Slashdot Mirror
NVIDIA's New GPUs Are Very Open-Source Unfriendly
An anonymous reader writes: The Nouveau driver developers working on open-source support for the GeForce 900 Maxwell graphics cards have found this new generation to be "very open-source unfriendly" and restricting. NVIDIA began requiring signed firmware images, which they have yet to provide to Nouveau developers, contrary to their earlier statements. The open-source developers have also found their firmware signing to go beyond just simple security precautions. For now the open-source NVIDIA driver can only enable displays with the GTX 900 series without any hardware acceleration.
IIRC, This has always been the case.
With Valve pushing for Linux gaming, they need to apply some inside pressure on AMD/Nvidia to make their shit work at 100% with Linux.
Since we know neither company is willing to do the work themselves, that means they need to release full documentation so the FOSS people can develop/maintain proper Linux support.
Why should Nvidia care about open source? I know this is slashdot, but I think it is past time we start to question the underlying assumptions behind making everything open source. Who does it benefit? Very few. Who does it hurt? Many. How many hardware and software companies have been destroyed by this "open source everything" agenda? Meanwhile, developers are constantly asked to give up the rights to their OWN work while companies like google and apple cash in on it with BILLIONS of dollars of income. If open source is so good, where is OUR share of the rewards for all this work?
How many of those linux machines that were required to post this comment also requires a high end GPU. I would venture to guess close to zero. Why sould a GPU manufacturer spend a lot of time supporting such a small user base?
That might have been a good troll in 2005, but it's at best a 2/10 in 2015.
wat
WTF is this post supposed to mean?
You are talking complete bullshit. .Net core as open source to add portability, and it will release windows for free (!) for the raspberry PI. The rPI move is to make people who would otherwise use open source based OSs, to install windows. Often the rPI is the first device people are confronted with linux. They are giving it a try there, as a safe test-bed. Of course, they are more comfortable with windows, and would install that if they could. Therefore, windows for rPI will be very successful. .Net move was to strengthen the .Net platform, make people depend on it, and then make it closed-source again in future versions. People will either have to abandon their code, or switch to windows. They will switch to windows.
Only in recent months, Microsoft is moving towards users. It has released
The
It looks like Nvidia's starting to abuse their market status by trying to force everyone onto their systems or at least to make it difficult to have alternatives. You can see a similar situation in the current adaptive sync Gsync / Freesync conflict where one became VESA standard (Freesync) and the other became proprietary and in general more expensive. I'm honestly considering avoiding Nvidia products at the rate they're going.
The slowest of these cards does over 2 teraflops, there's no way you can remotely use that level of performance and features in games with an open source driver anyway.
Why not? And what about other GPU-intensive operations, such as real-time rendering or (some) bitcoin mining?
What I'd like to know though is what features are locked out of the OSS driver?
wat
WTF is this post supposed to mean?
Simple, The graphics card GPU is faster than the main CPU can shovel data for it to process.
Not that I agree with the statement though. I think we are seeing a lot of processing being off loaded from the CPU and pushed onto the GPU. There is a lot of stuff a GPU can do much faster than the CPU, especially when doing modeling and rendering of physical objects or other math that lends itself to being done on GPUs.
"File to fit, pound to insert, paint to match" - Aircraft Maintenance 101
I mean these are so ridiculously powerful cards that if one buy one, that may be because you wanted to run some demanding and advanced game. There are at least a handful available now for linux desktops. But if you use an open source driver, and it manages to run the game without crashing or debilitating bugs, the driver will likely bottleneck you so much you get like 10% or 20% of the performance.
Way to waste a computer upgrade, both GPU and CPU - you do need to upgrade the latter to play advanced and recent games, too.
Because they have TRADE SECRETS to protect. Secretes which are both theirs and ones that they have licensed and contractually are bound to protect.
I don't think they are anti-open source, they are just trying to protect their intellectual property. They are still releasing drivers for these devices and although you may not be entitled to see the source, you can still use that open source operating system with that shiny new video card.
"File to fit, pound to insert, paint to match" - Aircraft Maintenance 101
If (or when) NVidia stops putting effort into supporting Linux enough to produce drivers that are of a comparable quality to their larger markets is when you'll really start to hear an outcry. People are complaining now, but that's nothing compared to what will happen if or when NVidia decides that Linux is just not worth any effort to put any quality amount of effort into.
Of course, as I said... by that time it will be too late.
So... AMD or NVidia... it reminds me of an election where there are are really only two viable candidates and both of them suck.
File under 'M' for 'Manic ranting'
Now that's better. I'd give that 4/10; I almost thought you were serious until I saw the part about the future success of Windows on the Raspberry Pi.
