Live data from Hacker News

CUDA on Windows Subsystem for Linux 2

devblogs.nvidia.com

31–40 of 221 posts

Re: CUDA on Windows Subsystem for Linux 2

#31

Having set this up, this is going to be absolutely huge for ML. A lot of nonsense is getting cut down by this: 1. You don't have to install special CUDA-specific drivers that are behind the normal gaming drivers anymore. You'll soon be able to just use the Regular nVidia Drivers (and even now, all you need to do is install a Beta driver version). That's huge for someone just starting out, they don't have to have "the…

> without the technical issues of Desktop Linux (aka the Wayland Nightmare) Nobody is forced to use Wayland. It’s still quite experimental, I think only Gnome enables it by default.

Xming on Windows isn't exactly a walk in the park either -- I'd still rather use the native, proprietary build of VSCode on a proprietary OS with the SSH extension if I'm on a high-DPI screen.

Re: CUDA on Windows Subsystem for Linux 2

#32
Throwing out a random data point: Until now, NVIDIA has been very anti-virtualization (on the consumer side), going as far as to engineer its drivers to detect in-use virtualization (Hyper-V, Xen, QEMU, etc.) and fail on purpose [1]. I'm curious to see how they now handle this scenario (given WSLv2 runs in a virtual machine). Perhaps they just commented those checks out in their 'specialized' drivers, an interesting development for enterprising individuals looking to enable consumer GPU pass-through for general purpose virtual machines and containers.

[1] https://github.com/riverar/Remove-HypervisorChecks

Re: CUDA on Windows Subsystem for Linux 2

#34

This is BIG for machine learning adoption! 2 years ago it was a nightmare to setup everything on a Windows box, and wasn't working in any VMs (WSL or VirtualBox). Very excited about this.

This feels like it's gonna be terrible for people already doing ML. Now, the following has more than a bit of elitism, I'm aware of that, so please don't comment to point that bit out... I'm afraid that searching for answers to specific ML questions will start feeling like trying to google some Windows problems (every few years when I make the mistake of trying to help someone out) where it's all "download this, then…

Don't be too elitist and scared, machine learning developers using Cuda and WSL are far from the windows non technical average user.

You can take a look at the Github issues for Microsoft's projects related to WSL if you don't believe me. They are normal developers like you and me.

Re: CUDA on Windows Subsystem for Linux 2

#35
post #27

This is finally the year of the Linux desktop. However, it is running on Windows as WSL. With this support now for CUDA now, it can be argued that the best, most versatile developer experience is on Windows. You now have access to all the Windows specific tools (such as Visual Studio) as well as all the Linux tools in a very seamless environment.

Next step, Nvidia will support only WSL and not normal Linux? Good riddance, we need less blobs.

Actually, this might be a great price discriminator for NVidia. They have tried to differentiate their consumer and professional GPUs so they can charge different prices. I could see NVidia only providing WSL/Windows drivers for their consumer GPUs. If you want native Linux, then you would have to buy a professional GPU at higher cost.

By great, I mean great for NVidia, not for us.

Re: CUDA on Windows Subsystem for Linux 2

#36

Pity. CUDA was/is proof positive that Linux on the desktop is perfectly feasible and that you can use it to both do a UI and do meaningful computation on the same machine without getting tied down into all kinds of licensing schemes. Opening this further up to Windows gives fewer people a really good reason to try out Linux as their daily driver. I personally don't get why any developer would prefer Windows over Linu…

First a developer you don't do only programming: you maybe have to work with documents, do a presentation, a spreadsheet and thus use Office. You maybe have to make some icons or logos and thus you need Adobe Illustrator. You maybe need to connect to a printer like the one that we have at my office that good luck doing that on Linux. You maybe need a CAD software to designa PCB, you need whatever other software that is Windows-only.

Second there are situation where as a developer you are forced to use Windows. A lot of SDKs, compilers, debuggers, programmers for microcontrollers, FPGA, etc runs only on Windows, and if you have to do with firmware development you are basically forced to either use Windows or spend a week just to make a LED blink (if you manage to do that).

Same reason for using macOS: I use a Mac mainly because I need to make iOS software and the only way is to use a Mac. By the way macOS to me is a good compromise, you have all your UNIX tools in a system that you can use to run also proprietary software.

On my server I use Linux, even if recently I started using more and more FreeBSD and I'm liking it a lot (recently they added too much useless bloat in Linux like systemd and other crap, while on FreeBSD you have a simple system that is stable as a rock)

Re: CUDA on Windows Subsystem for Linux 2

#37
post #31

Earlier quoted context omitted.

> without the technical issues of Desktop Linux (aka the Wayland Nightmare) Nobody is forced to use Wayland. It’s still quite experimental, I think only Gnome enables it by default.

Xming on Windows isn't exactly a walk in the park either -- I'd still rather use the native, proprietary build of VSCode on a proprietary OS with the SSH extension if I'm on a high-DPI screen.

https://x410.dev/ is trivial to set up, and once done, it "just works".

But yeah, with VSCode WSL remoting, there's just no particular reason to do that anymore.

Re: CUDA on Windows Subsystem for Linux 2

#38
post #11

Earlier quoted context omitted.

This feels like it's gonna be terrible for people already doing ML. Now, the following has more than a bit of elitism, I'm aware of that, so please don't comment to point that bit out... I'm afraid that searching for answers to specific ML questions will start feeling like trying to google some Windows problems (every few years when I make the mistake of trying to help someone out) where it's all "download this, then…

Hope we don't see things like: "Google announces today that to work around Windows' file path length limitations, Tensorflow's directory structure is being rewritten to use three characters or less per directory."

What sold me on NT 3.1 was the in-the-box Microsoft Transaction Server.

MTS is the single largest reason why SQLServer_Linux was a shoehorn refit, not anything like the job that it could have been.

Out of the box CICS interop was bigger than my memory of the first browsers for W/NT. Want to trade with the world and speak EDIFACT, or do X.500 real money movement? You had codes running before you could put the screws on your local IBM account rep to get out of bed for you.

I last looked at HN just as Lockdown was announced in England. The last discussion I perused was showing a project that provided secure remote access to sensitive ports. Which I think is stable CICS fayre since about ‘85 on the cross platform side I would guess at least twenty years older for anything/360.

Re: CUDA on Windows Subsystem for Linux 2

#39

Earlier quoted context omitted.

Other than C# and other Microsoft specific stacks, I've found coding on Windows to be absolutely horrible compared to Linux.

Can you be more specific? What exactly is horrible on Windows when you've tried coding?

- Powershell sucks

- Graphical installers / no package manager

- No search results / docs for most languages

- Weird bugs due to being second class for most languages

- Have to use docker in VM

- Bad env for writing scripts

- We deploy on Linux

Re: CUDA on Windows Subsystem for Linux 2

#40
post #31

Earlier quoted context omitted.

> without the technical issues of Desktop Linux (aka the Wayland Nightmare) Nobody is forced to use Wayland. It’s still quite experimental, I think only Gnome enables it by default.

Xming on Windows isn't exactly a walk in the park either -- I'd still rather use the native, proprietary build of VSCode on a proprietary OS with the SSH extension if I'm on a high-DPI screen.

Microsoft has also planned to provide a remoting tool based on RDP: https://devblogs.microsoft.com/commandline/the-windows-subsy....
Post reply on HN