NVIDIA Cleans Up GSP Firmware Binary License

[oiaohm]

Indeed, the problem affects legacy Nvidia GPUs devoid of micro-controller which are not so old in present days.

GeForce RTX 20 series and newer that 2018 and newer that have the micro-controller.. So yes 4 year old cards.

Moving from Nvidia proprietary drivers to MESA drivers causes the end-users to lose many benefits in some browsers, as example chromium or chrome in terms of video acceleration and video decode.

AMD key example. AMD has closed source userspace and open source user space but open source kernel level driver share between both of them.

There is no particular reason other than Nvidia unwillingness to work with others why Mesa and nvidia closed source userspace could not use the same kernel mode driver.

Nvidia has not given clear statements on firmware older than GeForce RTX 20.

adaptive power management which allows to increase and decrease performance and power consumption

This bit is critical for why Mesa with Nvidia is the way it has been. There has been no point attempting to implement video acceleration and video decode without these controls. All you do is end up doing on default Nvidia GPU power settings if you attempt to run video acceleration and video decode is have the GPU screw up badly as it runs it self out of power and proceed to lock up and do many other stupid things.

The reality here is Mesa drivers have been unable to developer legally for Nvidia hardware because of not having the legal right to ship the firmware to run the power management. Without the means to control the power management many of the advantages of using Nvidia hardware are totally not usable and attempting to use them will just lock the card up..

Yes the problem of older card of Nvidia are not solved. The firmware blobs for older cards lot of cases are embeded inside the provide object file not their own firmware files either that brings another issue of legality.

[MorrisS.]

GeForce RTX 20 series and newer that 2018 and newer that have the micro-controller.. So yes 4 year old cards.



AMD key example. AMD has closed source userspace and open source user space but open source kernel level driver share between both of them.

There is no particular reason other than Nvidia unwillingness to work with others why Mesa and nvidia closed source userspace could not use the same kernel mode driver.

Nvidia has not given clear statements on firmware older than GeForce RTX 20.

adaptive power management which allows to increase and decrease performance and power consumption

This bit is critical for why Mesa with Nvidia is the way it has been. There has been no point attempting to implement video acceleration and video decode without these controls. All you do is end up doing on default Nvidia GPU power settings if you attempt to run video acceleration and video decode is have the GPU screw up badly as it runs it self out of power and proceed to lock up and do many other stupid things.

The reality here is Mesa drivers have been unable to developer legally for Nvidia hardware because of not having the legal right to ship the firmware to run the power management. Without the means to control the power management many of the advantages of using Nvidia hardware are totally not usable and attempting to use them will just lock the card up..

Yes the problem of older card of Nvidia are not solved. The firmware blobs for older cards lot of cases are embeded inside the provide object file not their own firmware files either that brings another issue of legality.

What would be the solution to this problem other than to change GPUs?

[oiaohm]

What would be the solution to this problem other than to change GPUs?

It is fairly much Nvidia need to-do things that then allow the open source developers to-do things.

Before Nvidia cards required signed firmware the mesa developers would reverse the card and write their own firmware of course this does come with is fair share of opp we allowed a lower model card to out perform higher model card and opps we nuked the card issues. Both did not make Nvidia exactly happen because opps we nuked the card did not stop some underhand RMA those and the performance ones causes Nvidia market issues.

https://lpc.events/event/16/contribu...2/1919/nouveau in the times of firmware.pdf

Signed firmware requirement appears on nvidia cards in 2014. "GeForce RTX 20 series and newer that 2018" So there is still 4 years problem that open source developers cannot work around in any country that support DMCA instead these firmware files need to be license by Nvidia so the open source developers can use them without them those cards cannot be used with full functionality. Of course those firmware files you need the documentation on the API/ABI so that driver can give those firmware files directions Nvidia open source developers have been fairly good and doing this themselves but would be faster if Nvidia just released the documentation of course if they ever wrote of themselves.

