Nvidia, free software, and Linux

swick · March 29, 2015, 8:43pm

New thread because it has pretty much nothing to do with the original thread.

Continuing the discussion from Poor Gaming Performance in Linux:

Dear Swick,

at first I'd like to apologise for editing your post slightly so that readers of my post can easier understand the points I'm replying to. Sorry!

And now into the business: I'd say that this is now heading towards the more philosophical direction about what's good and right. Everybody has their own opinions about that. And it's OK as long as people don't go running around claiming that their way is the only way it should be.

(As a funny side-note, we've now reached the same central argument between GNU/Linux users that lead @MEC_777 to erroneously believe that Linux drivers for AMD's hardware were superior. Like he described it, there's this decade old ideological split in the community about this matter.)

From my perspective you seem to be essentially saying that a company should not provide the very best it has to offer. Should not provide anything but the standard features. Should not distinct itself from its competitors. And should opt to a poor committee designed standards that are not as good as they could have been. And should not try to compete. And so on.

That's in opposite direction of what free market should be about. There's no API that "should not exist" in my opinion. I tend to think that there are various levels of "vendor lock-in". One some levels it is acceptable and perfectly fine thing. However I want to make one thing clear now: The word itself, "vendor lock-in", contains a certain level of ideological package with it which I am not entirely comfortable using it in this case.

Certain people like you use ”vendor lock-in” very loosely to attack companies. I don't think that just providing unique solution to a problem or a need that clearly exist should constitute to "vendor lock-in" just because only one company is offering that feature or set of features that no other company has access to. (The ”Free software” also carries an ideological package with it. That's why certain people thought that ”Open Source” is a better choice of words.)

If that solution proves to be very good and popular (as in competitive solutions appear that duplicate the functionality of the original solution) it makes sense to standardise it and enable it for everybody. Especially after the company that innovated the solution has made back their investments into the research and made a reasonable profit from it.

If the original company tries to prevent competitors from providing similar feature to compete with their feature, or prevent their customers from moving to use the competitive feature, or prevent the standardisation of less used solutions, then that is what vendor lock-in is all about. That's against the free market.

Additional side-note: I'd keep software non-patentable because I see software as pure mathematics and it's very difficult for me to see why pure software processes should be patentable when they're nothing but abstract ideas without an actual physical implementation. Software can be closed but not patentable.

And one more additional point: The way you use ”vendor lock-in” in this case discredits the actual dangerous vendor lock-in situations, in my honest opinion. There are companies out there which are constantly seeking ways to fix the market and try to prevent their competitors from providing solutions like e.g. Oracle, Novell, IBM, Microsoft, Apple, and Google to name a few.

These companies have grown so big on certain markets that sometimes they try to actively make it so that it's as difficult as possible (e.g. cost is too high) for their customers to move from their product to a competitive product. The classic example of that is the Microsoft and their text processing program Microsoft Word. Another classic example is IBM and their punch cards.

nVidia is nowhere near the top on the list of the worst offenders. nVidia tends to open new markets and compete within that market with their own rules and they do allow others to participate (much like Microsoft in their early days). nVidia is ”open”. I'll come back to this point later when I talk about VDPAU.

I am pro choice. If it's your choice to not support nVidia then it's OK. I am actually grateful to you for your dedication to AMD and Intel because by supporting their ”free” and ”open” solutions it means that nVidia will have to compete honestly. We are benefiting from that kind of competition.

Competition forces both companies to try to do their best, and it ensures that customers don't have to ”suck it up”. That would happen if and when one company has too much leverage over the market like it was in 2005-2007 when nVidia arguably had a monopoly like position in Linux 3D market (and from certain perspectives they still hold on to that kind of position).

I support free software because opening the source code for public scrutiny gives certain freedoms to its users as you are aware. It improves technical side of the software especially because it enables more competition on that market where it is being applied to. Closed source is not an entirely bad thing (but it opens the door to monopolies and trusts which are a inherently a bad thing. That's why proper competition and regulations are needed.)

If nVidia, or any company for that matter, gains a complete monopoly in computer graphics market it would be a really bad thing. It doesn't if it is nVidia. AMD would be just as bad as a company in that kind of position. Any individual or organisation who has power wants to stick to that power as much as they can and haul even more power to themselves. That leads to the failure of the system if and when there's nothing to balance things out between the parties.

I see need to support proprietary software so that utilising as many free software or open source components as possible becomes a functional reality. Without nVidia's GNU/Linux support, I'd be forced to use Microsoft Windows because Intel and AMD just do not deliver in my opinion a product that serves my needs. But there's also the aspect of competition. When free software and proprietary software compete against each other both ideologies are forced to improve themselves and their intellectual products on the market.

I see no problem in compromising little bit of the feature set I get from nVidia in-order to support AMD because I see value in their current attitude towards the more free and open driver development. However AMD doesn't deliver a product that serves my specific needs. They might do so in the future. At which point I'd be more than happy to try them again. But at the moment I'm still quite a bit disappointed in abilities of their product lines.

I could tell you a story why I think the ATi as a company was the worst thing on the planet and how they cheated GNU/Linux users and betrayed trust of many. But let's move on. You see, I could talk about this topic for hours. But let me move on and talk little bit about the VDPAU. You see, I'm not entirely sure if you've understood what VDPAU (or OpenCL) actually is.

VDPAU was developed by nVidia (and still is for the most parts). VDPAU is just one of the examples that shows how nVidia sees free software and open source market as a opportunity for itself. They saw that there was a need on GNU/Linux and UNIX-like systems for a hardware accelerated video processing.

nVidia then made VDPAU to implement their PureVideo on UNIX family of systems and UNIX-like systems (Linux, Solaris, various BSDs, etc.). And then gave it away to FreeDesktop.org as a royalty-free API for adoption by free software community. The only similar things that exists to my knowledge on Linux are Intel's VA-API (which is very much Intel specific since Intel doesn't support VDPAU) and AMD's XvBA (which is AMD's answer to VDPAU which only works on GNU/Linux and only on AMD's graphics cards unlike nVidia's VDPAU).

VDPAU interface is used access certain video processing capabilities of the graphics card. It is implemented in the graphics drivers. VDPAU itself only defines a small standard set of functionality but it can offer more than that if the manufacturer extends it which is entirely possible. Both nVidia and AMD have their own extras in their respective cards you can access through VDPAU.

VDPAU is a very good at what it does to the point that AMD also implemented it in their drivers. Now nVidia and AMD can both compete in the accelerated video for Linux market. VDPAU is probably the most common way to access nVidia's PureVideo and AMD's UVD capabilities. Software developers can safely implement VDPAU in their software and everybody wins.

Note that AMD's UVD's feature set is different compared to nVidia's PureVideo capabilities. In my opinion PureVideo is better because of its performance and extra capabilities although for common usage it probably doesn't make much different which one you have. You see, I don't have extensive experiences from AMD's UVD.

Both OpenCL and OpenGL work pretty much the same in principle. Both APIs provides a standard API/Interface to a card that implements certain standard features in drivers and also certain features that are specific only to the manufacturer. This way the companies can distinct themselves by offering additional capabilities on top of the standard features.

Extending has its problems too which easily could lead us to the talk about why DirectX is so popular among the game developers. However I am not going there. I am also not denying the raison d'être of these extensions. Some of them could and will become official part of the standard someday and everybody will benefit.

AMD doesn't have as much money to spend on software development as nVidia does. That shows in the amount of features they provide and in the stability of their driver.

en.wikipedia.org

VDPAU

VDPAU (Video Decode and Presentation API for Unix) is a royalty-free API as well as its implementation as free and open-source library (libvdpau) distributed under the MIT License. The VDPAU interface is to be implemented by device drivers, such as Nvidia GeForce driver, nouveau, amdgpu, to offer end-user software, such as VLC media player or GStreamer, a standardized access to available video decompression acceleration hardware in the form of ASIC blocks on GPU chips, such as PureVideo or Unifie...

Next I'll go through your points...

