend 16GB minimum. These videos probably were recorded using Intel version of Android Studio through Rosetta 2, but AFAIK there is an Apple Silicon version of Android Studio somewhere.

Saheber commented 2 years ago

@Saheber How about rent a Macbook Air and see what happens before buying the thing?

https://www.youtube.com/watch?v=f_0M3BuIbJk

https://www.youtube.com/watch?v=-HOWvvOnhHQ

Both testers recommend 16GB minimum. These videos probably were recorded using Intel version of Android Studio through Rosetta 2, but AFAIK there is an Apple Silicon version of Android Studio somewhere.

Ok then I will loook bout it then and make sure i save more and buy a 16gigs thanks for the info though

LeonSolisPedro commented 2 years ago

On your UTM test, are using a qcow2 or a raw volume? Those writes make me suspect qcow2, which, while the default in UTM come with a noticeable performance impact. If you don’t mind losing state saving and decreased image size, raw should result in improved disk I/O values. …

I re-run the test using a the raw volume instead of qcow2 and the results where a little better. Geekbench 5 Scores now the same as parallels (and it even beats it sometimes) But Cinebench scores seems to be always lower than Parallels (one hundred points) which is strange...

I also run Ubuntu with 3D Acceleration on UTM and compared with Parallels (which also uses virt-GPU I think) they are both the same, but UTM is 1 micro fps behind (I use this website: https://solarsystem.nasa.gov)

I think I'm gonna stuck with UTM, the price for Parallels is ridiculously expensive

brunocastello commented 2 years ago

@LeonSolisPedro about Cinebench scores, Parallels is more advanced than QEMU in terms of GPU virtualization/emulation and has DX11 support. They will always beat QEMU in that aspect unless you do some VFIO or pass through a real GPU to QEMU.

If you're not into AAA games, UTM will be just fine for you.

skol101 commented 2 years ago

I don't know what params should I run windows 7 in UTM on my M1 Pro Max host, but reality is that the guest is incredibly slow. I had to buy Parallels Desktop because UTM is just getting on my nerves. And don't get me started on trying to get USB flash working on the guest or shared folders.

osy commented 2 years ago

I hate to break it to you but you can’t run windows 7 on parallels for M1…

LeonSolisPedro commented 2 years ago

@LeonSolisPedro about Cinebench scores, Parallels is more advanced than QEMU in terms of GPU virtualization/emulation and has DX11 support. They will always beat QEMU in that aspect unless you do some VFIO or pass through a real GPU to QEMU.

If you're not into AAA games, UTM will be just fine for you.

Thanks!

So... Can we close this issue now?

skol101 commented 2 years ago

I hate to break it to you but you can’t run windows 7 on parallels for M1…

Well, I'm switching to Win 11 ARM, but still have to get files from the Win7. Man, I can't get shared folders to work on Win 7 in UTM.

jjangga0214 commented 1 year ago

Is there a plan for hardware graphic acceleration on Windows?

moreaki commented 1 year ago

Is there a plan for hardware graphic acceleration on Windows?

I would transfer the value of my currently owned three licenses from Parallels and pledge support to whomever is working on GPU support in QEMU. I bet others would follow.

Not only that, but I used to pay for over 20 licenses for employees for years. More than happy to redirect respective funds to an Open-Source project, if this means we will start getting faster results.