NVIDIA Announces New Legacy Linux Support

[Gusar]

I was referring to this article: http://www.phoronix.com/scan.php?pag...tem&px=MTE3MzY

I don't see how what's in that article is connected to the FU, despite what the last sentence says. correlation != causation
What makes sense: a) Nvidia is looking into supporting Optimus because Linus showed them the middle finger, or b) Nvidia is looking into supporting Optimus because the graphics stack is now capable of it thanks to dma-buf and xrandr1.4?

I'm pretty sure xrandr 1.4 never existed. I believe xorg decided not to release that and skipped to xrandr 1.5.

What was initially meant to be 1.4 was scraped, so now the output/offload slave stuff required for Optimus is 1.4. There will be no number-skipping.

Other than the potentially false remark here about licensing issues, it would seem to make sense that they implement the kernel one and remove their custom one since it would reduce bloat in their driver from a programmer's perspective.

Why move from an established working solution they have to something new, when there aren't apparent benefits? What would be the incentive to invest in the effort? Also, have you seen the size of the Nvidia driver? I don't think they care about avoiding bloat.

Switching from twinview to xrandr1.2 is different, it means existing DE tools for monitor configuration now work. Also, twinview didn't support rotating a single display in a multi-monitor scenario, whereas xrandr1.2 does. There might be a few other benefits that I can't recall currently. There are no such benefits in switching kernel modesetting implementations.

Somehow it doesn't seem all that tiny since linux runs on a lot of embedded devices (tablets, smartphones, etc).

Yeah, but we're talking here about desktop Linux.

In terms of desktop users (laptops too), it also doesn't seem that small.

But it is. Thinking otherwise would be naive. Nvidia supports the Linux desktop at all only because of two reasons - they can reuse a lot of stuff from the Windows driver, and there are high-paying customers in the workstation space. The effort required to then support the desktop after you already have a workstation driver isn't that much.
Though sometimes we do get goodies that appear desktop-only (VDPAU), desktop Linux is nevertheless low priority for Nvidia. Because the market share simply isn't there.

[logistiker]

Please elaborate on these "things".

I was referring to this article: http://www.phoronix.com/scan.php?pag...tem&px=MTE3MzY

and xrandr1.4 came to fruition.

I'm pretty sure xrandr 1.4 never existed. I believe xorg decided not to release that (which kind of sucked at the time because we thought nvidia couldn't support xrandr in their blob without 1.4 which turned out to be false) and skipped to xrandr 1.5.

They said they don't have plans to support Optimus "at this time", which was in a post more than two years ago. And they didn't need to give any reason, it was clear to anyone having a clue - there was no business incentive, and the rest of the stack wasn't capable. I said this a long time ago already: http://www.nvnews.net/vbulletin/show...4&postcount=96 <- because I had a clue, rather than believing in magical middle fingers.

Well for the people who don't understand all the mechanics of a graphics stack, I suppose they could have elaborated. On the other hand, the elaboration could have evoked another reaction where people think they're just blaming open source instead of writing their own implementation which would make it somewhat of a catch 22.

Their custom implementation predates KMS. That's why they have it. And it's not the only such thing, Nvidia had a few things before the open stack got it - in addition to modesetting in the kernel, there's also accelerated indirect rendering (nvidia had it before AIGLX existed), and redirected direct rendering (nvidia had it before DRI2 existed), and possibly a few things more.

Other than the potentially false remark here about licensing issues, it would seem to make sense that they implement the kernel one and remove their custom one since it would reduce bloat in their driver from a programmer's perspective. Whether or not they would implement an fbcon driver at that point is still speculation. :-/

We're still very, very tiny compared to the amount of Windows users.

Somehow it doesn't seem all that tiny since linux runs on a lot of embedded devices (tablets, smartphones, etc). In terms of desktop users (laptops too), it also doesn't seem that small. I understand most computers sold come preinstalled with windows on it but who keeps the statistics of what people do with their computers after they buy them? If we're to believe that the sole statistic of determining the market share of linux desktops is based on the number of units sold where it's preinstalled, we may always remain a small market share (i.e. in the desktop space) even though that may not be the case.

[Gusar]

Perhaps it's just coincidence that nvidia started looking into optimus support after Linus' FU and the implementation of prime dma-buf happened to land in xorg 1.13 and kernel 3.5 right around that time.

There is no coincidence that Nvidia started looking into it as soon as dma-buf and xrandr1.4 came to fruition. That the FU happened around the same time, that's a separate event and believing it had anything to do with it is silly.

