NVIDIA Begins Providing Open-Source 3D Driver Support For GeForce GTX 900 Series (phoronix.com) 63
An anonymous reader writes: In late 2014 NVIDIA announced their GPUs would begin requiring signed firmware images before the open-source driver could enable hardware acceleration. That led the Nouveau developers to call the latest GPUs "very open-source unfriendly", but that criticism can now be laid to rest as NVIDIA has finally released the signed firmware and basic open-source driver code. The open-source driver can now move on with its open-source 3D enablement for Maxwell GPUs and the NVIDIA developer is hoping it will be ready for the next kernel cycle (Linux 4.6).
still a binary blob then? (Score:1)
soo...
we are still waiting for the source-code for the binary blobs, and keys required to sign our own firmware?
Re:still a binary blob then? (Score:5, Insightful)
I hate what you wrote because I love open-source but I have seen all these complaints before.
Company doesn't open source and they are the devil
Company partially open sources and they are the devil for not opening it all
Company completely open sources and then we hear that they are just trying to get free labor and they are still the devil.
Re: (Score:2)
To what company does the last statement apply? I can't think of a single instance. Any instance that comes to mind refers to a partially open source situation, not a completely open source situation.
Pretty much every time you see someone complaining about firmware. It doesn't run on the CPU, it could just as well have been read from an EEPROM chip but for cost reasons they want the driver to send the blob during initialization instead. Personally I never understood the vile difference between loading a proprietary blob to a chip once with flashing and sending it in each time on boot, but apparently RMS and friends go mental over the latter. Also, often the firmware is blocking things that could harm th
Re: (Score:1, Insightful)
THERE ISN'T ANY CARD THAT DOESN'T HAVE PROPRIETARY TECHNOLOGY IN IT THAT THEY DON'T WANT TO JUST GIVE AWAY TO THE WORLD?
Um, apart from AMD GCN which has proprietary technology in it and yet they still document the card's registers and ISA.
In fact, Intel manage this on their processors, which is a way more competitive and secret world than GPUs.
The reason nVidia hide shit is because their shit smells bad and they don't want that to get out. It has nothing to do with "secret sauce".
Documenting your products'
Re: (Score:2)
Re: (Score:1)
20 end
laid to rest? (Score:3)
requiring signed firmware is still open source unfriendly! if the firmware can be changed, we want an open source version of that too! we also want to be able to run our own code on it. signed firmware is a hostile statement saying that you don't want anyone else to be able to write firmware for this card.
Re: (Score:1)
Signed firmware isn't bad since generally firmware can't be changed. Firmware doesn't need to be signed but Nvidia is probably doing this for security reasons.
Re: (Score:2)
Firmware is basically just another word for "software that runs on a peripheral processor." It can almost always be changed. Technically the Android install on your smartphone is "firmware" and the broken router software on most cheap routers is "firmware." Typically both can be replaced with 3rd party builds - unless the firmware is signed, in which case you're fucked if there's something wrong with it.
The caveat with this, of course, is w
Re: (Score:2)
[...] if the firmware can be changed, we want an open source version of that too! we also want to be able to run our own code on it. signed firmware is a hostile statement saying that you don't want anyone else to be able to write firmware for this card.
The signed firmware is not intended to interfere with the consumer / user, in fact one of the key justifications is continuing to provide post-manufacture updates of the video card firmware to provide fixes and enhancements, while preventing counterfeiting where low-end cards are re-flashed with bogus firmware that factory overclock it and reports itself as a more capable higher-end (more expensive) model. Nvidia claims to have found unauthorized manufacturers / re-packagers selling such cards in Asia.
User
Re: (Score:2)
Users / non-Nvidia developers do not have the necessary technical documentation to produce their own firmware, so there is no lost of functionality or flexibility. Their video cards designs, ASIC, and firmware are all proprietary design, with almost no technical documentation available to open source developers.
right... because reverse engineering isn't a thing and patents don't contain any information. what this does is prevent people from being able to experiment and succeed at writing their own firmwares.
Re: (Score:1)
Except in the cases of some AMD and nvidia models where the power saving features weren't programmed with proper settings, which led to gray-screening. End users had to modify the card firmware to change the stepping values (or disable power saving entirely in some cases). There were also a few cases where overclocking the cards wasn't allowed via the drivers (softlocks), and that also required manual editing and flashing of firmware to get around.
