Two years ago I was working on a large GL project, and basically the only workable drivers were either intel, or nvidia. With nvidia being far better off pretty much anything, from driver quality to debugging tools.
I'm using an HD4400 on my primary laptop, mostly to have a low-end system for performance tuning. While everybody says that intel should be one of the best supported OSS drivers (if not the only), the i915 driver constantly breaks with a million of little issues which are fixed in one release of the driver and broken again on the next. In the end, all these issues pop up in the wild, where I cannot even restrict users to a single driver version (because many times issues are chipset-specific).
Despite whatever I read here, over and over again, the nvidia driver has always been rock-solid for me, with intel being at least on par, and AMD being not even remotely comparable.
I had generally harder times with intel even, due to unpredictable gpu stalls. Even when it comes to power management, the latest i915 driver fails to come back from suspend.
I really want to switch to something higher quality and OSS, so I'm hoping AMD steps up the game.
I do not need another half-assed driver like i915.
That said, the ghost of ATI still haunts AMD. To buy from them next time I'll have to get a great price or they'll have to be way ahead of Nvidia which they were in 09 with the 5870.
But I do care about general performance and driver quality, especially as a developer. I love the fact that i915 is always the first to get support for the latest kernel features, it's probably going to have the first usable vulcan driver, YET every time I find dozes of minor (and not so minor) regressions, to the point that I'm simply tired.
The catalyst drivers were _never_ an option on linux. When it comes to development I cannot afford the system to fail on me twenty times a day just because I'm playing around.
I miss the days of mach64. I hope AMD delivers.
Are there no automated unit tests for the Intel driver?
I can also confirm the GPU stalling on Intel.
Even though other chips perform ok in the benchmarks and some AAA games, when it comes to less popular 3D apps/games or writing own graphics code NVIDIA just eats its competition for breakfast. We sometimes get better GL performance out of a 2 year old Tegra4 than out of everything that the competition offers today. And this is entirely down to drivers, not because the Tegra4 is such an advanced chip. Not even talking about the K1 or X1 here which are different levels entirely.
Edit: inserted a "not"
Brand loyalty is useless, because brands don't care about consumers.
It may only be their weak market share that inspires their generosity, but it's hooked me and certainly has me rooting for their continued existence.
I can only hope that Vulkan will change the game a little bit, so shader hacking won't affect perceived performance that much.
This generation the Fury and Fury X aren't quite as good as the 980 Ti (turns out Maxwell is really good) but aren't bad either.
I like AMD, but CUDA >>> OpenCL, so for any scientific computing Nvidia is the way to go. For gaming the 970, 980, and 980 Ti are too good this time around. AMD isn't far behind however, and next generation could easily flip things around.
Benchmarks are useless for most things because they are always fabricated or don't refer to real world scenarios. For example, Firefox JS engine is always very fast in benchmarks, but in the real world we know how it goes. Same with everything.
So the only thing that matters is your opinion.
People always assume only Gamer care about recent hardware.