I've addressed the fact that making money on a new market is not a bad thing. And it has not been ”the worst route” to do so with a proprietary drivers or signed firmwares. nVidia was very open about it and alerted Noveau developers and talked with them about it. Noveau drivers work properly with any nVidia card, even the Falcon ones which you're talking about. I'd suggest you read:

http://www.phoronix.com/scan.php?page=news_item&px=MTc5ODA

Besides AMD already has similar firmware solution to protect their chips. With nVidia now issuing similar signed firmware images to Noveau project isn't too different from what AMD is already doing.

I'd like to see the source for ”nVidia pressuring the kernel developers to re-license their code”. I don't remember that event and I'd like to read more about it before giving away my thoughts on the matter. However if you're talking about the DMA buffer thing where nVidia wanted to enable Optimus for the Linux users. But to do that they would have had to use DMA-BUF to achieve that so the Intel and AMD stuff would have worked together. But that DMA-BUF only allows GPL'd drivers and nVidia asked to change it so that it would allow also their proprietary driver. The ”stealing code” argument is pure propaganda and fearmongering by FSF people in relation of this topic. Anyway that didn't happen and that's why we don't have nVidia's Optimus on Linux officially supported by nVidia's proprietary driver. This is the reason for it.

http://www.tomshardware.com/news/nvidia-gpl-linux-driver,18572.html
http://www.phoronix.com/scan.php?page=news_item&px=MTA1OTU
http://www.phoronix.com/scan.php?page=news_item&px=MTIwNDI

I'd also like to see your sources for nVidia holding back their OpenCL implementation in hopes of people turning to CUDA. I don't remember that event. Not at least in the way you described it since you don't have to use CUDA API to get good GPGPU support. That's not true.

Yes it's true that OpenCL on nVidia cards is implemented using CUDA platform interface. But from the point of the source code or performance it doesn't make any difference. The same OpenCL code runs both in Intel and AMD implementation of OpenCL. Besides nVidia's OpenCL performance is on-pair or faster than with the Intel's and AMD's relative performance depending on what kind of GPU is being used.

CUDA as a platform is overall better designed, its documentation is better, and the libraries are better in my honest opinion. However I think it's fantastic that we have these two (very different) solutions for the sake of competition. Not to mention that CUDA as a platform doesn't only support OpenCL. OpenGL CS, Microsoft's DirectCompute and C++AMP can also be used through CUDA.

Now that I remember you should also know that the nVidia drivers also have support on Linux for the nVidia GameWorks. It comes with support for VisualFX, PhysX, and OptiX. OptiX is especially interesting because while ray tracing might not be a thing right now in games but in future it will be. OptiX is useful in various science and engineering applications at moment. The point is that nVidia driver don't offer just ”performance”.

nVidia's driver does offer performance but also comes with loads of extras for the kind of engineering stuff I do that I find useful. AMD doesn't deliver this kind of stuff. There are similar things that offer something like that but nothing that's quite as good.

Unlike with the AMD's proprietary driver, the nVidia's proprietary Linux, FreeBSD, Solaris, and Windows drivers use the same common core. When Windows driver is improved those same improvements also go to the Linux driver almost directly. And yes when Windows driver gets crippled then Linux, FreeBSD, and Solaris drivers are also crippled in most cases. But that's the magic behind the nVidia's driver superiority on Linux.

Linux, FreeBSD, and Solaris will receive the support for new nVidia graphics cards the same day Windows receives the support. When OpenGL version is updates Linux, FreeBSD, Solaris and Windows receive the support at the same time. And then there's the performance parity between Windows and Linux version.

And that's why you're also little mistaken. Nvidia's proprietary driver doesn't limit the number of monitors per single card. As long as you're using single card you can have as many monitors as the card supports. If you have, let's say GTX 750 with 2x DVI and 2x HDMI, then you can use all those four ports with four monitors.

However the driver does indeed limit the number of total monitors in a multi-GPU set-up to three instead of the previous four in consumer GeForce cards. This limitation doesn't exist on professional Quadro cards.

This change apparently happened because nVidia changed their driver core to accommodate the rule changes on Microsoft Windows regarding the multi-GPU multi-monitor set-ups. The normal GeForce cards have the Base Mosaic feature and the so called SLI Mosaic feature is still there in the professional Quadro cards and it works as intended.

