zeus83

Hi, If you're on Nvidia gpu then try Moonlight - open source Nvidia game stream client. There are no issues with controls at all. Amongst everything I tried it gives the best experience in remote gaming. As a second option you may give a try to Steam Remote Play. For met it doesn't work that well, but still better then any of Teamviewer, Splashtop, RDP and etc. The third option is Google Remote Desktop. It's better use as remote desktop, but games are still possible with it as well.

Hi, Are you playing locally or remotely ? Firstly, I may recommend following this guide to setup a gaming vm. Additionally try executing some benchmarks like CPU-Z internal benchmark, Unigine Valley or 3DMark Time Spy as GPU benchmarks and post results.

I've eventually found why the Gears 5 performance was so bad after host resumed from suspend mode. The reason was TSC clock source. Upon resume from sleep the TSC timer was turned off in favour of HPET. Feb 12 11:25:55 zeus-unraid kernel: TSC synchronization [CPU#0 -> CPU#1]: Feb 12 11:25:55 zeus-unraid kernel: Measured 108 cycles TSC warp between CPUs, turning off TSC clock. Feb 12 11:25:55 zeus-unraid kernel: tsc: Marking TSC unstable due to check_tsc_sync_source failed Feb 12 11:25:55 zeus-unraid kernel: TSC found unstable after boot, most likely due to broken BIOS. Use 'tsc=unstable'. ... Feb 12 11:27:43 zeus-unraid kernel: kvm: SMP vm created on host with unstable TSC; guest TSC will not be reliable The host switched clock source to HPET instead of TSC: root@zeus-unraid:~# cat /sys/devices/system/clocksource/clocksource0/current_clocksource hpet TSC is a complicated thing, more info can be read here Pitfalls of TSC usage And the only working solution for me was to add a kernel parameter tsc=reliable which tells kernel to skip TSC sync checks . I don't know what issues might be in the long run because of that, but so far I didn't notice anything bad to both host and guest os. I tried then to reboot and put host into sleep again and wake up. Immediately after that I checked that the host timer was still TSC. I run the Windows 10 VM and benchmarked Gears 5. I got ~70 FPS vs. 30 FPS before that. I've also noticed more modest performance improvements in other bechmarks like Division 2 : 83 FPS vs. 79 FPS. So far the Gears 5 was the only game I tested that benefit such big upon TSC fix. I've googled a timer benchmark TimerBench that is useful to test the Windows guest os timer performance. And it immediately shows the difference before and after:

Hi, I have similar build running Ryzen 3600. Did you try removing those lines: <synic state='on'/> <!-- https://mathiashueber.com/performance-tweaks-gaming-on-virtual-machines/ --> <stimer state='on'/> <!-- https://mathiashueber.com/performance-tweaks-gaming-on-virtual-machines/ --> Those lines enabled causes a great shuttering and lags in my Windows 10 VM. Another thing to look at is: <timer name='tsc' present='yes' mode='native'/> You have to check if you Linux host is using TSC timer indeed: $ cat /sys/devices/system/clocksource/clocksource0/current_clocksource tsc If you put your host system into sleep then upon resume the TSC timer will most likely be marked as unreliable and dropped in favor of HPET (at least this is what I observed on AMD hardware). TSC timer is crucial for Windows 10 VM to run smooth. If this is the case for you then you may try to add a kernel parameter 'tsc=reliable'. Additionally try running this software https://www.thesycon.de/eng/latency_check.shtml to monitor your DPC latency.

Additionally my current setup is Nvidia unRAID 6.8.0. Ryzen 3600, Nvidia GTX 1650 as primary GPU and Nvidia RTX 2060 Super as secondary which I pass to VM.

