What I still don't understand is why hardware manufacturers like AMD and nVidia keep their drivers closed source. If they are making profit from the hardware only, why close the software? What is so secret about it that they have to keep it closed source? In fact, I think open sourcing it should help them reduce the cost since the community would be doing some of the work on their behalf. Am I missing something?
Here's an example: NVIDIA had a frame smoothing scheme in its SLI drivers for years to cover up the latency spikes inherent in SLI setups. About a year ago, TechReport, PCPerspective, and others started to look at Microstutter in SLI/CF rigs and discovered the issue was latency. They also found out AMD was far, far worse in this then NVIDIA, and AMD is still trying to fix its drivers to be competitive in this area.
So there is a commercial advantage NVIDIA gained by keeping the details of their drivers private.
Nouveau, despite the lack of documentation is thousand times better than the obfuscated nv driver Nvidia themselves provided.
The latter recently release some documentations to nouveau development realizing they are losing grip against competitions.
With times, Nvidia either cooperate with Nouveau developers of face irrelevance in the future.
Not on OSS drivers, which is what I was responding to.
Even then the statement might be wrong. Since Nvidia doesn't participate in developing open drivers we don't know how much time and knowledge they could contribute.
I might be the best chess player in the world, but nobody will know if I refuse to play chess.
Not that I want to downplay the radeon drivers or their developers, they are doing good work and the driver is good, but we don't know how good nouveau would be if they would get proper support from Nvidia.
So on the latest generations of Nvidia hardware the closed source Nvidia driver is in many cases faster on Linux than on Windows. A great point for Linux gaming and Nvidia, but it already shows that the often sub-par performance of AMD's closed drivers is not caused by Linux, but by the driver itself.
My two conclusions on that:
- Nvidia, give the nouveau developers more information, so that they can work out the issues with their drivers
- AMD, get the developers of your closed drivers the whip, so that they come up with an equally performant driver
It's funny, i came to the exact opposite conclusion looking at that recent article from Michael that compared fglrx and nvidia drivers across linux and windows.
I couldn't really see the difference between how fglrx was comparing to windows and what nvidia's driver was doing.
Both were mostly even in a lot of tests, and way behind in a lot of others. Maybe you give nvidia a slight edge, by 1 or 2 percent, but that's basically meaningless.
A far cry from what everyone always says about how great Nvidia's driver perform compared to windows and AMD.
On the other hand, it's virtually impossible to gain anything meaningful from the "benchmarks" michael is doing, so maybe on real tests NVidia would pull ahead. I just haven't seen it from what Michael is posting.
For examples, here are two representative tests.
Here, both the linux drivers are essentially even with their windows counterparts. AMD is about 1% slower on linux, while NVidia is about 1% faster. Which means it's essentially dead even and a wash. No victory by anyone.