Open-Source NVIDIA Outlook Brighter Due To GSP Firmware, But Major Challenges Remain

[qarium]

You should stop moving the goalposts

what nvidia wants to perform here is not yet proofed in practice...
it could work in theory but as soon as some security holes show up and this need change in the firmware api this will result in chaos ...
and could result in strong firmware version numbers and also mesa code version numbers means different driver for every different firmware version api... or people stay on old firmware because new firmware with security fixes break someting...

people right now believe this is "nice" only because nvidia did make the kernel driver opensource but it could result in a desaster ,.,

people who want opensource driver should not buy nvidia ... Nvidias way to do this is not yet proofed to be practical.

[arQon]

meanwhile polaris doesn't have hevc amf support

Apparently the Windows driver doesn't either, which implies that the HW isn't actually capable of it for some reason.

For most of the rest of your points, while they're fair ones to make you're basically asking RTG to spend money that they don't / didn't have, on an EOL architecture, for the sake of a tiny niche of what's already less than 1% of their market. You certainly have a right to be irritated by their decision not to do that, but I don't think you could really consider it a surprise.

[arQon]

Nvidia's been working kinda slow on getting their driver into the kernel (have they even been working on it at all) I wonder if nouveau won't just beat them to it, still, even if it does, it's a moot point unless the nouveau performance is on par with the proprietary drivers.

There is realistically no chance of *either* of those things ever happening, let alone both.

> they aren't bound by the limitations of not being allowed to let the linux driver surpass the windows driver, stupid feature parity bs (wasn't even feature parity, was just "windows performance must be equal or better than linux performance")

I have no idea where that train of thought came from, but that's not how things work. Nobody at nvidia GAF about the platform in the way you're suggesting, and while MS would certainly be happy to enforce that themselves, they can't. If anything, from nvidia's perspective the profit from HPC / DC / CUDA is a compelling reason for the *opposite* to be true: the gamer market isn't that important to them any more, it just has a lot of momentum from the old days still.

[Quackdoc]

Apparently the Windows driver doesn't either, which implies that the HW isn't actually capable of it for some reason.

For most of the rest of your points, while they're fair ones to make you're basically asking RTG to spend money that they don't / didn't have, on an EOL architecture, for the sake of a tiny niche of what's already less than 1% of their market. You certainly have a right to be irritated by their decision not to do that, but I don't think you could really consider it a surprise.

what? I use hevc AMF on my rx 580 on windows daily... how does the windows driver not support it? and EOL or not, EOLing a card released not that long ago is scummy behavior

[arQon]

what? I use hevc AMF on my rx 580 on windows daily...

Then it's been fixed in the last 2 years at most, because only h264_amf worked prior to that.

> EOL or not, EOLing a card released not that long ago is scummy behavior

No argument there. That's not quite what I said though, which was that *Polaris* was EOL: that is, that it was an obsoleted architecture that development was abandoned on because it wouldn't be used in any future products, and there was a whole new bunch of hardware that needed that manpower instead.

I was being generous when I described your use case as a tiny subset of what was already a tiny platform - because not only was it both of those things, it was also on legacy HW, and legacy HW doesn't get you YT videos or tech site articles like your newest shiny stuff does. That new shiny stuff also needs working drivers, and a missing feature there might result in lost high-margin sales.

Like I said:
> You certainly have a right to be irritated by their decision not to do that, but I don't think you could really consider it a surprise.

[fitzie]

The naivety/cluelessness of some people is outstanding. I can almost guarantee you that NVidia's decision to opensource a tiny part of their graphics stack (even calling it open sourcing is a stretch, they basically put some logic into closed firmware on the onboard graphics chip, i.e. GSP and the open source part is an interface + glue code to that firmware) had nothing to do with recent sales, due to the fact that such decision are preplanned years before.

NVidia did this because Linux is moving to Wayland and since their EGLStream effort didn't get community buy-in, they were forced to do this in order to properly support GBM/KMS etc etc for Wayland. There were other reasons as well, i.e. their current distribution method doesn't work well with some distros.

saying that nvidia didn't do this because of sales is pretty naive. I was talking about sales in the enterprise and car applications, you know the sectors where all nvidia growth is occuring? and where the customers actually get to demand things of vendors?