Re: (Score:2)
If not, then lets get all the torches and pitchforks ready. Not to mention a big OSS to burn into nVidia's front lawn.
If so, then people need to shut up about it. Companies (that actually develop stuff) need to protect their IP.
Re: (Score:2)
The proprietary drivers (and firmware) include agreements that nvidia made with other companies, so they can't open source them. The problem here isn't even the existence of the firmware that was always loaded, it was that the normal workaround no longer worked, because nvidia was seeing scammers loading firmware for more advanced cards on less advanced cards- this didn't improve the cards, but it DID let them lie about what they were. I'm glad they are coming up with a solution for the open source stuff.
The trouble with unsigned firmware (Score:2)
Re: (Score:2)
requiring signed firmware is still open source unfriendly! if the firmware can be changed, we want an open source version of that too! we also want to be able to run our own code on it. signed firmware is a hostile statement saying that you don't want anyone else to be able to write firmware for this card.
If you could load your own firmware you would probably be a little bit closer to being able to bypass HDCP (High-bandwidth Digital Content Protection). Chances are Nvidia have signed a crap load of agreements that prevent them from letting you do that. They could probably invest a load of time in letting you run your own firmware, but have the windows driver scan for that and disable HDCP in this case but even this may prove awkward if it made it any easier for you find a way around the HDCP in older cards
heh (Score:1)
Only software written by Stallman himself is good as everything else is a potential patent troll and/or vendor lock-in.
Re: (Score:2)
Only software written by Stallman himself is good as everything else is a potential patent troll and/or vendor lock-in.
Preach it, bro.
https://www.youtube.com/watch?... [youtube.com]
Personally, I don't care. (Score:4, Insightful)
For me FOSS is a tool, it's not something to get emotional or tied up about. I pay Nvidia money Nvidia pays people to develop drivers that work. I paid AMD/ATI money and they said "Ha, here's a shit ton of specs, write them yourself". Sorry. My job isn't to write display drivers, my job is to use the display drivers.
I suppose I could try growing my own food too, but I (gladly) pay someone else to do it for me. Even if it is a bit 'closed source'.
AMD's opensource is good (Score:4, Informative)
And how good is that AMD driver?
Their opensource driver (yup, they are also having some of the opensource driver developers on their own payroll) is actually pretty good.
(As often shown on phoronix benchmarks).
To the point that the open-source driver is the officially supported driver for older hardware platforms that get dropped out of catalyst. Usually by the time a card isn't supported by catalyst anymore, the opensource driver is demonstrating nearly as good performance (and on a few occasions, even better).
The situation is quite different from Nvidia.
With AMD:
- the closed source driver is so-so. It's buggy and crashy, but at least there's 1st day support for newer hardware generations or newer features.
- the open source driver is really good (thanks to actual input by AMD, both documentation *AND* paid position)
- AMD is putting lots of effort in that direction. It progresses very slowly, takes time, but looks promision. AMDGPU is such an exemple (moving to a stack which is mostly open-source, with catalyst being only a proprietary openGL library running above the opensource component as an alternative to the opensource Gallium3D/Mesa's OpenGL state tracker).
With NVidia:
- the closed source driver is high quality. But it's basically an almost straight recompile of the windows stack. So you're fucked up if you need a feature that Linux does differently (hybrid Intel+Nvidia grpahics on laptop was such an exemple). Also fuck you if you use a newer kernel than what they are currently supporting (you can't use a rolling distro or a 3rd party repo to get the latest kernels, because Nvidia's drivers rely on their own special shim driver). And don't forget "fuck you too" if you have older hardware, they'll drop support from newer drivers, and only seldom support older generations of driver for you.
- the open source driver is a mixed bag anywhere between total crap and more or less working. Not at all the fault of the developers. They are basically on their own it's a miracle what they managed to pull off with such a meager support.
- Nvidia mostly doesn't give a crap about opensource development. From time to time, they might decide on a whim to be nice for once and throw a bone. Usually when it also helps Tegra developement and just happens to have some use for desktop cards.
- The exception is the Tegra mobile platform. Given the overly dominant position of Linux in the embed world, Nvidia are regularily providing some help to the opensource Tegra support.
