Red Hat Developers Continue Working On OpenCL/Compute For Nouveau

[coder]

As for getting the signed firmware blobs, probably the reason for it still not happening is that the greater community is giving up on nouveau. That's not a good thing. I have hope that Red Hat may be able to convince NVIDIA to do the right thing here and release the blobs so the driver can be made useful.

Don't hold your breath.

Jen-Hsun Huang¹:

Nvidia is a Software Company

I think they're threatened by open source drivers. They lose too much control, like what APIs they support² and where/how customers can use their hardware³. Note that the last policy was enforced via the EULA on their proprietary drivers.

References:

1. https://forums.geforce.com/default/t...-future-fermi/
2. https://devtalk.nvidia.com/default/t...upport-opencl/
3. http://www.datacenterdynamics.com/co...25.fullarticle

[coder]

after installing an OS and having the screen stretched over your monitor at around 640x480 resolution is not good for end user experience.

nouveau could also server the niche of enabling users to run old NVIDIA hardware on new software stacks (eg latest Kernel). So if NVIDIA drop support for old hardware in their proprietary driver, it doesn't mean the user is forced to use old proprietary drivers which only support older software stacks.

These might be reasons for Nvidia to keep Nouveau on life support. The real test will be to see if they provide info to support reclocking on Maxwell, once they drop it from their proprietary drivers. That should be a while, though.

[torsionbar28]

It must be nice being a company making more money than any other, while getting other people to do stuff for them for free.

Is this really how you think it works? SMH

[starshipeleven]

You seem surprised by his setup. It's pretty normal.

If he is just browsing the internet, even GMAs from socket 775 motherboards are fine.
I have right here a HP Compaq dx2200 (desktop PC) that has a ATI Radeon Xpress 200 iGPU and it is still running KDE perfectly fine with OpenGL acceleration.

Most workstations have/had some kind of integrated graphics for office use while offering a dedicated GPU for heavier use or for using higher-than-FullHD resolutions.
If you are running Linux and the iGPU is supported (and you use a normal monitor) then you can usually remove the dedicated GPU.

If you wanted to run Windows instead it's more likely that the old NVIDIA card has a driver for Windows 10 than the iGPU.

[cybertraveler]

If he is just browsing the internet, even GMAs from socket 775 motherboards are fine.
I have right here a HP Compaq dx2200 (desktop PC) that has a ATI Radeon Xpress 200 iGPU and it is still running KDE perfectly fine with OpenGL acceleration.

Most workstations have/had some kind of integrated graphics for office use while offering a dedicated GPU for heavier use or for using higher-than-FullHD resolutions.
If you are running Linux and the iGPU is supported (and you use a normal monitor) then you can usually remove the dedicated GPU.

If you wanted to run Windows instead it's more likely that the old NVIDIA card has a driver for Windows 10 than the iGPU.

As I said: "or they will have worse performance using the integrated graphics over the discrete card with the nouveau drivers." I'm sure that there are lots of cases where nouveau driving a discrete GPU will out-perform the integrated GPU. A faster GPU can make browsing websites a smoother more pleasant experience. Furthermore, the discrete GPU may offer additional benefits like audio over HDMI, faster 2D & 3D acceleration (useful for some desktop effects and maybe even WebGL), faster video decoding (via hw acceleration), multi-monitor setups (integrated GPUs often have only 1 video out port) and kernel mode setting (for more aesthetically pleasing visuals while booting and more stable resolution switching when booted). Some details on the amazing work of nouveau here:

https://nouveau.freedesktop.org/wiki/FeatureMatrix/

So again: there's nothing odd about someone using an old machine with an old discrete NVIDIA GPU for basic tasks, even when there is an integrated GPU in the same system. The user may have to deal with more noise and higher power consumption if they keep the discrete GPU, but that's a trade off that each individual can figure out for themselves.

[cybertraveler]

... The real test will be to see if they provide info to support reclocking on Maxwell, once they drop it from their proprietary drivers. That should be a while, though.

I really hope they do. I hate seeing perfectly good hardware wasted due to stuff like this; especially when there are lots of users with this hardware and lots of bright minds ready and able to implement the reclocking when NVIDIA just dump the stuff they need.

[GI_Jack]

Michael I too was wondering this. The most likely explanation I came up with is this:

1. Premise: Red Hat customers are increasingly utilizing and relying upon GPU compute on the Red Hat platform. (I don't know if this is true; it's an assumption)
2. Premise: NVidia make very popular hardware for use on the Red Hat platform. (I also don't know how popular NVidia is versus AMD for GNU/Linux compute work)
3. Red Hat recognise that they are competing with Microsoft.
4. Red Hat recognise that Microsoft have a recorded history of utilizing their near-monopoly to make special deals with hardware manufacturers to promote their software & services at the expense of their competitors.
5. Red Hat recognise that if NVidia are tempted into signing one of the agreements mentioned above, then all of a sudden NVidia may stop supporting/maintaining their proprietary drivers on the Red Hat platform. In this situation, Red Hat would not legally or practically be able to maintain the drivers themselves and rapidly add support for new NVidia hardware. This would pose a large financial risk to their business.
6. Red Hat decide to mitigate this risk to a potentially growing and lucrative area of their business, by immediately starting work on building and maintaining robust Open Source compute drivers for NVidia hardware.
7. When these compute drivers are fast, well documented, well integrated and stable (complete with re-clocking support), Red Hat will likely promote them to their customers over the NVidia proprietary alternative. For customers that don't adopt them, they may simply maintain documentation for a fast migration path from proprietary drivers -> Open Source drivers, so that in the event NVidia / Microsoft do become a problem, they can quickly move their customers over to the Open Source drivers with low friction.

This theory I have is more plausible when you consider that Valve are almost exactly following this pattern in a different market to mitigate the threat of Microsoft using their near-monopoly position to take over the extremely lucrative PC game distribution and social platform market. Valve are investing in Open Source, GPU, driver development.

This is the best I've got

I seriously doubt that nouveau will ever be a serious for a high performance drivers. It will work well enough for Live CDs, especially the CentOS installer that need decent graphics support, and as the default install to ensure a working desktop for day to day functions.

As for FOSS friendly video card, at this point, go with AMD. Factory support is worth it.

[Danielsan]

It makes a lot of sense using Nouveau: to sell a product that is ready for production and don't need further drivers installation; to sell a product that you can promote as secure, blob binaries are unsafe; to sell assistance on Redhat products only.

[edwaleni]

Is this may be related to the new restrictive license of the GeForce drivers which forbid the use of these CG in data centers? This licence is only about the software part, not the hardware. And redhat has many customer in the data center market which use GeForce...
I may be wrong...

This is it. RedHat wants to expose OpenCL capabilities to processes running in containers without having to map the device via host kernel.

[antus]

Is this may be related to the new restrictive license of the GeForce drivers which forbid the use of these CG in data centers? This licence is only about the software part, not the hardware. And redhat has many customer in the data center market which use GeForce...
I may be wrong...

This would have to play a part. I believe this to be true, and if they can prove the compute works they can charge those customers to develop for the clocking problem. People who build small to medium scale super computers are the types who might be able to raise the funds to sponsor development effort for their particular GPU platform, but they would need some confidence that it can work which slow but functional compute might provide.

In reality its probably part all of the reasons in this thread, and if they are only funding a small number of developers to keep the driver moving then it probably is a good insurance policy with no down sides.