This is something to be aware of. Nvidia OS drivers and hardware firmware does not appear to have a formal API/ABI between them instead the driver and firmware historically written at the same time and just modified to match.. This is different to AMD/Intel and most other vendors who have a defined API/ABI from firmware to drivers. This difference may equal that Nvidia has no documentation instead its been used the code as documentation.

Remember the hell that came to play when Intel with mainline Linux released a new firmware with changes API/ABI to driver in such way the new driver would not work with old firmware the Linux kernel open source mainline developers don't tolerate this. AMD and Intel have been obey the Linux kernel rules on API/ABI stability from firmware to OS driver. This is another area that may require change on Nvidia side.

Yes the thing to remember before signed firmware hit there were many cases that the mesa nvidia driver was out performing the Nvidia closed source drivers. So nvidia closed source drivers being superior to mesa ones may be nothing more than fiction caused caused by the monopoly control of firmware Nvidia had due to signed firmware and horrible restrictive license preventing distribution and use of the required firmware files to use the hardware correctly.

You cannot separate clause is that is removed in this release is really horrible because it means that where that clause is legal you must use the firmware with the nvidia closed source driver you cannot use it with mesa alternative. The Nvidia drivers for the 2014-2018 cards still have the cannot separate clause in there licenses. This change does not fix all the cards that have the signed firmware problem that prevents open source driver developers from doing their work. It would be good if Nvidia cleans up their license on all firmware on cards that can only be operated with firmware signed by Nvidia so that open source developers can maintain and make drivers for these cards.

Of course it going to take some time for the open source driver developers to get use to using what has now been made usable. Nothing changed quickly.

[MorrisS.]

It is fairly much Nvidia need to-do things that then allow the open source developers to-do things.

Before Nvidia cards required signed firmware the mesa developers would reverse the card and write their own firmware of course this does come with is fair share of opp we allowed a lower model card to out perform higher model card and opps we nuked the card issues. Both did not make Nvidia exactly happen because opps we nuked the card did not stop some underhand RMA those and the performance ones causes Nvidia market issues.

https://lpc.events/event/16/contribu...2/1919/nouveau in the times of firmware.pdf

Signed firmware requirement appears on nvidia cards in 2014. "GeForce RTX 20 series and newer that 2018" So there is still 4 years problem that open source developers cannot work around in any country that support DMCA instead these firmware files need to be license by Nvidia so the open source developers can use them without them those cards cannot be used with full functionality. Of course those firmware files you need the documentation on the API/ABI so that driver can give those firmware files directions Nvidia open source developers have been fairly good and doing this themselves but would be faster if Nvidia just released the documentation of course if they ever wrote of themselves.

This is something to be aware of. Nvidia OS drivers and hardware firmware does not appear to have a formal API/ABI between them instead the driver and firmware historically written at the same time and just modified to match.. This is different to AMD/Intel and most other vendors who have a defined API/ABI from firmware to drivers. This difference may equal that Nvidia has no documentation instead its been used the code as documentation.

Remember the hell that came to play when Intel with mainline Linux released a new firmware with changes API/ABI to driver in such way the new driver would not work with old firmware the Linux kernel open source mainline developers don't tolerate this. AMD and Intel have been obey the Linux kernel rules on API/ABI stability from firmware to OS driver. This is another area that may require change on Nvidia side.

Yes the thing to remember before signed firmware hit there were many cases that the mesa nvidia driver was out performing the Nvidia closed source drivers. So nvidia closed source drivers being superior to mesa ones may be nothing more than fiction caused caused by the monopoly control of firmware Nvidia had due to signed firmware and horrible restrictive license preventing distribution and use of the required firmware files to use the hardware correctly.

You cannot separate clause is that is removed in this release is really horrible because it means that where that clause is legal you must use the firmware with the nvidia closed source driver you cannot use it with mesa alternative. The Nvidia drivers for the 2014-2018 cards still have the cannot separate clause in there licenses. This change does not fix all the cards that have the signed firmware problem that prevents open source driver developers from doing their work. It would be good if Nvidia cleans up their license on all firmware on cards that can only be operated with firmware signed by Nvidia so that open source developers can maintain and make drivers for these cards.

Of course it going to take some time for the open source driver developers to get use to using what has now been made usable. Nothing changed quickly.

So, there is no alternative to Nvidia proprietary drivers for recent legacy Nvidia GPUs at this moment. End-user can take benefit only by Nvidia proprietary drivers. Matter is whether to get them from repositories or from kernel itself by the partial open-sourced drivers. In this case end-user faces with the activation process of integrated Nvidia drivers.

[oiaohm]

So, there is no alternative to Nvidia proprietary drivers for recent legacy Nvidia GPUs at this moment. End-user can take benefit only by Nvidia proprietary drivers. Matter is whether to get them from repositories or from kernel itself by the partial open-sourced drivers. In this case end-user faces with the activation process of integrated Nvidia drivers.

No correctly functional alternative. Nouveau developers have attempted to have something no matter what.


Maxwell has second version the Maxwellv2 (yes annoying from retail package and printing on cards there is no difference between maxwellv1 and maxwellv2)

following is the list of you are possible cursed if you are Maxwell and 100 percent cursed by the remainder while Nvidia license on firmware stays the way it is.

https://nouveau.freedesktop.org/CodeNames.html

NV110 family (Maxwell)
Code name Official Name
NV117 (GM107) GeForce GTX (745, 750, 750 Ti, 840M, 845M, 850M, 860M, 950M, 960M)
Quadro K620, K1200, K2200, M1000M, M1200M; GRID M30, M40
NV118 (GM108) GeForce 830M, 840M, 930M, 940M[X]
NV120 (GM200) GeForce GTX Titan X
NV124 (GM204) GeForce GTX (970, 980)
NV126 (GM206) GeForce GTX (950, 960)
NV12B (GM20B) Tegra X1
NV130 family (Pascal)
Code name Official Name
NV132 (GP102) NVIDIA Titan (X, Xp), GeForce GTX 1080 Ti
NV134 (GP104) GeForce GTX (1070, 1080)
NV136 (GP106) GeForce GTX 1060
NV137 (GP107) GeForce GTX (1050, 1050 Ti)
NV138 (GP108) GeForce GT 1030
NV140 family (Volta)
Code name Official Name
NV140 (GV100) NVIDIA Titan V, NVIDIA Quadro GV100

Older than that you have reversed engineered solution to fire up the cards using open source drivers and that not without it problems.

Not exactly the activation process. Issue is Nvidia has a default firmware on cards that most of the time good enough to fire up the GPU to run bios and the like but anything past that you are in trouble. Please note I said most of the time Nvidia OEM/ODM have manged to release cards that you leave them sitting in bios they overheat due to slightly wrong firmware.

Yes the only way to fix that problem is Nvidia do action and change the license on the firmware required and possible give more documentation.

It quite a few cards still effected by the Nvidia made their firmware signed so we cannot do any power management problem. The reversed power management solution before that is not ideal either. Making firmware without full understanding of the internals of the hardware is not without it major risk of bad problems.

Nvidia been a problem for quite some time with Nvidia being the only party that can really correct the problem.

Legacy drivers think about it AMD and Intel legacy drivers get new features well after AMD and Intel has given up on them Nvidia users have been missing out on this as well.

Nvidia having the claim to being the highest performing has made them quite a pain to their users really. But just like apple you have the fan boys.

[MorrisS.]

No correctly functional alternative. Nouveau developers have attempted to have something no matter what.


Maxwell has second version the Maxwellv2 (yes annoying from retail package and printing on cards there is no difference between maxwellv1 and maxwellv2)

following is the list of you are possible cursed if you are Maxwell and 100 percent cursed by the remainder while Nvidia license on firmware stays the way it is.

https://nouveau.freedesktop.org/CodeNames.html

Older than that you have reversed engineered solution to fire up the cards using open source drivers and that not without it problems.

Not exactly the activation process. Issue is Nvidia has a default firmware on cards that most of the time good enough to fire up the GPU to run bios and the like but anything past that you are in trouble. Please note I said most of the time Nvidia OEM/ODM have manged to release cards that you leave them sitting in bios they overheat due to slightly wrong firmware.

Yes the only way to fix that problem is Nvidia do action and change the license on the firmware required and possible give more documentation.

It quite a few cards still effected by the Nvidia made their firmware signed so we cannot do any power management problem. The reversed power management solution before that is not ideal either. Making firmware without full understanding of the internals of the hardware is not without it major risk of bad problems.

Nvidia been a problem for quite some time with Nvidia being the only party that can really correct the problem.

Legacy drivers think about it AMD and Intel legacy drivers get new features well after AMD and Intel has given up on them Nvidia users have been missing out on this as well.

Nvidia having the claim to being the highest performing has made them quite a pain to their users really. But just like apple you have the fan boys.

It's not so much that Nvidia has fan boys, rather is that many end-users have Nvidia cards. So end-user is forced to make the choice more convenient. To use Nvidia proprietary drivers is the logical solution to avoid what thou statest in thy last post.

[oiaohm]

It's not so much that Nvidia has fan boys, rather is that many end-users have Nvidia cards. So end-user is forced to make the choice more convenient. To use Nvidia proprietary drivers is the logical solution to avoid what thou statest in thy last post.

There is a catch of course.
1) Nvidia users are a minority. 80%+ of x86 users don't have a dgpu at all.
2) Nvidia will end binary driver support before the hardware dies. This proble is a problem for parties like Redhat that can have support contracts on hardware longer than Nvidia supports their drivers for.

Convenient choice is one thing. Logical choice is don't buy Nvidia for systems in general. Yes only buy nvidia for special use systems because Nvidia has had users on a treadmill of you must by new card even when old card is not broken because we are removing driver support.

I said Nvidia fans because Nvidia users normally end up justifing their hardware by saying it has the best drivers and other stupid things that are in fact disprovable. Having to admit Nvidia is good because their silicon design is highly performing compared to everyone else and vendor lockin moves is the truth. Over and over again Nvidia driver stack has been shown lacking. zink with shadow tome raider recently shows the opengl section of the Nvidia drivers is under performing.

The horrible fact here lack of competition normally leads to lack of quality. Nvidia closed source drivers have had lack of competition because they vendor locked out their competition in a few different ways. Firmware of course is one. Another is Nvidia cuda being restricted from being used by other parties and not making working opencl to force those with Nvidia to use cuda.... and I could go on how Nvidia has worked to trap users and users have been falling for it.

Yes just because something is the best at something does not mean they have got to be the best by fair and consumer friendly ways.

[MorrisS.]

There is a catch of course.
1) Nvidia users are a minority. 80%+ of x86 users don't have a dgpu at all.
2) Nvidia will end binary driver support before the hardware dies. This proble is a problem for parties like Redhat that can have support contracts on hardware longer than Nvidia supports their drivers for.

Convenient choice is one thing. Logical choice is don't buy Nvidia for systems in general. Yes only buy nvidia for special use systems because Nvidia has had users on a treadmill of you must by new card even when old card is not broken because we are removing driver support.

I said Nvidia fans because Nvidia users normally end up justifing their hardware by saying it has the best drivers and other stupid things that are in fact disprovable. Having to admit Nvidia is good because their silicon design is highly performing compared to everyone else and vendor lockin moves is the truth. Over and over again Nvidia driver stack has been shown lacking. zink with shadow tome raider recently shows the opengl section of the Nvidia drivers is under performing.

The horrible fact here lack of competition normally leads to lack of quality. Nvidia closed source drivers have had lack of competition because they vendor locked out their competition in a few different ways. Firmware of course is one. Another is Nvidia cuda being restricted from being used by other parties and not making working opencl to force those with Nvidia to use cuda.... and I could go on how Nvidia has worked to trap users and users have been falling for it.

Yes just because something is the best at something does not mean they have got to be the best by fair and consumer friendly ways.

OK but Nvidia is a market company. Thou states that a market company should make free open sourced software. This is a non sense. I prefer Nvidia cooperate with open sourced developers to get alternative properly solutions. If all software was free no enterprise would produce it, because pieces of software are also goods to sell. How doest thou think to sell a free item? How can be CUDA free if it is an item an enterprise realize and sell? Why an enterprise should realize free item? A company should just provide a technical and informative support. This is where Nvidia is going to improve. Nvidia made free Physx driver code. Does Phisic work on linux? I don't think. So, it's not only a problem dealing with Nvidia. A company has to preserve industrial secrets. The reasoning is valid on avoiding hardware affected by copyright and as thou statest just Intel Nvidia and ARM should be the best proposals. Probably just elementary hardware as Adreno or Raspberry should be used by linux users. The other possibility will be when Microsoft will move to Linux operating system making the driver realization convenient by a partnership and linux users too. Nvidia proprietary drivers are good, and its support is not so bad, indeed, last drivers support until 800 series GPUs and some maxwell chip on 700 series as well. On the other side, intel and AMD support is rather miserable as for proprietary drivers. Moreover, the real problem on deprecation is Wayland where linux is moving. Indeed, if a driver works well, the interest to update is minimal. What Nvidia could set, it is to make free the modification of their proprietary drivers.

[NobodyXu]

If all software was free no enterprise would produce it, because pieces of software are also goods to sell. How doest thou think to sell a free item? How can be CUDA free if it is an item an enterprise realize and sell?

First of all, nobody is arguing that every software should be open source, but rather just GPU driver.
And open sourcing GPU driver doesn't hurt Nvidia at all, becuase you would need a Nvidia GPU to use the GPU driver anyway.

Second, open source != free, open source means you can obtain the source code and modify it freely, not that the companies cannot charge you for services based on it.

Ubuntu, Redhat, etc is able to profit from maintaining and releasing open source software.
One key service they provide for Linux kernel is live patching and extended LTS support, which is quite expensive and is sold on per-machine basis.

They also provide Software Defined Storage (Ceph), Fully managed OpenStack and Kubernates, all of them are open source.

Please check this out https://ubuntu.com/pricing/pro

Why an enterprise should realize free item? A company should just provide a technical and informative support.

That's exactly what Ubuntu and others are doing: The software itself is open source, but they charge you for the services.

A company has to preserve industrial secrets.

Sorry but Intel and AMD release their own GPU driver plus a few more, and it does not leak any IP secret.

They can put all the secret or IP into their firmware, which works just fine.

[oiaohm]

Nvidia made free Physx driver code. Does Phisic work on linux? I don't think. .

https://gameworksdocs.nvidia.com/PhysX/4.1/documentation/platformreadme/linux/readme_linux.html

As NobodyXu stated your argument MorrisS was basically garbage. NobodyXu missed that this point about Physx is also garbage but it also points out something really bad.
Physx under Linux is not driver and never has been. Physx is nothing more than a fancy front end to cuda. Under Linux you can use Physx on AMD hardware using hip because it userspace. Under windows you cannot use Physx with AMD because its a windows driver...

Of this gets better when you know that the early versions of Physx for windows were also a totally userspace solution. The only reason Physx is a driver under windows vendor locking to prevent AMD and Intel and other third parties providing their own CUDA replacements to have games work that use Physx.

MorrisS there are many miss alignments like this where Nvidia does one thing for Linux native programs then totally different thing for Windows users majority of the time this turns out to be make sure Windows users buy Nvidia hardware over competition by making tech only work on Nvidia hardware that does not have to be limited to a single vendors hardware. Physx is one of the examples of Nvidia doing intentional vendor lock-in.