>Why sould a GPU manufacturer spend a lot of time supporting such a small user base
I don't know, maybe because most super computers on the fucking planet use GPUs? Why would scientists want a GPU manufacturer to support the operating system they do most of their work on? Oh, I can't think of a reason.
Meanwhile, we're trying to do some work in ROS. I certainly don't want CUDA cores to help speed up the processing and filtering of tens of thousands of LIDAR points. Nor could I possibly use shaders for anything outside of gaming.
This much sarcasm is killing me. Please get better opinions before I die.
My post wasn't a question of "what does 2 teraflops mean", it was a question of what the fuck "there's no way you can remotely use that level of performance and features in games with an open source driver anyway." is supposed to mean.
It's a gibberish sentence.
I don't know, maybe because most super computers on the fucking planet use GPUs?
Still a very small market. Lets see, they can spend resources working on the next card that can make them million or spend the same resources suppoting a small market that may make a few $100K. If you ran the company which would you choose?
Why would scientists want a GPU manufacturer to support the operating system...
It is not NVIDIA's job to support scientists. Their job is to make money for their stockholders.
Nor could I possibly use shaders for anything outside of gaming.
How is a private company obliged to support your project?
Sorry but "they re not allowing me to do what I want" just sounds very entitled to me.
PS. Using profanity just makes you appear to be an illiterate idiot.
Its been doing well, but...
Omne ignotum pro magnifico.
But if you use an open source driver, and it manages to run the game without crashing or debilitating bugs, the driver will likely bottleneck you so much you get like 10% or 20% of the performance.
So you're implying that an open source driver can't be as good as a closed source driver?
How is the open/closed status of the source code even remotely relevant to it's performance. There is a lot of open source software that is considerably faster than the closed source equivalent.
You're posting complete and utter garbage.
Most likely because they aren't doing any of the firmware signature verification in hardware and so releasing specs that would allow people to write drivers to load arbitrary firmware would make it trivial for malware to do the same (not that a bit of reverse engineering won't do that eventually, but companies do love security by obscurity).
I am TheRaven on Soylent News
What a fucking jerk.
I find it remarkable that people so often confuse pure FOSS driver support with linux support.
How many times does it have to be stressed that nvidia supports linux by providing quality closed source drivers?
If that's not good enough for a minority of linux users, well tough. It's not everybody's problem and anybody who wants to do real work with their nvidia gpu on linux probably finds that the nvidia provided driver works great.
Why would those scientists you mention need to use FOSS drivers?
Valve has little to no Linux gaming clout. Ya they released a rebadge of Ubtunu with Steam on it. Yay. So far it has had very little influence. Most people continue to game on Windows (and to a lesser extent OS-X). They are not migrating in droves, nor are there droves of people who used Linux but didn't game that are now. Valve has changed very little in the Linux gaming space, as of yet,
The Unity engine and Kickstarter have done a lot more for driving any sort of Linux gaming than Valve.
Most of nVidia's gaming customers play on Windows, and they don't care about closed source drivers. Indeed, binary drivers are the way of things, the users would be extremely mad if you gave them source packages and told them to download a compiler. On OS-X it is all Apple's way, all the time. You gets the drivers you gets from Apple and live with it. Only in the Linux arena is there any wish for OSS drivers, and then only form a minority of their customers. Most of nVidia's Linux customers are high end enterprises, doing simulations or CAD work. They want certified binary drivers, because they want everything to be verified to work.
Valve really doesn't have much they can do to change nVidia's mind. I mean maybe if Valve themselves made Steam Machines and they could threaten to change vendors, but they don't, all kinds of hardware companies make them and they all do business with nVidia.
Well, they do support the OS, just not the OSS drivers. For someone who's primary concern is getting their work done on these machines the differnce is not that important. The fight for an OSS driver is mostly due to philosophical and tinkerer issues.
Since the very reason given since the discussions began 15 or so years ago, Nvidia, and most of its competitors (Intel being a special exception for an unrelated reason) have always said that due to fears and concerns about reverse engineering, they - Nvidia and ATI, now AMD, have been slow and limited in making available any documentation or assistance that could directly or indirectly ease reverse engineering of its technology, its intellectual property (IP); not to Open Source / Free Software developers, but to potential and current 3D video card competitors.
Providing the direct firmware blobs, even if encrypted (to be decrypted in memory on the video card) does reduce the effort of a reverse engineering attempt. Perhaps legal or senior management has overruled the previous plan to make encrypted firmware blobs. I believe there was one or more blogs entries written about methodologies of bypassing the decryption of encrypted firmware blobs even when/if the decryption key(s) are secure stored in the Nvidia GPU, or at least recovering the decryption key which undoes a lot of work by Nvidia, and may cause violate terms of various patent / IP licensing agreements.