Things seem to indicate that they snapped to attention when Linus did that.

Please elaborate on these "things".

Sometimes I wonder why they said they simply wouldn't support optimus but never gave any particular reason

They said they don't have plans to support Optimus "at this time", which was in a post more than two years ago. And they didn't need to give any reason, it was clear to anyone having a clue - there was no business incentive, and the rest of the stack wasn't capable. I said this a long time ago already: http://www.nvnews.net/vbulletin/show...4&postcount=96 <- because I had a clue, rather than believing in magical middle fingers.

so that they wouldn't have to write a custom implementation (like their custom kernel modesetting implementation :-) ).

Their custom implementation predates KMS. That's why they have it. And it's not the only such thing, Nvidia had a few things before the open stack got it - in addition to modesetting in the kernel, there's also accelerated indirect rendering (nvidia had it before AIGLX existed), and redirected direct rendering (nvidia had it before DRI2 existed), and possibly a few things more.

Wait what? I'm pretty sure linux users pay money for nvidia hardware just the same as windows users. We may not be as many as windows users but we are a growing community.

We're still very, very tiny compared to the amount of Windows users.

[logistiker]

LOL!

Nvidia doesn't provide equal support for the Linux desktop because there's no business incentive to do so.

Wait what? I'm pretty sure linux users pay money for nvidia hardware just the same as windows users. We may not be as many as windows users but we are a growing community. I'm sure Valve recently throwing their weight behind linux also tends to help that argument as well.

[logistiker]

Err, Optimus falls into the low-priority desktop part.
(I know Optimus is on laptops, I use "desktop" as meaning "simple end user running Linux on their home machine", as opposed to high-paying customers doing heavy-duty graphics stuff on high-end workstations)

And *rolls eyes* that them looking into Optimus support now has anything to do with Linus' FU. Seriously guys, what is it with giving the FU such magical abilities? I really, really, really don't get it.

Perhaps it's just coincidence that nvidia started looking into optimus support after Linus' FU and the implementation of prime dma-buf happened to land in xorg 1.13 and kernel 3.5 right around that time. Things seem to indicate that they snapped to attention when Linus did that. Sometimes I wonder why they said they simply wouldn't support optimus but never gave any particular reason like it would be better if there was infrastructure for it like there is now in xorg and the kernel so that they wouldn't have to write a custom implementation (like their custom kernel modesetting implementation :-) ).

[Gusar]

In regards to the "stfu", it's exactly that response which doesn't get nvidia to provide equal same time support to their linux drivers as they do their windows drivers. This is why Linus' big FU was important.

LOL!

Nvidia doesn't provide equal support for the Linux desktop because there's no business incentive to do so. That's all there is to it, trying to come up with anything else is just silly. And Linus' FU changed exactly nothing regarding that. Meh, I suppose this myth of the legendary Linus FU will stay around forever, despite being total nonsense.

[logistiker]

Until you find a manufacturer that does better, I'd say stfu.

I could say intel (open source drivers, KMS and as of 2.20.7, hybrid graphics support) but obviously their GPU isn't nearly as capable as nvidia GPUs are. In regards to the "stfu", it's exactly that response which doesn't get nvidia to provide equal same time support to their linux drivers as they do their windows drivers. This is why Linus' big FU was important.

[Gusar]

Heh, and their lack of optimus support even now (albeit that seems to be changing in part to Linus' big FU) shows how high-priority even their X customers are.

Err, Optimus falls into the low-priority desktop part.
(I know Optimus is on laptops, I use "desktop" as meaning "simple end user running Linux on their home machine", as opposed to high-paying customers doing heavy-duty graphics stuff on high-end workstations)

And *rolls eyes* that them looking into Optimus support now has anything to do with Linus' FU. Seriously guys, what is it with giving the FU such magical abilities? I really, really, really don't get it.

[bug77]

Heh, and their lack of optimus support even now (albeit that seems to be changing in part to Linus' big FU) shows how high-priority even their X customers are.

They just announced they'll provide drivers for 6000 series till 2017. Until you find a manufacturer that does better, I'd say stfu.

[logistiker]

And to answer the obvious question of why don't they provide it then: Simple, they have no incentive to. The desktop is a low priority for them, and all their high-priority customers are using X.

Heh, and their lack of optimus support even now (albeit that seems to be changing in part to Linus' big FU) shows how high-priority even their X customers are.