Hi everyone, I've made some investigations into this by myself. First I switched to bare metal setup and ran Gears 5 internal benchmark which showed solid 84 FPS at Ultra settings no v-sync of course. Then I try to stream with Moonlight from the bare metal Windows 10 to my another machine and I was surprised to see only 45 FPS... I checked my other games Just Cause 4, Witcher 3 and 3DMark Time Spy benchmark and noticed a significant performance loss when streaming with Moonlight/Steam/RDP this didn't matter. I've noticed that FPS monitor showed GPU usage at around 60-70% when streaming, however when playing locally it was always 90-100%. I should have mentioned that when I tried Unigine Valley benchmark it wasn't affected that way. OpenGL benchmarks weren't affected either. So I googled this issue and found many complaints from OBS users on FPS drop when OBS program is opened and even no streaming is ongoing. I tried many suggested tweaks like disabling all Game mode stuff, however what helped me was disabling WIndows 10 aero effects (I could not believe this at first but tried as a last hope) and it immediately worked. Streaming FPS for Gears 5 became like 80 FPS vs. 84 on local setup. Everything else I tested remotely also was fixed and GPU usage returned to 90-100%. I didn't investigate which Windows 10 visual effects were causing the issue, rather disabled them all as desribed here: https://www.windowscentral.com/how-disable-system-visual-effects-boost-performance-windows-10 However I suspect it has something to do with ALT-TAB app previews, may be. I disabled all Game mode stuff as well as advised. But only got an effect when disabled all visual effects. This wasn't the end of the story though. When I first start my headless VM and ran Gears 5 I happily saw 66 FPS where I previously see barely 30. Other games and benchmarks have improved as well. I put my unRAID into sleep than and start it next morning and couldn't believe it my Gears 5 were capped at 30 FPS and GPU usage was merely 30-40%. Everyhing else weren't affected and ran smoothly. I've figured out that when you restart the host then Gears 5 performance reverted back to normal. However once you put host into sleep mode and then run Gears 5 it's awfully running at 30-40% of GPU power. And the only way I've found to fix this is to restart the host. So I've put a lot of hours into googling the same issues and haven't found anything to fix it. Everything else I have installed is running with no issues now and only Gears 5 has this issue. So currently I gave up.

Hi, I'm running RTX 2060 Super and in 3DMark Time Spy the GPU performance tests is no difference to bare metal (default settings). Altough there is a CPU test as well and it's there where is a loss, and it depends on you VM configuration (how many cores you're passing and etc.) . Games normally got more performance downgrade because they're more depend on CPU & memory subsystems, but pure graphic benchmarks are mostly showing a bare metal performance.

Hi guys, Has anyone encountered the poor GPU performance after waking up from suspend (S3 sleep) ?

You may try to test your setup with Unigine Valley benchmark. It is not CPU binded and you GPU load must be close to 100% there. I have RTX 2060 Super as well on Ryzen 3600 and in this test in VM I'm getting around 100 fps average (streaming from headless VM) vs. 110 fps on bare metal Windows installation. If you have your monitor attached I believe the peformance hit would be even less. Although I'm experiencing different issues with my setup, related to game streaming, because I"m running a headless VM setup. Whatever streaming software I used (Steam/Moonlight/even RDP) my FPS immediately dropped by 50% and GPU utilisation was hardly 60% in certain benchmarks (3DMark) and games (Gears 5). However when I start investigation I've found the very same performance hit on my bare metal install as well. I partially solved my problem by disabling Windows 10 aero effects and all game mode stuff.

Hi, are you streaming your game or using a monitor?

Hi everyone, My host specs are: Ryzen 3600 cpu, Asus RTX 2060 Super 8Gb, 32Gb ECC RAM. I've setup a Windows 10 VM according to all best practices found here. I assign 4 cores of my 3600 to VM (not isolated). I pass through my second gpu which is RTX 2060 Super and was completely satisfied with overall VM performance until I ran Gears of War 4/5. I have a headless setup with either Stream/Moonlight streaming to the client machine. I benchmarked the system with Unigine superposition test and it showed 5-10% drop compared to baremetal when I have display attached and 10-20% drop when streaming. So I'm totally ok with this. I tested Just cause 4/Witcher 3 and they were able to run and MAX settings with > 60 fps when streaming , so no issues there as well. However with Gear of War 4/5 I get around 20-30 fps with recommended settings and FPS almost doesn't change when I set graphics either to LOW or ULTRA. When I running the game the GPU load is merely 30% even running on base clock 1500 Mhz (when I run other games the gpu is overcloked to 1900). This all makes me believe that there is a CPU limitation. However when monitoring the CPU load it's also low. Internal Gears of War 5 benchmark shows little GPU bound around 10% and the issue seems to be in CPU render latency. I tried almost all tweaks to my VM however this doesn't change anything. The CPU/GPU load is still low in this game and therefore FPS is low as well. Anyone had similar issue?

Try the solution from here: https://forums.unraid.net/topic/34889-dynamix-v6-plugins/?do=findComment&comment=522886 unraid doesn't support s3 sleep officially so it seems it won't put disks into standby again after wake up because it thinks they are already in standby.