You can call it ”an artificial limit” and ”pissing into the pool of existing customers” because that's what it is. What else can I say? It's a result of one company having too much leverage on the market that is not most significant for them.

https://bbs.archlinux.org/viewtopic.php?id=193314
https://devtalk.nvidia.com/default/topic/809498/2x-gtx970-sli-mosaic-4-monitors/
https://devtalk.nvidia.com/default/topic/579449/linux/basemosaic-v295-vs-v310-vs-v325-only-up-to-three-screens-/1/
http://us.download.nvidia.com/XFree86/Linux-x86_64/325.15/README/xconfigoptions.html

But in all honesty no matter what graphics manufacturer you support that multi-GPU set-ups in Linux are kind of messy and you're unlikely to get all the benefits from having one. It probably results performance setbacks actually if I've understood correctly. And it doesn't matter what card you're using. AMD doesn't even support this on Linux to my knowledge. If anything I think this gives AMD a spot they could serve segment of the market who do require cheap multi-GPU set-ups with more than three monitors. But they're not investing into it.

We could also debate whenever or not FreeSync would even exist if nVidia didn't come up with G-Sync or Adaptive V-Sync first. FreeSync was AMD's response to G-Sync. And nVidia has the right to charge any arbitrary sum of money they want for their technology.

http://www.geforce.com/hardware/technology/adaptive-vsync

Why would you care if they charge extra 100 USD for it? If it was the only solution, and you felt that you need it, then you'd be likely to buy it. It's simple business. And going back to the theme of this post... nVidia saw a problem and offered a solution.

FreeSync is exactly what AMD should have come up with in-order to offer a competitive solution to the market. FreeSync is the very opposite of G-Sync which is proprietary and exclusionary. AMD's FreeSync was adopted as a VESA standard and is royalty-free, free to use, and at least in theory, has no performance penalties.

And about the reports that nVidia won't be supporting it... Well, yeah... nVidia has to support it eventually because Adaptive Sync aka. FreeSync becomes mandatory with the DisplayPort 1.3. At the moment with 1.2a it's optional.

Please go on. I can take time on weekends to correct the misinformation and misunderstandings you might still have. The point is that Intel, AMD, and nVidia are all companies in the business of making money. AMD has found out that by supporting free and open source software they can make money even if they aren't offering the best performance or the best feature set. And that's good! They're able to serve audience they didn't have before. nVidia on the other hand has found out that the best way to make money is to be the best in the performance, offering the widest range of features, and supporting the professionals out there and their special needs.

One of my personal dreams is to push free software to schools so that children would be able to see that there are more choices than just Microsoft Windows or Apple's OS X and iOS. The free software and the world would benefit greatly from that. I agree with the RMS that closed proprietary software in schools is harmful to society in terms of monetary value much like cigarettes are (for different reasons obviously).

But I also defend the closed source where it makes sense. Closed source wasn't that bad of a thing before the internet came and changed everything. Now it's entirely possible for Amazon, Sony, Microsoft, Apple, etc. to spy on you and disregard your civil and human rights. It's entirely possible for Sony to remove features from the products they sold. It's possible for Amazon to remove eBook you bought like they did with the 1984 Kindle edition.

What I seek is the golden middle way and the ability to have different choices. Free software is not the ultimate answer to all the problems but it is a great way to provide some choice for the people and balance the negative effects of the proprietary software.

Yours truly,

Hopeakettu

P.S. I quite literally just wrote this in an hour or something so if you find mistakes then please report back so I can fix them. That includes spelling and grammar mistakes. You see, I am not a native English speaker. English is the third language I've learned. Some of the things I've written here will probably appear bit "strange" to you guys. I'll keep drinking beer in a meanwhile. Cheers and have a nice Sunday!

Hey, I appreciate your response but there are quite a few statements I disagree with. First something more general:
I find it weird that you defend Nvidia! I would also be confused if you defended any other for-profit company. You only have to justify your buying decisions for yourself, it doesn't matter if I believe that your decision is wrong or stupid. Nvidia does things which are bad for you, the user and you should not defend such behavior, no matter which company, even it it makes sense from the perspective of the company.
It's a thing I notice way too much where users try to understand a decision of a company which is bad for the user. Then they say that the company had to do something this way because it makes sense for the company from a monetary viewpoint, thus justifying the bad decision. It's a stupid thing to do.
You're the user, whatever is bad for you is simply bad for you, you should not care why the company decided to make the decision!

Now I'll go over your list of arguments and try to explain how each decision Nvidia made is bad for the user (if it makes sense for Nvidia to do or not is completely irrelevant).

being open about a decision that's bad for users is not good in any way
the decision to only allow signed firmware means that the whole work of reverse engineering and writing a free software firmware is wasted
you can only run firmware on your GPU which is signed by Nvidia, effectively taking away freedom to use the hardware you bought at your liking
Nvidia can now turn off hardware features in the firmware and you can't turn them back on

AMD GPUs allow you to run arbitrary firmware. You're just wrong on this one (find me a source if you still think you're right).

Yes, that's exactly what I was talking about. You call it asking, I called it pressuring. It's probably very subjective and down to your own interpretation.
Anyway, this is a great example, showing how not playing by the rules of the kernel (i.e. drivers should be free software) results in a bad experience for the user. In this case: GPU offloading doesn't work with Nvidia hardware. (Remember, we don't care why Nvidia doesn't support offloading, we don't care that it's their license choice, we only care that it's bad for us, the users.)

Well, we're at OpenCL 2.0 now and Nvidia still only supports OpenCL 1.1 (iirc). So if you want to develop something that runs on Nvidia hardware you can use OpenCL 1.1 which is really bad to develop for, or use CUDA which Nvidia invests a huge amount of effort.
That's bad for a developer (who is also a user of the GPU) because he can't choose between OpenCL 2.0 and CUDA but essentially has to choice done by Nvidia, and that is using CUDA.
It's also bad for a user of applications because the developer now has to spend double the effort (CUDA + OpenCL) to make the application perform good on all vendors GPUs.

OpenCL 2.0 and HSA are very powerful and really good from a developers perspective. Which one is better however is not relevant. All the cool CUDA features could have gone into OpenCL extensions or at least be opened up for other vendors. Neither of those happened which led to the problems I described earlier.

And again, it's not good for a user, no matter how "cool" or "interesting" the so called "features" are: Nvidia is taking away choice from the user! All those technologies are available on most other GPUs the only thing that happens is that Nvidia gives developers a fast implementation for their GPU and a slow implementation for every other GPU. That would not be a problem if the code was open source and other vendors could adapt the code to run better on their own GPU, but it's not.
For a user it simply means that Nvidia performs better, which also means that it makes no sense to buy a similar fast GPU from another vendor. In short: it takes away choice.

Here you're wrong in a few ways:
1. AMD also shares a lot of the code in Catalyst between Windows and Linux
2. I'm pretty sure that the drivers don't use the same "core" but instead use a separate core for each OS with the same interface and share all of the "non-core" code

Anyway, the interesting part here is that you justify Nvidia's decision to remove a feature. Again, as a user I don't care how or why, the only part that matters is that Nvidia removed that feature!

That's an artificial limit! The hardware clearly does support it so they take away features from you.

I don't care why it happened! It happened! Stop justifying bad decisions!

[...]

Again with the "I don't care!". Now there is an open standard and Nvidia doesn't support it. There isn't more to it.

I think you're wrong on this. To my knowledge Adaptive Sync is not mandatory in DP1.3, but that's beside the point: Nvidia can still choose not to support Adaptive Sync simply by not supporting it in their driver.

I have yet to see a single instance where proprietary software makes sense. True, sometimes a piece of proprietary software might be better in some way than free software, but it doesn't change the fact that the piece of proprietary software would make more sense if it were free software.

swick · March 29, 2015, 8:44pm

Replying to myself because apparently I just went over the max-size of a post.