So if you want opensource drivers not only for ideological reasons but also practical reasons (like having a rolling distro and/or latest kernels).
AMD is the definitive GPU maker to go to.
If you want the best performance ever while not minding whatever code you run, Nvidia is you best choice.
(Just hope you won't land on one of the few "not supported" sore points, like laptops).
Re: (Score:2)
You have a great post, but it's Linux specific. There is only one choice on FreeBSD and that's NVIDIA. They're the only company to support FreeBSD. AMD has no commercial driver for FreeBSD and the open source attempts are weak on BSD over the last 5 years. FreeBSD and DragonFly developers have put in recent efforts but it's hard to track all the crazy changes in the linux kernel lately.
Re: (Score:2)
hybrid Intel+Nvidia grpahics on laptop was such an exemple)
Sorry to rain on your parade but this is shit on AMD cards too. The closed source driver AMD just discontinued (or not updated so it can run on fedora 23) actually worked better on my work Dell E6540 laptop. Some people I work with who started just before me are lucky enough to have the Nvidia version of this laptop and this works better with Linux.
Re: (Score:2)
Well that's nice for you. Do you normally go into topics and say "I don't care about this, I don't see why anyone would?"
Re: (Score:2)
When we see the same thing posted over and over for a decade, yes.
I'm still buying Nvidia cards because they work. I do have some AMD paperweights if anyone wants them.
Re: (Score:2)
I'm still buying Nvidia cards because they work. I do have some AMD paperweights if anyone wants them.
They must all be old, right? If any of them are vaguely current, send me one (I'll pay shipping obviously) and I'll write an article about how the drivers are or aren't now
Re: (Score:2)
OS: FreeBSD
Card: Radeon 7870 [amd.com]
Meanwhile my old Nvidia cards are still cranking away with working drivers (with acceleration!)
It's like I paid Nvidia money and they delivered a product I could use.
Unless you're volunteering to read through AMD's spec sheets and writing me a driver for free.
Additionally under Linux AMD has a 'bug' (Feature?) where you can't even use them under OpenCL without a device attached. There is a way to fake it with a resistor but if you're going to bank on headless GPU computing you'd
Re: (Score:3)
Radeon 7870
Those older radeon work very nicely with the opensource drivers (r600 driver on radeon kernel module), and AMD is devoting resource to this driver.
The problem here isn't AMD's own effort. The problem is you're using an OS that is NOT officially support by AMD anyway.
AMD do support Linux. They don't officially support BSDs (though there has been some announcements that they might eventually).
Usually driver support for BSD comes in the form of mending Linux driver code into working with BSD kernel.
Meanwhile my old Nvidia cards are still cranking away with working drivers (with acceleration!)
It's like I paid Nvidia money and they delivered a product I could use.
Unless you
Re: (Score:2)
The problem is you're using an OS that is NOT officially support by AMD anyway.
So what you're saying is I should continue to pay money to the company that develops drivers for my OS? How is it that Nvidia can support as many OSes as they do for nearly the same cost as AMD?
It's a thing that is currently being worked on, and that should disappear once the stack has finished migrating to AMDGPU.
How long does it take to fix a bug this big?
Whereas with Nvidia? It's just "go fuck yourself" (unless you're referring to Tegra).
But I don't care Nvidia provides hardware that works with the software. Most people don't care. We don't have time to sit around and wait for someone to fix it eventually. I'm glad my job doesn't depend on that AMD "bug" being fixed. It's going on 3 years now(?). I'm not pa
Re: (Score:2)
NVIDIA sells more GPUs than AMD in discrete cards. They also aren't in a losing battle with Intel on x86. I'd say NVIDIA has a lot more money to play with.
On the AMD side, they have some decent GPUs, with no OS support. They also have several failed die shrinks and haven't released a real CPU product since 2012 for the desktop. They're probably really hurting right now.
Re: (Score:3)
I paid AMD/ATI money and they said "Ha, here's a shit ton of specs, write them yourself". Sorry. My job isn't to write display drivers, my job is to use the display drivers.
That's weird. I paid AMD and I got decent binary drivers. Not perfect, but decent for my purposes. I know they also provide some specs for open source devs, but so far I haven't found the open drivers good enough.