|
|
Log in / Subscribe / Register

Ekstrand: Introducing NVK

Jason Ekstrand announces a new Vulkan driver for NVIDIA hardware on the Collabora blog. It seems to be off to a good start, but there is some work yet to do:

Normally, I would have submitted the merge request long ago. There are far more alpha-quality drivers already in Mesa. The problem is that we really need a new kernel uAPI to support Vulkan properly and I don't want to be stuck supporting the current nouveau uAPI for the next five years.


to post comments

Ekstrand: Introducing NVK

Posted Oct 4, 2022 21:14 UTC (Tue) by jhoblitt (subscriber, #77733) [Link] (7 responses)

I am honestly a bit conflicted about this. The NVidia binary drivers have long been a pita but also representated the best Linux GPU experience. On the other hand, this feels like rewarding NV's poor behavior by solving their support problems for them and potentially rewarding them with additional hardware sales.

I actually replaced an NV GPUs in primary desktop a few months ago with a Radeon card because of the frustrating 4 display limit for NV cards (yes, I know about Quadro cards, I used to run 2x for additional displays). It has been a painless experience and ffmpeg decode acceleration works fine. I haven't tried gaming but it sure seems that the NV monopoly is over.

Ekstrand: Introducing NVK

Posted Oct 5, 2022 3:42 UTC (Wed) by dxin (guest, #136611) [Link] (3 responses)

Fundamentally question is, do open/closed-source driver make (my and your) life better/worse.

Ekstrand: Introducing NVK

Posted Oct 5, 2022 8:19 UTC (Wed) by mfuzzey (subscriber, #57966) [Link]

Leaving aside the philosophy there are cases where open drivers are a huge practical improvement.

All closed drivers have the problem of platform availability. If the vendor only provides it for X86 and you want to use ARM or RISC-V you're out of luck.
They also have no guaruntee of support over time. When you're putting an embedded device in the field and need to support it for 10 or 15 years this can be a big issue.

There can also be toolchain issues. For example distribution may require certain build time options and complain (or refuse to run) binaries that don't respect that. Eg a number of years ago Android switched to wanting all binaries to be built with -PIE for security reasons. This is problematic for anything that can't be rebuilt.

For a "standalone" driver, implementing a single known and mostly stable interface it may not seem that important being mostly a question of the number of bugs and the level of trust we can have without source code.

However graphics drivers are not standalone; there is a whole, very complicated, fast moving, stack stradling both kernel and userspace and including things like X and Wayland.
One huge problem with closed userspace graphics drivers is that the are written for a single stack (ie a given kernel driver and a given higher level userspace framework like X11, Wayland, Android, ...). This means that the closed driver is often useless to those who want to use it outside of its original context.

Closed source graphics driver also reinvent everything each time whereas the open source equivalents have lots of common code meaning that, once the hardware specific parts are done they all tend to advance together.

Ekstrand: Introducing NVK

Posted Oct 5, 2022 8:58 UTC (Wed) by LtWorf (subscriber, #124958) [Link]

When using proprietary drivers kernel upgrades go from seamless to needing extreme care and supervision.

Ekstrand: Introducing NVK

Posted Oct 5, 2022 17:05 UTC (Wed) by jhoblitt (subscriber, #77733) [Link]

No doubt this will improve the experience for a huge number of users and it is valuable work. I am lamenting that we have a major GPU vendor holding the user base hostage while not committing the relatively few resources required to work with the community.

There may be a chance that the situation will change as GPU compute grows in popularity. I am hoping there is at least one data-center customer large enough for NV to care about (aka a "hyperscaler"), that isn't designing their own hardware, and which won't allow drivers which taint the kernel.

Ekstrand: Introducing NVK

Posted Oct 5, 2022 16:45 UTC (Wed) by flussence (guest, #85566) [Link] (2 responses)

I agree that we shouldn't reward the company, but at the same time we ought to acknowledge NV's current users are people too and not everyone can afford to exit on a whim; it's no better laying the blame on them for owning a GeForce than it is to blame all of someone's misfortune on living in Texas or Florida. It's going to be next to impossible in coming months to convince someone to buy a used NVidia card, what with Intel routing them with a better proposition at the mid-range and the second-hand market being tainted with a flood of hastily abandoned burned-out cryptomining trash.

Besides I think anything this does to improve NV's public image will be negated by the new 4-digit 4-slot cinderblock of death they recently announced. It seems obvious from their actions that the company is only interested in an audience with more money than sense. Caveat emptor to anyone who decides to buy NVidia *now*, but this effort will at least let existing owners weather their unfortunate situation, and it'll reduce the rate of e-waste in the long run.

And there's the most important motivation of all, having a FOSS driver humiliate the proprietary one in benchmarks.

Ekstrand: Introducing NVK

Posted Oct 5, 2022 20:12 UTC (Wed) by NYKevin (subscriber, #129325) [Link] (1 responses)

> and the second-hand market being tainted with a flood of hastily abandoned burned-out cryptomining trash.

You understate this problem in two different ways:

1. Miners probably tend to take better care of their GPUs than gamers (because miners are operating out of datacenters and gamers aren't), so the miners are actually selling a superior product compared to the average person.
2. Miners all abruptly stopped buying new GPUs at the same time, so now there's a supply glut of new GPUs that were "supposed to" go to mining, but were never even opened. On top of that, the 40x series just came out, so at least some retailers will probably try to sell off their remaining 30x cards quickly (IMHO a mistake, because nobody wants the 40x cards at current prices, but meh, inflation could obviate that problem, so we'll see what happens), which in turn implies discounts on unopened 30x cards.

The net result of these things is that 30x NVidia cards are currently a buyer's market, regardless of whether you buy new or used.

Ekstrand: Introducing NVK

Posted Oct 6, 2022 11:22 UTC (Thu) by paulj (subscriber, #341) [Link]

Also on 1, there isn't even good evidence to think ex-mining cards are less reliable than ex-consumer (i.e. gaming) cards. There's a tech channel on Yt who did some comparisons I think, between ex-mining and new and found no difference in reliability.


Copyright © 2022, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds