4 ms·

I'm happy to hear that gaming works well, and apparently my decidedly older CPU (AMD Phenom X4 965) supports the technology as well, which is interesting to lea

by paler 11y ago

I'm happy to hear that gaming works well, and apparently my decidedly older CPU (AMD Phenom X4 965) supports the technology as well, which is interesting to learn. However, my main reason for needing Windows is for music recording and mixing via Cubase and other DAW software. This requires really low-latency access to my USB and PCI sound cards. Does anyone have experience with running applications like this in a VM? Is it feasible? I would be thrilled if I could finally move to Linux or FreeBSD entirely and just keep a Windows 7 VM, all sandboxed up with no Internet access, just for the music recording.

cmdrfred 11y ago
I have found that a lot of the audio editing software you can get away with quite low specs. I record my podcasts on a 8 year old laptop running Windows XP. It rarely connects to the internet and I record to a thumb drive that is backed up so there is nothing to lose on it. In this particular case something similar might be the best option. For pod-casting anything over 2 or 3 ms in my headphones throws me off my game.
- plaguuuuuu 11y ago
  Unlike podcasts, music recording and mixing is extremely CPU intensive. Especially if you run a lot of analog simulating synths and stuff.
  - cmdrfred 11y ago
    I stand corrected.
  - jsolson 11y ago
    Virtual CPUs with VTx can be made to run at essentially native speed for most workloads on most fairly modern CPUs. Memory performance can be subpar, although there's public documentation on techniques that can improve that: https://help.ubuntu.com/community/KVM%20-%20Using%20Hugepages https://help.ubuntu.com/community/KVM%20-%20Using%20Hugepage... I/O performance is a whole other story, dependent on many workload and hypervisor specific factors.
ferrari8608 11y ago
Really, the 965 supports it? I have the 970 and was under the impression the Phenom II's in general just don't support it. Can you show me where you found that information?
- paler 11y ago
  https://en.wikipedia.org/wiki/List_of_IOMMU-supporting_hardware#CPUs_2 https://en.wikipedia.org/wiki/List_of_IOMMU-supporting_hardw...
fpgaminer 11y ago
For lack of a response from someone who has actually tried getting ASIO working on such a setup, I suggest you just try it. Grab a Live CD and a spare external hard drive and just give it a whack. Passing through the sound cards should work. USB might be trickier if you're not just passing through a whole USB hub. But considering how new IOMMU is, I imagine the intersection between people who have tried it and people who use Cubase is incredibly tiny.
jsolson 11y ago
The simplest thing to do of course is just try it. You can install Windows in a VM without a product key and try it without limitation for something like 30 days. That said, "really low latency" is relative. Device assignment will give you excellent I/O performance in terms of sending data to/from the device, particularly for bulk DMA transfers. Where you (likely) still pay a penalty is in terms of interrupt delivery, as on most chips these are still routed through the host even in "passthrough" mode. Even so, you're looking at delays at the outside on the tens of microseconds scale, depending on which microarchitecture you're on and how well tuned the fiddly bits of your hypervisor (presumably KVM for most users) are. Some configurations will be much better behaved here. If your audio hardware goes into polling mode (or can be made to do so) when it is recording you can potentially avoid the interrupts entirely. (source: I work on Google Compute Engine, in particular on virtualized networking, historically on performance. Virtual interrupt delivery latency is a topic near and dear to my heart, although I don't know a damned thing about audio gear and its tolerances there :)
anonbanker 11y ago
I personally install KXStudio[0] for guys just like you almost monthly. here's some appropriate[1] pornography[2]. Imagine being able to have a skype call with someone, and be able to record their vocals (provided they have a good mic) apply effects, and then record _and_ pipe the vocals back to the vocalist in realtime. Or make recording snippets of audio off youtube trivial. Or stream live instruments in other rooms over wifi/ethernet using only a raspberry pi and a good microphone. Or stream 8 of those tracks/instruments (on different devices) to one computer for mastering/final mixing. It's audio nerd nirvana, basically. What you need: Wine Wineasio (for Wine support in FL Studio or others) KXstudio 1. http://kxstudio.linuxaudio.org/index.php http://kxstudio.linuxaudio.org/index.php 2. https://www.youtube.com/watch?v=GBD-wsMiZCk https://www.youtube.com/watch?v=GBD-wsMiZCk 3. https://www.youtube.com/watch?v=PhfmGGPcwNc https://www.youtube.com/watch?v=PhfmGGPcwNc

by paler 11y ago

cmdrfred 11y ago

I have found that a lot of the audio editing software you can get away with quite low specs. I record my podcasts on a 8 year old laptop running Windows XP. It rarely connects to the internet and I record to a thumb drive that is backed up so there is nothing to lose on it. In this particular case something similar might be the best option. For pod-casting anything over 2 or 3 ms in my headphones throws me off my game.

plaguuuuuu 11y ago
Unlike podcasts, music recording and mixing is extremely CPU intensive. Especially if you run a lot of analog simulating synths and stuff.
- cmdrfred 11y ago
  I stand corrected.
- jsolson 11y ago
  Virtual CPUs with VTx can be made to run at essentially native speed for most workloads on most fairly modern CPUs. Memory performance can be subpar, although there's public documentation on techniques that can improve that: https://help.ubuntu.com/community/KVM%20-%20Using%20Hugepages https://help.ubuntu.com/community/KVM%20-%20Using%20Hugepage... I/O performance is a whole other story, dependent on many workload and hypervisor specific factors.

ferrari8608 11y ago

Really, the 965 supports it? I have the 970 and was under the impression the Phenom II's in general just don't support it. Can you show me where you found that information?

paler 11y ago
https://en.wikipedia.org/wiki/List_of_IOMMU-supporting_hardware#CPUs_2 https://en.wikipedia.org/wiki/List_of_IOMMU-supporting_hardw...

fpgaminer 11y ago

For lack of a response from someone who has actually tried getting ASIO working on such a setup, I suggest you just try it. Grab a Live CD and a spare external hard drive and just give it a whack. Passing through the sound cards should work. USB might be trickier if you're not just passing through a whole USB hub. But considering how new IOMMU is, I imagine the intersection between people who have tried it and people who use Cubase is incredibly tiny.

jsolson 11y ago

The simplest thing to do of course is just try it. You can install Windows in a VM without a product key and try it without limitation for something like 30 days. That said, "really low latency" is relative. Device assignment will give you excellent I/O performance in terms of sending data to/from the device, particularly for bulk DMA transfers. Where you (likely) still pay a penalty is in terms of interrupt delivery, as on most chips these are still routed through the host even in "passthrough" mode. Even so, you're looking at delays at the outside on the tens of microseconds scale, depending on which microarchitecture you're on and how well tuned the fiddly bits of your hypervisor (presumably KVM for most users) are. Some configurations will be much better behaved here. If your audio hardware goes into polling mode (or can be made to do so) when it is recording you can potentially avoid the interrupts entirely. (source: I work on Google Compute Engine, in particular on virtualized networking, historically on performance. Virtual interrupt delivery latency is a topic near and dear to my heart, although I don't know a damned thing about audio gear and its tolerances there :)

anonbanker 11y ago

I personally install KXStudio[0] for guys just like you almost monthly. here's some appropriate[1] pornography[2]. Imagine being able to have a skype call with someone, and be able to record their vocals (provided they have a good mic) apply effects, and then record _and_ pipe the vocals back to the vocalist in realtime. Or make recording snippets of audio off youtube trivial. Or stream live instruments in other rooms over wifi/ethernet using only a raspberry pi and a good microphone. Or stream 8 of those tracks/instruments (on different devices) to one computer for mastering/final mixing. It's audio nerd nirvana, basically. What you need: Wine Wineasio (for Wine support in FL Studio or others) KXstudio 1. http://kxstudio.linuxaudio.org/index.php http://kxstudio.linuxaudio.org/index.php 2. https://www.youtube.com/watch?v=GBD-wsMiZCk https://www.youtube.com/watch?v=GBD-wsMiZCk 3. https://www.youtube.com/watch?v=PhfmGGPcwNc https://www.youtube.com/watch?v=PhfmGGPcwNc