About a few things I skipped at the beginning of your post:

I disagree with your interpretation. I'm all for competition but competition should be useful for the user. I don't think that proprietary APIs are in any way useful for a user because it means that when a user wants to use a program that is designed to run on the proprietary API, the user has no choice of the hardware.
We can see this every now and then when people choose their GPU by vendor because the program they want to use only supports CUDA or CPU computing.
That's bad, and that's why a new feature or API is only useful for the user if it is free software or a standard. Competition has to be in this frame to be useful.

Why do you care as a user? You should not! You care about choice and good value for your money and that's why you should say that every API should be open.

When two free software implementations compete against each other I'm more happy as a proprietary software and a free software implementation compete. Saying "I'm for proprietary software because it forces competition" is crazy because the same competition exists with free software.
Also, I'm not saying that Nvidia supporting Linux is a bad thing, all I'm saying is that Nvidia generally makes bad decisions for the users.
So, yes, VDPAU and Linux support are good things but I'm still going to call them out on all the other shit they're doing, and for me personally, I'm not willing to support the company in any way. It's a decision everyone has to make for himself but I'll gladly help to explain my reasoning.

Hopeakettu · March 29, 2015, 11:19pm

Hi, I wanted to let you know that I have read this. And I don't know when I am able to make "a full reply". But in-general I do agree with some of your points. And disagree with others. I too think that companies should be called on their shit (although it's apparent that we draw the line in different height).

But something stuck me odd... You wrote that:

However nVidia already has (at least partial) OpenCL 2.0 support in their Linux driver. When you do:

$ strings /usr/lib/x86_64-linux-gnu/nvidia/current/libnvidia-opencl.so.340.65 | grep CL_COMMAND_SVM_

You get:

CL_COMMAND_SVM_MAP
CL_COMMAND_SVM_UNMAP
CL_COMMAND_SVM_MEMCPY
CL_COMMAND_SVM_MEMFILL
CL_COMMAND_SVM_FREE

And these are macros that are only part of the OpenCL 2.0 as kindly noted by AMD's and Khronos Group's documentation:

However I see no mention of the OpenCL 2.0 in the nVidia's developer documentation. And that is little fishy. I guess people at nVidia are still working on it. After all AMD got their OpenCL support out only few months ago in September, 2014. And Intel got their support rolled out in August, 2014 which was pretty quick. I guess the nVidia's OpenCL implementation makes it either A. complex to include new stuff, or B. they have something going on there.

However we have to remember that the guy leading Khronos Group at the moment, Neil Trevett, is an nVidia employee. Not to mention that nVidia was in a big role in designing the OpenCL 2.0, and nVidia would only lose in the long run if they would not implement OpenCL 2.0. I'd expect nVidia to have OpenCL 2.0 out sometime this year.

I'll try to find time to reply to you in a more complete manner. You're pretty interesting guy.

Commissar · March 29, 2015, 11:25pm

how did I get here....AND

dez · March 29, 2015, 11:29pm

Xaom · March 29, 2015, 11:29pm

anon93339890 · March 29, 2015, 11:32pm

I think you are lying about those big butts

Commissar · March 29, 2015, 11:33pm

Novasty · March 29, 2015, 11:33pm

So.......... Tits or Ass?

anon93339890 · March 29, 2015, 11:33pm

I like fore arms

Commissar · March 29, 2015, 11:34pm

B_Type13X2 · March 29, 2015, 11:34pm

I protest your implication that I do not indeed like or enjoy big butts. We will have a dual at dawn, good sir, first person to die loses.

Commissar · March 29, 2015, 11:35pm

Xaom · March 29, 2015, 11:35pm

anon93339890 · March 29, 2015, 11:36pm

I think you misunderstand I am in capable of dying since I am a Wizard, let me show you my selfe:

Commissar · March 29, 2015, 11:37pm

B_Type13X2 · March 29, 2015, 11:39pm

That is a fair point, I guess I will, just have to dismember you, and put your torso and head into 2 different safes and drop each safe in 2 different oceans and in the deepest parts of those oceans.

Commissar · March 29, 2015, 11:43pm

Xaom · March 29, 2015, 11:52pm

NO ITS NOT!!!
FLKDS:JFKLSD:FJKSLGH:W#RUPQWBU$#

Kat · March 30, 2015, 12:12am