in: https://bugzilla.mozilla.org/show_bug.cgi?id=974684 we updated nVidia drivers for only windows 8. Is it possible we could apply this to windows 7 as well to simplicity of management?

Current version on windpows 7: 314.07 Current version on windpows 8: 335.23 Who is the SME here from ateam or releng?

Probably isn't one, really - gfx developers are the only ones who might have any idea other than "well, we could try it and then try to make sure we run all the test suites that would be affected and see what happens."

Gfx devs as philor points out. Upgrading would probably require us to fix or disable tests and uplift the patches to all release branches before we could enable this. We would need one machine attached to a dev master to run all test suites (probably few times to prevent new intermittent oranges). I would only recommend this change if jrmuizel believes there is a good technological reason not to be on that version of the drivers.

I don't know of any reason to upgrade the drivers. It's more valuable to us for the machines to have different driver versions because it increases our testing matrix.

Q: how bug a hassle is it mgmt-wise? If Jeff and the gfx developers prefer it this way, that likely means maintaining different versions.

I know that one of the issues we keep running into is that the nvidia driver is now issuing popups that ask if you want to upgrade to a newer version of the driver. There is some chance that a test will accidentally click on this popup and upgrade the driver for that test run. Even if nothing clicks on the popup, I'm assuming that there may be some fallout from having it appear on the screen in the middle of a test to begin with. Running newer versions of the drivers is probably the easiest way to prevent this, and it means that the system that controls what versions get installed on each platform doesn't have to have rule based exceptions. Needinfoing Q to answer coop's question with more detail if he has any.

Management issues aren't that bad. It is mostly just simplicity in management of installer files. There are subtle differences in how installer versions allow the install process to be customized. I am working to remove certain components during install to avoid the aggressive driver upgrade push from NVidia. It is also very likely windows 10 will need another new version of the drivers then we are managing 3 installs. It was more a quick request to see if I could remove the overhead while I have the patient open.

If possible, can the drivers be upgraded to the latest 358.xx drivers if the test slaves are running Nvidia D3D11 Fermi/Kepler gen GPUs? The latest drivers support the latest OpenGL ES 3.2 extensions. https://developer.nvidia.com/opengl-driver lists the new OpenGL extensions supported. http://www.nvidia.com/download/driverResults.aspx/92212/en-us http://www.nvidia.com/download/driverResults.aspx/92218/en-us 358.50 WHQL is the newest driver, but I think if possible, wait for the next 358.xx WHQL driver release. I'm not sure what GPUs the test slaves run, if they are running older Tesla D3D10 gen hardware(GeForce 8/9/100/200/300 series), Nvidia no longer support Tesla gen hardware in the latest mainline driver, only the legacy 340.xx driver branch supports Tesla gen now. http://nvidia.custhelp.com/app/answers/detail/a_id/3473 http://www.nvidia.com/download/driverResults.aspx/89879/en-us http://www.nvidia.com/download/driverResults.aspx/89883/en-us

I'd like to see these upgraded too! I filed bug 1315871, but I'll mark it as a duplicate of this one.

I am going to close this as won't fix. Since we are going to move away from the IX hardware, and the nvidia drivers will be no longer used.

Reopening as nvidia drivers will be used in the cloud for the foreseeable future

What is the desired version? The one mentioned above seems to be specific to hardware.

I see two options here (and frankly, these same things should probably be under consideration for Win8 and Win10). 1) Figure out whatever Windows Update offers by default for this hardware and go with it. I'm going to go out on a limb that the vast majority of users don't go out of their way to update their video drivers, so should we? 2) We collect driver version telemetry data, so we could run an analysis to see which is the most common version for Win7 users and go with that. Probably not a ton of work for someone who knows their way around our data pipeline. In general, I think we should aim to have our CI be more representative of what our users are doing, so one of those options seems preferable vs. picking one out of a hat.