Other Linux Things I Learned From AMD At GDC 2014

[bridgman]

But you avoided the real question: UVD is low priority due to scarce resources, Mantle is low priority due to scarce resources, changelogs are low priority due to scarce resources, so where do the developers take the time from for such major changes?

No, you keep changing the question

The people working on IP review for UVD support on older hardware are different (I mean really, totally, completely different) from the people working on the Catalyst driver. We can't move people back and forth between those two areas, at least not without 5-10 years of lead time.

Why do you think Mantle is low priority due to scarce resources ? The idea is to stop *changing* it before we go implement it everywhere.

Where do you get this "low priority due to scarce resources" thing from anyways ? Things are high or low priority independent of resources, it's just that the more resources you have the further down the priority heap you can work at any given time.

There *is* a real trade-off between change logs and other driver work since in the end both activities need the same skill set. Just to be clear, are you saying that working on change logs is more important than working on bug fixes, feature enhancements and performance improvements ? Maybe that is the case and we should divert resources, I don't know -- but my impression was that improving the driver was still felt to be more important by our users than improving the change logs.

[johnc]

When I read "Shanghai" the first thing that popped into my mind was "minimum wage". Hopefully I'm wrong on that and they've actually got really solid software engineers in their Linux organization.

[johnc]

There *is* a real trade-off between change logs and other driver work since in the end both activities need the same skill set. Just to be clear, are you saying that working on change logs is more important than working on bug fixes, feature enhancements and performance improvements ? Maybe that is the case and we should divert resources, I don't know -- but my impression was that improving the driver was still felt to be more important by our users than improving the change logs.

I do see your point here, but here is the reality from the consumer perspective: your primary competitor is providing a superior driver while simultaneously releasing change logs.

It's just business.

I think AMD is erring here though by not sufficiently fertilizing their Linux efforts, especially with Steam Machines coming and the revolution of mobile... AMD risks being left behind. JMO. Nobody pays me a few million to be a CEO though, so what do I know.

[bridgman]

I do see your point here, but here is the reality from the consumer perspective: your primary competitor is providing a superior driver while simultaneously releasing change logs.

It's just business.

I think AMD is erring here though by not sufficiently fertilizing their Linux efforts, especially with Steam Machines coming and the revolution of mobile... AMD risks being left behind. JMO. Nobody pays me a few million to be a CEO though, so what do I know.

Fair point. We have a bunch of developers working on the open source drivers and documentation -- NVidia does not, but they do have better changelogs.

If we killed the open source driver effort we could probably have really nice changelogs -- where do you think the $$$ are best spent ?

[johnc]

Fair point. We have a bunch of developers working on the open source drivers and documentation -- NVidia does not, but they do have better changelogs.

If we killed the open source driver effort we could probably have really nice changelogs -- where do you think the $$$ are best spent ?

It's a judgment call of course. My judgment is that a majority of potential consumers -- gamers, mobile users, potential Steam Machine buyers -- would prefer the best driver possible above the driver being open source. And I think the current overall market choices bear this out (Windows, Apple, Android, etc.).

I know that's not the best choice for fans of FOSS, but that's how I would be thinking if I was AMD. It's pretty clear that the #1 PC game distribution service is pushing Linux and OpenGL hard and has every interest in effecting a major upheaval in the PC game space. It's not obvious that FOSS provides a serious advantage in the minds of the average consumer.

[Vim_User]

No, you keep changing the question

I don't think so.

The people working on IP review for UVD support on older hardware are different (I mean really, totally, completely different) from the people working on the Catalyst driver. We can't move people back and forth between those two areas, at least not without 5-10 years of lead time.

OK you are right, UVD does not belong in this equation, sorry. But nonetheless Mantle and changelogs do.

Why do you think Mantle is low priority due to scarce resources ? The idea is to stop *changing* it before we go implement it everywhere.

From AMD employees, posted by Michael on Phoronix in this article:

The common response I received came down to that they're currently looking at the feasibility of Mantle on Linux and that it comes down to a resource issue in committing to any Mantle Linux support plans.

That clearly sounds like: We can't do it now, resources are scarce.

Where do you get this "low priority due to scarce resources" thing from anyways ? Things are high or low priority independent of resources, it's just that the more resources you have the further down the priority heap you can work at any given time.

That is implied by saying things like quoted above about Mantle or something like this:

The common theme of the response was "the cost of the engineering time" and then posing the question whether Linux users would like to see more bugs and features worked on rather than proper release notes.

Which comes down to: We do not have enough resources to do both. It isn't a matter or high and low priorities if you do not have enough resources to ever reach to the bottom of the heap.