Nvidia could possibly go out of business if they were barred from obtaining necessary licenses allowing them to implement video codecs in hardware in their future products.
I suspect this, or some benign reason (Nvidia's Linux developer were simply busy with in-house development, or on holiday) is the culprit.
* Unrelated pure speculation:
My pet theory about why Intel has been so open with their open source driver support for Linux, is that it is intended to be a) to support their APU processors and b) to try to help AMD in its secondary market (video GPUs) rather than their primary market (x86 compatible processors) which Intel knowns AMD needs to keep being a viable option, as AMD's x86 processors alone the past few years could of easily drove it out of business.
To avoid more anti-trust violations / investigations Intel needs at least one viable x86 competitor to remain alive. Preferably neither too far ahead nor behind, so that Intel continues to dominate the CPU manufacturing sector, it has at least something that is realistically a potential threat to their business. Just not a strong potential threat. But by possibly supporting AMD's secondary product line by providing an open book to their GPU's documentation and interface via their driver source code, Intel can provide a subtle nod to technologies, or other solutions that AMD could re-implement to improve their (AMD's) video card offerings.
In summary Intel can stand to help AMD in their video cards to keep AMD alive, which serves a critical purpose to Intel, as Intel needs someone that can be seen as potentially a rival CPU manufacturer.
Regarding Intel's domination of microprocessors:
While ARM processors have shipped in record numbers the past few years, they are manufactured by various companies who pay ARM a royalty (per unit made AFAIK), so Intel remains the single largest designer and manufacturer of CPUs. Although ARM Inc. has experienced explosive growth and tremendous profitability, it is still a tiny company in relative terms, such as market capitalization (a common benchmark) compared to Intel.
Why would those scientists you mention need to use FOSS drivers?
Because they're not playing games, they're using the GPUs' computational power, and the proprietary drivers don't work for that presumably.
Serious comments are useless when the discussion is being drowned in comments from paid shills.
This discussion has made it obvious that Slashdot has been completely taken over by shills, and serious discussion can't be done here any more.
Since you can not seem to be able to have a civil conversation I will leave it here.
If AMD wants an opportunity for a couple more points of market share, here it is. Be friendlier to Open Source than your competitor.
the major advances in civilization are processes which all but wreck the societies in which they occur - A.N. White
Is Valve some sort of major customer of Nvidia GPUs?
No, but Valve's users are.
Video games are a leading application for GPUs. The four hardcore video game platforms are Nintendo's AMD-powered console, Sony's AMD-powered console, Microsoft's AMD-powered console, and the PC. With another company owning the console space, NVIDIA's GPU business has to compete for PC makers and PC users with other GPU makers (AMD and Intel). And if PC games work poorly with NVIDIA products, PC users will have little reason to buy NVIDIA products. Valve runs a leading video game sales channel on PC and has developed several pairs of popular PC games, and it is looking to extend this sales channel to Steam-branded PC hardware that does not ship with a Microsoft operating system. So in order to be part of this segment of the market, NVIDIA has to ensure that its products continue to "work 100% with Linux" without causing serious instability.
That being their drivers suck. Also that writing GPU drivers is hard and the OSS community hasn't done a good job.
AMD released a bunch of hardware info, and what code they could (they can't just open up all of their proprietary driver, there are things in it they legally can't release). There were claims of an absolutely amazin' driver that would be made, better than Windows, that there were thousands of skilled OSS programmers who were chomping at the bit to work on it.
Well that was mostly just people bragging on places like /. who didn't know what they were talking about, someone who'd fooled around writing a NIC or SATA driver and thought it was easy. Turns out graphics drivers are REALLY COMPLEX and each generation of hardware needs a new one. So the AMD OSS driver has been pretty poor quality. I mean it works, and supports some features, but it has some stability issues and is nowhere near the full feature set.
So ya, not really helping them. What the OSS community wants is for someone to write an nVidia quality driver, and open it up. Do all the work and then hand it out. Doesn't seem like anyone is interested in doing that. In part that is because some of what makes those closed drivers good is IP that gets licensed that can't be open sourced.
Because they have TRADE SECRETS to protect. Secretes which are both theirs and ones that they have licensed and contractually are bound to protect.
I don't think they are anti-open source, they are just trying to protect their intellectual property. They are still releasing drivers for these devices and although you may not be entitled to see the source, you can still use that open source operating system with that shiny new video card.
