Bickering Continues About NVIDIA Using DMA-BUF

[mememe]

/siren

the 'market share' argument

1% of computers run gnu/linux or some variant of gnu/linux - this is utter bullshit

gnu/linux runs on all kinds of systems all over the world from set top boxes to webservers to super computers

lets flip this round for a moment...

if you count any electronic device with a cpu as a computer what is the percentage of these systems that run say a microsoft operating system?

and out of the top 100 most powerful computers in the world what percentage of these run gnu/linux?

the 1% figure comes right out of the m$ propaganda department as they only count consumer P.C.s and laptops

Except that set top boxes, webservers and supercomputers don't use Nvidia chips and drivers. Neither AMD nor Intel have really opened the guts of their proprietary drivers but instead opted to support the Linux community through documentation. We already have relatively decent drivers provided by the Nouveau project - the problem is that the graphics stack is pretty far behind when it comes to supporting recently introduced OpenGL stuff, like tesselation. We're still stuck with GL 3.0, and we're still a far cry from reaching the performance of the proprietary drivers.

[D0pamine]

Except that set top boxes, webservers and supercomputers don't use Nvidia chips and drivers. Neither AMD nor Intel have really opened the guts of their proprietary drivers but instead opted to support the Linux community through documentation. We already have relatively decent drivers provided by the Nouveau project - the problem is that the graphics stack is pretty far behind when it comes to supporting recently introduced OpenGL stuff, like tesselation. We're still stuck with GL 3.0, and we're still a far cry from reaching the performance of the proprietary drivers.

and the likelyhood of nvidia becoming a supplier of hardware to set top boxes/supercomputers/ect is greatley diminished because of the lack of documentation

nobody is suggesting that nvidia, amd, intel and any other hardware mfr should write open drivers or open up their existing code but just give documentation to allow the community to develop their own drivers this is what happens anyway through reverse engineering however you cant expect 0 day support from reverse engineered development nor can you expect it to perform as well as it might

and like i said - what software are you running that requires anything above opengl3.x ... i can only think of a couple of things that require even opengl 3.0

[blackiwid]

Except that set top boxes, webservers and supercomputers don't use Nvidia chips and drivers. Neither AMD nor Intel have really opened the guts of their proprietary drivers but instead opted to support the Linux community through documentation. We already have relatively decent drivers provided by the Nouveau project - the problem is that the graphics stack is pretty far behind when it comes to supporting recently introduced OpenGL stuff, like tesselation. We're still stuck with GL 3.0, and we're still a far cry from reaching the performance of the proprietary drivers.

as far as I know opengl 4 is done now in mesa, because of that they raise their major version number. And of course that works with a free driver, and be it only with the intel one and not the amd one? So your general statements are just not true.

about the speed problems, thats also not true, the intel driver is very fast it gives you nearly the full speed this hardware is capable to produce so even again a general statement is not true in this cases.

But one thing we see on the amd side, if you are not commiting yourself to the opensource way, and use 99% of the developers to make the closed one better, yes the free one are not so good than the closed one. So just commit to it and you get a really good driver (intel).

ok they have some chipsets they did not release but thats no problem just dont buy this anti-freedom-hardware. and they will not sell it again.

[Ex-Cyber]

as far as I know opengl 4 is done now in mesa, because of that they raise their major version number.

They just hit 3, not 4. According to the todo list in the repo, most of the 4.0 work hasn't even been started yet.

[XorEaxEax]

1. Nvidia says "Fuck you" and doesn't care about Optimus on Linux anymore -> Bad for the users

Bad for the minority of Linux users who bought NVidia optimus, it's not as if NVidia's lack of Linux support for optimus is a state secret.

It should be clear by now that if you want a laptop running Linux you are best off choosing an Intel solution.

Also I've seen no confirmation from NVidia that they are actually going to support optimus on Linux even if they could somehow use DMABUF.

2. Nvidia decides to implement something in their blob to work around the issue -> Even more propietary things used in the end, instead of a well documented kernel interface

Which is what they've always done, you know for those proprietary driver which all NVidia fans are saying runs so great, so what's the problem? Simply that NVidia are not so eager to support their own technology on Linux when it's not targeting those Linux markets where they make money (hint! not the end user desktop and certainly not laptops).

So, right now, please tell me, who benefits from the "dma-buf" block ???????????????????

The Linux kernel and Linux users benefits since this is a much larger issue than 'optimus'.

If the kernel developers makes it easier to maintain proprietary drivers then we will see an increase of companies only providing proprietary drivers, and those only for architectures said vendors find 'worth their while' rather than supplying/assisting with open source drivers, leading to less open source drivers and less value for Linux as a working-everywhere-on-anything kernel.

And that is ignoring all the other problems proprietary drivers bring along, like security and stability.

Are those who for some reason have bought an optimus powered laptop on which they want to run Linux (despite NVidia going out and saying 'we have NO plans of supporting optimus on Linux') worth jeopardizing hardware vendors returning to providing proprietary blobs rather than open source drivers in-tree as it's no longer such a hassle to do so? Nah, I don't think so, and more to the point it seems that most kernel devs in question doesn't think so either.

'Optimus' is a small blip in computing history and only pertains to a computer segment in which Linux barely even registers, yet some people are trying to portray optimus as some 'make it or break it' thing for Linux, it's quite hilarious.

[D0pamine]

isnt optimus just a crappy solution to the performance hit caused by compositing ?

[Sonadow]

blah blah blah...

wake me up when Nouveau can deliver something like THIS on ULTRA settings on any 8-month-old game:

[D0pamine]

blah blah blah...

wake me up when Nouveau can deliver something like THIS on ULTRA settings on any 8-month-old game:

WOW thats impressive i'm going out right now to buy a nvidia e-penis and then i'll be able to masturbate like a frenzied monkey too!!

EDIT - isn't that batman? I didn't realise that batman was available for gnu/linux

[Sonadow]

EDIT - isn't that batman? I didn't realise that batman was available for gnu/linux

Lay off the sarcasm. Anyway, it seems that you didn't get the point I was trying to make.

Everybody knows that the Nvidia blob for Linux shares code with the Windows blob so performance is mostly similar across the board. IF Batman was available for Linux the Nvidia blob will have generated similar results, give or take 10-20 fps here and there (that bench was done using DirectX 9 btw, not DX 11).

My point is that unless Nouveau can offer comparable framrates, it's not going to attract any user who puts performance over freedom, especially where high-end cards are concerned (they are high-end cards precisely because you WANT their superior performance). Heck, the Nouveau driver does not even need to offer performance parity; I'd bet most people will be more than satisfied if it could provide at least 60-65% of the blob's performance.

[D0pamine]

Lay off the sarcasm. Anyway, it seems that you didn't get the point I was trying to make.

Everybody knows that the Nvidia blob for Linux shares code with the Windows blob so performance is mostly similar across the board. IF Batman was available for Linux the Nvidia blob will have generated similar results, give or take 10-20 fps here and there.

My point is that unless Nouveau can offer comparable framrates, it's not going to attract any user who puts performance over freedom. Heck, the Nouveau driver does not even need to offer performance parity; I'd be more than satisfied if it could provide at least 60-65% of the blob's performance.

the nouveau driver has been created by the community ie people like yourself but who choose not to spend their time helping the dark knight trudge his way round the unreal 3 engine with absolutely no help or input from nvidia - it has been reverse engineered!

i'd wager the r600g could easily reproduce this even in wine

and no - sarcasm is what i'm good at!