There *is* a real trade-off between change logs and other driver work since in the end both activities need the same skill set. Just to be clear, are you saying that working on change logs is more important than working on bug fixes, feature enhancements and performance improvements ? Maybe that is the case and we should divert resources, I don't know -- but my impression was that improving the driver was still felt to be more important by our users than improving the change logs.

I am not saying that providing changelogs is more important than fixing bugs, but it is possibly more important than your development team thinks. Just have a look at any of the threads about a new Catalyst release, you will find many posts like, for example, this:

Anyone know if there's still that nasty input lag in Source games on Trinity? I'm not much tempted to try out this beta. I'm thinking some driver by the end of the year should make 2 year old hardware (by then) usable. Sad.

Posts like that really come up in every thread about a new driver release, people are annoyed by the fact that they have to actually install the driver, test the game, if the issue is not resolved uninstall the driver, all that happened in the past often enough in conjunction with a kernel downgrade due to Catalyst not supporting the latest stable kernel. If there would be a clear changelog that states what has changed for certain software that would give AMD a much better image.
Comes down to: If your customers are annoyed by you you are doing something wrong. Making funny statements about dropping FOSS support for better changelogs does not improve the situation, by the way.

[blackout23]

Fair point. We have a bunch of developers working on the open source drivers and documentation -- NVidia does not, but they do have better changelogs.

If we killed the open source driver effort we could probably have really nice changelogs -- where do you think the $$$ are best spent ?

"Better" changelogs is almost an understatement when you compare them to the catalyst changelogs.

http://www.nvidia.it/download/driverResults.aspx/73700/it

http://www.nvidia.de/download/driverResults.aspx/73693/de

They translate these long changelogs even into other languages and it's not Google translated. Grammar and punctuation is flawless. A real person wrote that for the Linux driver.

[bridgman]

I am not saying that providing changelogs is more important than fixing bugs, but it is possibly more important than your development team thinks. Just have a look at any of the threads about a new Catalyst release, you will find many posts like, for example, this:Posts like that really come up in every thread about a new driver release, people are annoyed by the fact that they have to actually install the driver, test the game, if the issue is not resolved uninstall the driver, all that happened in the past often enough in conjunction with a kernel downgrade due to Catalyst not supporting the latest stable kernel. If there would be a clear changelog that states what has changed for certain software that would give AMD a much better image.

OK, now we're getting somewhere.

Most of the "fixes" that people are testing for in the manner you describe do not come from a Linux driver developer fixing a Linux driver bug reported against Linux application X, but rather from a developer fixing a bug in shared code where the original bug was probably reported against a Windows application. The fix also *happens* to fix Linux application X as a side effect, but you can't get that from changelogs and AFAICS I don't see that in the NVidia changelogs that were linked either.

If we implement the plan you are arguing against, most of the code you care about would be managed in open source repositories with per-commit visibility, ie detailed changelogs would come "for free". I say "for free" because there is an added inspection/review cost to maintaining publicly visible code, but we're already funding most of that cost for the open source driver today.

[chrisb]

It's a judgment call of course. My judgment is that a majority of potential consumers -- gamers, mobile users, potential Steam Machine buyers -- would prefer the best driver possible above the driver being open source.

There are many governments, corporations, educational establishments etc. that are also potential customers, and who would prefer open source. For some this is practical - they might use, say, Red Hat, who won't support them if they taint their kernel with closed source drivers. For others, they may not trust the security implications of closed source kernel drivers. They may be building a 1000x PC cluster, and not trust that the driver will be maintained in the future, or be compatible with future releases of their chosen distribution. And for many, they may have sysadmins who don't want to play the game of trying to match kernel versions from their Linux distributions with the closed source driver, or deal with the problem of not being able to install a kernel update that may be incompatible with fglrx (perhaps this isn't such a big problem anymore, but it once was a big issue that distributions would release kernel updates, perhaps security updates, people would upgrade, and then find that they don't have accelerated graphics anymore).

Remember China: NVIDIA lost an order for 10 million GPUs?

The order was at least for ten million GPUs, which given the current low-end parts, would value the order at least 250 to 350 million dollars (USD). However, I've heard from a separate source that it was closer to the half billion dollar mark. This money will now be handed over to AMD since they have the officially-based open-source driver for their products.

[johnc]

Remember China: NVIDIA lost an order for 10 million GPUs?

I do recall Michael's shoddy reporting on that, yes. I do recall he was also the only source for that story. And I don't recall seeing any corresponding bump in AMD revenues in subsequent quarters.

Yeah it's possible it was a true story but we're talking super low-margin parts. And from the likes of it that was probably an IP grab anyhow.