OSDev.org

obiwac wrote:Well then how do the major OSes (eg. Linux, Windows) make 3D acceleration drivers for nVidia graphics cards? I'm assuming that nVidia makes the drivers for them...

Yes, that's the benefit of being an OS with a significant amount of users, hardware vendors will provide the drivers for you.

JAAman wrote:

Sik wrote: Modern 3D hardware acceleration has nearly no documentation available which I think is a much bigger issue =P

wrong...

most modern 3D hardware acceleration is fully and completely documented
I think what you meant to say is that it is not in our wiki... which is unfortunately true

Unfortunately it's not true. PowerVR SGX540, GPU I have on two of my machines (one mips the second arm), has 2 pages of the "documentation" in one manual and 8 pages in the second (they are just more talkative).
I know you meant desktop GPUs.

There are no reasons in not choosing IPv6 for the implementation. Don't hesitate yourself and do it, it's really a big deal for the future. Where even electric bulbs will have Internet access (for jabbering with other bulbs on social media and posting photos of themselves). Personally, I am so far away from the network in my project, that I even can't say why I'd choose one or another.

JAAman wrote: yes, the ~3% of systems using nVidia are not documented

Where did you get 3% from? http://store.steampowered.com/hwsurvey

A survey on a site devoted to gaming determines that many computers have discrete graphics cards. Now there's a surprise!

But I would also be interested to know where the 3% figure comes from. I'd be very surprised if it is correct. As well as desktops a lot of laptops have NVidia graphics.

I suspect that the true figure is somewhere between that pulled out of thin air and that from a biased survey.

Hi,

iansjack wrote:A survey on a site devoted to gaming determines that many computers have discrete graphics cards. Now there's a surprise!

But I would also be interested to know where the 3% figure comes from. I'd be very surprised if it is correct. As well as desktops a lot of laptops have NVidia graphics.

I suspect that the true figure is somewhere between that pulled out of thin air and that from a biased survey.

The true figures depend on what type/s of systems you're looking at and what time period your looking at. Until these things are defined nobody can agree on any statistics.

For "all computers of any kind sold in the last 6 months" (including systems embedded in things like vehicles, microwave ovens, networking routers, etc; and smartphones, game consoles, desktop, small and large servers, ...) I wouldn't even know where to start, but I'd be tempted to suspect "there is no video at all" has the largest share (and I'm not sure if "nothing" counts as documented or undocumented ).

For "all laptop/desktop 80x86 PCs sold in the last ~12 months" it's not too hard to find statistics showing Intel's integrated video at about 70%, NVidia at around 16% and AMD at about 13%; where (due to Ryzen) it'd be reasonable to expect AMD's share to increase a little in the next few years. For "laptop/desktop 80x86 PCs sold in the last 10 years" I'd expect Intel's share to be lower (maybe 50%?).


Cheers,

Brendan

Hmm... if only nvidia can make their own drivers as they keep the documentation to themselfs, then why are there some opensource nvidia drivers (like Nouveau) not made by nvidia?

My OS implements both IPv4 and IPv6.

EDIT: My home ISP also provides IPv6, so I had plenty of testing ground. Not everyone has that luxury though...

Yeah, ISPs are the big problem =/

obiwac wrote:Hmm... if only nvidia can make their own drivers as they keep the documentation to themselfs, then why are there some opensource nvidia drivers (like Nouveau) not made by nvidia?

Reverse engineering (pretty much constantly looking at what the hardware does to take a guess of how it works), which is a complete pain in the @$$. Takes a long while, can get you in legal trouble depending how you do it (or if you just happen to get on the nerve of the wrong lawyer), and by general rule it should be left in as a last resort.

Linux actually has two drivers for NVidia cards. Neuveau, which is reverse engineered and open source but is very slow and doesn't support a lot of features, and the proprietary driver, made by NVidia which is high performance and absolutely necessary for using the advanced features on the Quadro cards.
The big problem with the proprietary driver is it has terrible integration with the DRI system, this is by design, to make it harder to reverse engineer, but it makes it really fragile.
For example if you update the kernel images with the driver installed, it just refuses to boot the xserver again until you reinstall the driver.

StudlyCaps wrote:Linux actually has two drivers for NVidia cards. Neuveau, which is reverse engineered and open source but is very slow and doesn't support a lot of features, and the proprietary driver, made by NVidia which is high performance and absolutely necessary for using the advanced features on the Quadro cards.
The big problem with the proprietary driver is it has terrible integration with the DRI system, this is by design, to make it harder to reverse engineer, but it makes it really fragile.
For example if you update the kernel images with the driver installed, it just refuses to boot the xserver again until you reinstall the driver.

nouveau actually works quite well with older cards.

I use nouveau, and it's actually not much slower than the driver that Nvidia provides. But I think that the fact that it's open source makes a great advantage.

I find it fine for desktop use, but it definitely can't match the performance of the proprietary driver. Both from personal experience, and I've seen benchmarks which support that. Not that it's bad, just that it isn't as good.

I have to support an application for my day job though which uses OpenGL quadbuffer, and that just doesn't exist unless you're using the proprietary driver.