Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Which NVIDIA graphics card do you own? I’m surprised Sway doesn’t support certain graphics cards.

I believe the goal of wlroots is to solve this problem:

> wlroots provides backends that abstract the underlying display and input hardware

https://github.com/swaywm/wlroots



Sway generally doesn't support NVIDIA, or it's rather NVIDIA that doesn't support common kernel modes that Wayland needs.

Details here: https://drewdevault.com/2017/10/26/Fuck-you-nvidia.html


> Every GPU vendor but Nvidia supports these APIs.

> About a year ago Nvidia announced “Wayland support” for their proprietary driver. This included KMS and DRM support (years late, I might add), but not GBM support. They shipped something called EGLStreams instead, a concept that had been discussed and shot down by the Linux graphics development community before.

That’s pretty damning for Nvidia. I can see why the developer is so upset. Must be tough working on OSS in that type of hardware environment.

Looks like AMD is the way to go for desktop Linux boxes going forward. Although it looks like both Dell XPS [1] and System76 [2] is still using NVIDIA. Purism Librem uses the Intel i7’s embedded GPU [3]. I’m curious what the ideal Linux laptop is these days…

[1] https://www.dell.com/en-ca/work/shop/cty/pdp/spd/xps-15-9570...

[2] https://system76.com/cart/configure/oryp5

[3] https://shop.puri.sm/shop/librem-15/


The ideal Linux laptop is one with just Intel integrated, and no discrete GPU. Honestly, Intel GPUs are decent enough, even if you want to do some mild gaming. Handing two GPUs on a laptop in Linux doesn't have the best support. A lot of the Optimus/Bumblebee stuff is not very well maintained.


True, I don't do any laptop gaming. Intel GPU's seem to be sufficient as long as it can easily power a full HiDPI internal monitor + 4k external monitor... which Librem's Intel 620 GPU is listed as supporting but it looks like Librem's 15 v4 only has an HDMI port and not usb-c/displayport, which limits it to 4096x2160@30Hz instead of 4096x2304@60Hz. So I guess you'd have to look at Lenovo, HP etc for top tier..

Sadly my work uses all Macbrook Pros, which are amazing machines), but I'm scared to install Archlinux on them. The Archlinux Macbook wiki pages are 4yrs+ old.


Don't overestimate the Intel GPUs. My current system uses a 530 which is getting pushed to its limits trying to render the desktop at 60 Hz, either UHD on AC or FHD on battery. Forget UHD on battery, that doesn't even manage to move the cursor at more than 10 fps.

IMO the ideal GPU for a UHD screen would be either an Iris or a Vega 10. But it's difficult to find the former outside MacBooks and the latter with more than an FHD screen.


> even if you want to do some mild gaming

Just to back this up, I've recently bought a non-touchbar 13" MBP from 2016, and been amazed at the amount of games I've been able to play no issues on the Intel graphics in it.


Can you list some of them? Since my MBP with much faster Radeon 460 is struggling even with new isometric games like Pillars of Eternity 2 or games like Stellaris.


Last I used Optimus my CPU desoldered itself.

Gaming on Linux was vaguely toleable while it lasted but I'm back to Windows.


> Looks like AMD is the way to go for desktop Linux boxes going forward.

Problem is: desktop Linux has an irrelevant market share, and the trivial workaround (use intel IGP or go buy and AMD graphics card) do not work when you don't just manage your one PC but actually maintain a system that is produced in series...

And honestly being upset at nVidia leads nowhere. The architecture of Wayland is complete bullshit to begin with and I'm not sure any other graphic stack of competing systems went that way, EVEN for systems that have a single window manager (well, that's most of other systems... -- so the situation is strangely kind of reversed). So hopefully wlroots will eventually improve the situation or something like that, but is it even convenient to build something that does not link the graphics session to the low level graphics stack with that? I hope it is, because if not, that is hopeless... :/

And I've not even talked about extra deps like Cuda, or the general lead nVidia still have in perf/W and/or perf/price (IMO AMD is close, but no cigar)

The X proto was maybe shit, but at least the general approach did not lead to the kind of insanity we are currently seeing. And people discussing the Death of X when Wayland ecosystem is maybe still 5 years away from decent quality and some reasonable feature completeness... (remoting, anybody? anything good enough for Wine?)

Oh well, we will see...


It wasn’t just Wayland that used GBM, plus every other vendor supports it but Nvidia and it’s well documented in Mesa.

But I personally don’t care as long as Intel and AMD are available, it’s fine. I see Linux like MacOS, it’s an operating system targeted at a limited set of hardware. As long as you buy within this group 90%+ of the problems with Desktop Linux will go away.


The HP x360 with a Ryzen 2500u is rather good, and can be had for ~$400 on eBay.

It did take a full year for all the bugs to be sorted out (random kernel panics & freezing), these same bugs were problematic in Windows 10 tho. Neither OS had good platform support for quite some time...


Nvidia's drive doesn't support the GBM API on Linux, and the maintainer of Sway doesn't want to support Nvidia's preferred EGLStreams API.


Last I checked, sway detects if nvidia.ko is loaded (even if the nvidia gpu has no display) and errors out with some message to the effect of "f you for buying nvidia."


I grepped wlroots for "nvidia" and got no hits.



If this check were not here, Nvidia would still not work. This serves to reduce our bug report volume. However, by deleting these few lines of code (or specifying the appropriate command line argument), Nvidia could work tomorrow if they shipped GBM support.


If that check were not there, would intel for display + nvidia for compute work? Better chance than now.


The flag is there for you to override if you think you know better. We don't answer questions or provide support for any use-case with the proprietary driver.


Refusing to run if a particular kernel module is loaded (even if you don't touch it at all) is overstepping. You should at least provide an override flag.


There is an override flag, its linked in the code snippet.

Use "--unsupported-gpu" or "--my-next-gpu-wont-be-nvidia" to bypass the check and allow sway to launch.


Ah, er, jolly good then. Carry on.


I applied the `--my-next-gpu-wont-be-nvidia` flag to my hardware and replaced my Nvidia cards with AMD ones, and it is a quality of life improvement.


"--my-next-gpu-wont-be-nvidia"

--ill-pass.


It's implying I have a choice in the accelerators I could rent in a cluster... Yes, it's about using an intel GPU for display and a nvidia GPU to provide local development (latency being the primary reason).


It's just generally taking a reasonable stand ("I don't want to spend my time on Nvidia, especially not for free") into an unreasonable one ("Users who have Nvidia are bad and should be bullied").


Have you tried building from source and editing this code to not invoke?




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: