Kernel Developers Work To Block NVIDIA "GPL Condom" Effort Around New NetGPU Code

[ssokolow]

Quite frankly, I've never seen much issues. I don't game on Linux though.

What's your typical uptime? I'm used to getting at least two months of uptime on a single login session... though I do have to turn off desktop compositing to get that. I've never found a compositor that reacts well to being plumbed into the GPU for more than two or three weeks without a process restart.

RX 460 and 550-60 for example are capable of more 3D processing than GT 1030, but if you are using them just to drive screens and basic 3D for a GUI they won't heat up enough to even spin their fans.

Are those available in half-length? When I bought this mobo, I wasn't paying proper attention, some of my SATA connectors are off the end of the PCI-E x16 slot, I'm using all of them, and I don't plan on replacing my pre-UEFI, pre-PSP loadout until the mobo or CPU die.

[oiaohm]

Are those available in half-length? When I bought this mobo, I wasn't paying proper attention, some of my SATA connectors are off the end of the PCI-E x16 slot, I'm using all of them, and I don't plan on replacing my pre-UEFI, pre-PSP loadout until the mobo or CPU die.

I am using SAPPHIRE PULSE RX 570 ITX 8G G5 on a test bench because I wanted the bench with something reasonable short and light because it gets moved a lot. There is a 4G version of that card as well. Yes I also did not want to have to install Nvidia drivers for test bench stuff. You do loss a little performance going the ITX form factor of course but that RX 570 ITX still beats the crud out of all the GT1030 cards on the market even the 4G version.

Longest my testbench has been left on is 4 months straight with desktop compositing so it stable. Yes this card did replace the prior Nvidia card I had for test benchwork.

Being a test bench I am doing ctrl-alt-f1 and the like a bit as well that is a fairly good way to have Nvidia at times screw you over and that problem is solved by the AMD card.

There are short 550 and 560 as well out there in the market but I cannot state how good they work only the one I have.

[starshipeleven]

What's your typical uptime? I'm used to getting at least two months of uptime on a single login session... though I do have to turn off desktop compositing to get that. I've never found a compositor that reacts well to being plumbed into the GPU for more than two or three weeks without a process restart.

I don't usually leave my PC on for more than a few days as I have a home server for anything that needs to stay always on.
I did leave the PC on for somewhat between one and 2 weeks a few times with no ill effects, with KDE and Xorg.

Are those available in half-length? When I bought this mobo, I wasn't paying proper attention, some of my SATA connectors are off the end of the PCI-E x16 slot, I'm using all of them, and I don't plan on replacing my pre-UEFI, pre-PSP loadout until the mobo or CPU die.

most RX460 and 550-60 were using a PCB as big as the PCIe slot anyway (or it was longer for no real reason, like some cases where it was only carrying the 12V lines for an useless PCIe power connector). For most cards it's only a matter of how big the heatsink is. And that's why I said they can go fanless if not under load. A dual-slot heatsink is oversized.
The easier to find for me are Sapphire Pulse ones, also from XFX (look more or less the same). https://www.techpowerup.com/gpu-spec...e-rx-550.b4471
(it is kind of dual slot, in the sense that it needs 2 slots free even if technically it says it isn't)

The best for my usecase would have been the fabled single-slot XFX cards https://www.xfxforce.com/gpus/amd-ra...le-slot-design
but of course I didn't get one before they went out of stock because of reasons, so I had to scour ebay for broken ones and do heatsink swaps on Sapphire Pulse cards (I only needed to drill a bit the solid aluminum plate of the heatsink under the fan to make space for a capacitor, who cares, it's a piece of metal).

Still, you don't need to limit yourself to lower-end cards.
I had the same issue in the past with poor Sata port placement and I bought some special slim and angled Sata cables from Silverstone that fit under the GPU https://www.silverstonetek.com/product.php?pid=445
They are still available, at least from EU sellers in ebay and Amazon

[V1tol]

Thank you, guy from Facebook. I really started to enjoy gaming on my Nvidia laptop on Linux but looks like Nvidia can be banned completely And my notebook will become a potato. At least I can sell it to some kid who will use Windows and buy at least something AMD-based (if I find one of course).

[MrRtd]

Perhaps Nvidia should just stop updating their Linux driver.

They could as long as they're fine with losing customers. I know if they did that, my next gpu would definitely be AMD, and I say this as a long time nvidia customer.

[kpedersen]

They could as long as they're fine with losing customers. I know if they did that, my next gpu would definitely be AMD, and I say this as a long time nvidia customer.

Well Nvidia does stop updating their drivers. They just call it "maintenance mode". Which is a little bit fraudulent because they don't maintain it at all. Not even to remain compatible with current Xorg or Wayland compositors.

[bridgman]

Its not trade secrets. There is actual trademarked code within the NVidia blob that they cannot legally release as open source and Nvidia would have to go through the entire driver, remove and then replace all such code and due to the driver having a history of 20+ years its probably gotten to the point where they can't accurately identify what code is trademarked and what isn't anymore (this is due to the nature of software development).

The code would be "copyrighted by someone else and licensed under terms that prohibit publishing source" rather than "trademarked", but the rest of your sentence accurately describes the challenges with open sourcing code that was not written as open source from the start. It also exactly describes what we had to do in order to publish an open source version of our Vulkan driver (aka AMDVLK).

Note that this isn't specific to NVidia, AMD had the same problem with bootloaders. The only real way to get around this is to make a complete clean room implementation of a driver (which AMD did a decade ago for their graphics stack because they were completely desperate at the time however it took like 8 years for their open source drivers to become generally usable).

We didn't actually make a new driver; we leveraged existing open source driver efforts using the radeon KMS driver framework. If you mean "wrote new driver code as support for newer chips was added" then I agree.

The radeon driver code was then refactored to create the amdgpu driver with two goals - organizing around HW IP blocks as part of a larger corporate initiative ("SOC15") and being able to support our closed source userspace drivers.

The decision to work with a separate code base was not made "because we were completely desperate" but because sanitizing and refactoring the existing driver code to fit with upstream practices would have been even more work than building and maintaining a separate code base. This was particularly true for the kernel driver, since most of the associated functionality in the existing driver was userspace code written in C++, and the Windows team did not want to move back to C.

Note that there was no "clean room" requirement on the open source driver development - we had access to the closed source driver code although the sheer size of the closed source stack made it more difficult to extract useful knowledge from it than one might reasonably assume, so a lot of the code was written based on internal hardware design docco rather than the existing drivers. Key point though is that we dealt with IP issues via IP review and cleanup, not by clean room development practices.

One other big benefit of implementing new code was that it allowed us to perform IP review incrementally over several years rather than having to do one big-bang effort like we needed for AMDVLK.

[kpedersen]

.

That was a really insightful and interesting post. It seems to me that AMD's goals are certainly driven by business but are not sleazy or secretive.
I wish this kind of informal information was available from other companies too rather than creating this childish facade of a members only social club

[starshipeleven]

Thank you, guy from Facebook. I really started to enjoy gaming on my Nvidia laptop on Linux but looks like Nvidia can be banned completely And my notebook will become a potato. At least I can sell it to some kid who will use Windows and buy at least something AMD-based (if I find one of course).

AMD laptops should be in the works for most manufacturers now that Ryzen 4000 (the laptop CPUs from AMD) are thoroughly besting anything from Intel

[starshipeleven]

They could as long as they're fine with losing customers. I know if they did that, my next gpu would definitely be AMD, and I say this as a long time nvidia customer.

No offence, but you and anyone using NVIDIA on Linux in his own PC are irrelevant.

NVIDIA cares about businness customers on Linux, consumers are an afterthought.