I keep receiving mailouts which suggest that US patent rules have changed in recent years such that keeping trade secrets is an increasingly advisable business strategy, instead of acquiring patents.
I don't know if that's true, but it could be part of what's going on.
Because they have TRADE SECRETS to protect.
No. They don't want to protect the binary blobs from your eyes. They're not encrypting, they're signing. They want to prevent you from developing your own blob, by having your video card reject firmware not written by them.
I don't think they are anti-open source,
It's not a matter of opinion. They are anti-open source by definition, it's a fact dictated by their actions. They're locking down the cards that they manufacture in order to prevent their owners from writing open-source software to drive them. You can't get more anti-open source than this. Nvidia have always been anti-open source, and they are getting worse and worse with time.
You haven't paid attention lately then. Graphics cards and SSDs are the few computer components where lots of innovation still happen and where there still is need for, and room for, improvements. They have long been held back by the 1920x1080@60Hz plateau of affordable display panels and the bandwidth limits of HDMI/DVI/DP interfaces but now with the 4K panel boom, FreeSync/G-Sync and DisplayPort 1.3, things are moving again because now once again the graphics card has become the bottleneck. Not even tripple SLI of the very expensive Titan X is enough to run the latest AAA games at 3840x2160@120Hz with all settings at max. You are correct that the past half decade has been fairly boring in this area though, even though some great things have been brewing during this time (G/FreeSync, Mantle/Vulcan, HBM, etc).
My other account has a 3-digit UID.
How exactly does handing a binary blob to the Nouveau developers reveal any trade secrets? The binary blob is handed out in the driver anyway, it is just a pain to extract now.
This is just an attempt at killing Nouveau. It will most likely succeed.
Finally! A year of moderation! Ready for 2019?
Oh good someone else understands
Yet another person who does not know the difference between "I have a different opinion" and "troll". Here is a hint; "trolls" use profanity. As an ex-military friend of mine says, "Pot pot this is kettle, kettle. Colour check, over."
I don't understand what you mean by 'non graphics competitors'. Intel, AMD, and ARM cpu offerings already have integrated GPUs with dual-head capability (and have for a few years now). There are no non graphics competitors.
Currently the best open source kernel and driver compatibility is with the Intel and AMD integrated GPUs. That's what all the KMS work was responsible for giving us. The performance of integrated GPUs has increased steadily over the last few years and has reached a point now where most 3D games will run with modest (but not high-end) settings, and *all* 2D (aka desktop operations) will run faster than you can blink.
I splurged for a mid-range card for my windows gaming box, but all my workstations just use the cpu-integrated gpus these days for dual-head operation. And they're nice and quiet and fast.
-Matt
Just about everything the card does has already been done open source by nouveau, including the firmwares described in this article. It does change a bit between card generations, but not too much. You can already go read the reverse engineered source for these firmwares (go on, it is distributed with the linux kernel source). This is simply a matter of the card not accepting non-signed firmware. There are no trade secrets protected by this.
They should hide their secrets in the chips themselves, or in the firmware for them. Their stubbornness about redistributing firmware images is retarded since they're useless without having an nvidia card anyway..
The past half of a decade involves idle CPUs and GPUs, and people wondering why they're getting low FPS. Because programmers are bad. No one knows how to write scalable code.
stated goal of legitimizing Linux Gaming
NVidia have freely available, user hardened, linux drivers for all of their hardware, and a large scientific/gaming community that uses them. Same deal for NVidia's windows drivers.
Will they [boycott NVidia]? Probably not.
...because...
- They will flush 50% or more of their own revenue down the toilet.
- It sounds too much like extortion/ant-trust, and is probably illegal.
- NVidia already comply with their stated goal.
And did you exchange a walk on part in the war for a lead role in a cage? - Pink Floyd.
I've made fun of Intel's GPU offerings for years. The Intel i810 I had in my PIII was barely adequate for 2D desktop work. It didn't support VESA modes above 640x480 without a driver.
Intel GMA910/915 was a piece of junk too. Intel apparently filled so many warehouses full of them, they encouraged Microsoft to allow this non-Aero capable hardware to be stamped as "Vista Capable" and was central around that boondoggle.
Intel GMA945/955 was the definition of the bare minimum requirements to support Aero, and drastically underperformed the bargin bin offerings from the era by ATI (x1200) and nVidia (6150). Intel continued to sell these to Atom users years after they should have been killed.
On some Atom platforms Intel also packed PowerVR based GPUs (which hardware wise were ok) with complete shit drivers for both Windows and Linux.
That said when I built my latest desktop (I don't do gaming) I was satisfied with the built in graphics on the Haswell chipset. It can drive 3 monitors without issue. HD video no problem. Even the latest Atom offerings have GPUs based on a scaled down model of it.
Intel continued to sell these to Atom users years after they should have been killed.
Killing the Atom users seems relatively merciful rather than continually being sold Intel video cards...
Inheritance is the sincerest form of nepotism.
is going to run out of fingers.
Buy your next Linux PC at eightvirtues.com
18,688 Nvidia Tesla K20X GPUs
How many of those have been built?
Learn to write calm counterarguments and leave out the ad hominem attacks.
Without the sources we can't even begin to know what our computers are doing either.
We already have more open source than we can handle. The biggest bottleneck is funding, manpower and quality assurance.
There is a lot of open source software that is considerably faster than the closed source equivalent.
Can you mention an example?
My "deprecated" Geforce MX 440 card works simply great with Nouveau. I can watch accelerated videos (tested up to 720p) with vlc/mplayer.
Does this "acceleration" actually mean something more than just using an YUV overlay?
>> Simple, The graphics card GPU is faster than the main CPU can shovel data for it to process
So, basically to use a common example, the GPU is kicking out bitcoin so fast that the CPU can't keep it saturated? Would you care to rethink your claim?
I am very small, utmostly microscopic.
You know, sometimes this pretentious anti-entitlement bullshit is just getting too much on my nerves.
Sometimes it's just a waste of time to argue with such people. They just state something and then you are expected to come up with a referenced list of refutations, while in reality, they were just trolling/astroturfing/swaying public opinion.
Of course people, who are pushing ahead our capabilities are entitled to more demands than beancounters, who only focus on the balance of possible gains and expenses vs. risk.
Vis-a-vis jklovanc, this would have been a moot argument, however, as apparently, he already formed an opinion - and we all know how easy it is to prove someone wrong and have him/her admit it.
So I decided to do the next best thing and reinforce the opinions of entitled people to actually continue to feel entitled and push the boundaries further.
>> Simple, The graphics card GPU is faster than the main CPU can shovel data for it to process
So, basically to use a common example, the GPU is kicking out bitcoin so fast that the CPU can't keep it saturated? Would you care to rethink your claim?
Did you even read my post? I was paraphrasing what I understood somebody else to say, then disagreeing with them.
Most GPU's are not used for graphics processing alone anymore. If they where just for graphics, most game programs would saturate the CPU and the data buss to the GPU before the GPU processor would run out of free cycles. They are WAY more powerful than is needed just for graphics...
Your BitCoin example is a classic illustration of what GPU's are really doing these days.
"File to fit, pound to insert, paint to match" - Aircraft Maintenance 101
Do you have any reason to suspect so?
Nothing in the original posts gives any indication.
In fact the only thing mentioned is how unfriendly the new chips are to FOSS drivers. Nothing else of substance
All the other assumptions about scientists, inadequate linux closed source drivers for CUDA (or whatever they will be using) seem to me like unfounded wild speculation.
I'm sure whoever used such chips would want maximum performance anyway, so closed drivers would be the preferred choice even is FOSS drivers sort of worked
It isn't developing at all anymore. They are only adding mass, Watts, heat and fans.
AMD's shift away from VLIW toward SIMT doesn't count, then?
The architectures are still evolving. They're not just throwing more transistors at the same old ideas.
If nobody knows how to write scalable code, that's a computer science problem, not a problem in the vast legions of programmers.
Contribute to civilization: ari.aynrand.org/donate
I guess you don't use Ubuntu.
>> So what's your problem then, hacker?
Really? maintaining/repairing my own car is now hacking? Your stupid thinking exactly represents the problem not the solution.
>> replace the damned ECU
a) there isnt one available for my car (Jaguar)
b) even if there was, fitting it would break the warranty
c) why should I have to pay an extra 3k just to be able to work on my own car?
How many supercomputers use the GM series GPU which is what this conversation is actually about. The K20 units use a different GPU and are beyond this conversation.
Linux internal APIs get changed from time to time. Drivers that are part of the kernel tree get updated by whoever is changing the API. Drivers that are outside the kernel tree have to be updated to work with the API changes.
I've had my distro ship a new kernel and then had to wait weeks for the Nvidia driver to be updated. (Admittedly they're getting better at it, but it has happened.)
Yes. See Valve Finds Open Source Drivers To Be Great; Valve Developed an Open-Source Intel Vulkan GPU Driver For Linux; Valve Bootstrapped Source 2 Engine On an Open-Source Vulkan Driver.
PS. Using profanity just makes you appear to be an illiterate idiot.
Your lack of profanity exudes pontification... but it's the lack of depth in your miasma of thoughts that accentuate your defeat.