NVIDIA Developer Talks Openly About Linux Support

[movieman]

Secondly, while patented code is a problem against open-sourcing drivers, how would it make it difficult to release programming specs? These specs need not contain either patented code or patented hardware design.

How do you control patented hardware without documentation on how to control it? Certainly there's 'dumb' hardware which is largely hidden inside the chip, but even much of that requires at least some kind of configuration or some kind of documentation for optimal means of using the hardware (e.g. even documenting that one instruction requires a wait of two clock cycles before you can access the results without a stall could tell an expert quite a bit about the internal operation of the hardware pipeline).

Providing clear examples is difficult without making obvious which company I was working for, but I'll try. Our chip had what I shall call a widget unit, which was patented and required several pages of programming documentation; it provided a more optimised interface between the chip and the host system than just relying on the host chipset to do the work. I don't believe any of our competitors ever implemented a widget unit, because Microsoft eventually built similar but slower functionality into Windows itself in software, but if they had then the pages on how to program it to talk to the host system efficiently would have been a pretty huge flag that they were violating our patent. It simply could not work without the driver configuring it properly.

Of course in that case the patent was obvious and publicised enough that none of our competitors could have violated it accidentally: but some patents are so general or obscure that, as I said, you'd need a whole gang of lawyers just to tell you that you're violating it and you could lose many millions of dollars if someone discovers that you are. Who's going to put their job on the line by taking that risk?

[mtippett]

Firstly, inside the company, the person/team that designs hardware is most likely different from the person that writes the usual proprietary drivers. How do these separate teams communicate? The hardware designing team must be documenting the interface which is then being used by the driver writing team. I agree that this documentation won't be sane enough and clear enough for releasing to the general public. But that is all there is to it: the cost involved in releasing programming documentation is just the cost of sanitizing these internal documents which is far less than the cost of writing documentation from scratch.

Meetings, co-design documents, phone calls, emails. It is communicated very actively, but not in a pretty programming manual. Co-development is probably the best way to describe it.

Clean, nicely bound documentation disappeared about a decade ago.

Secondly, while patented code is a problem against open-sourcing drivers, how would it make it difficult to release programming specs? These specs need not contain either patented code or patented hardware design.

A recent post covered this nicely . It all comes down to clues about patent infringment (which in most cases in non-intentional).

[movieman]

Firstly, inside the company, the person/team that designs hardware is most likely different from the person that writes the usual proprietary drivers. How do these separate teams communicate?

In our case, we had a pile of internal specs explaining what the chip was supposed to do and defining exactly how it worked in pseudo-code, and a simulator built by translating that pseudo-code as directly as possible into C++... and we could go talk to the designers if that wasn't enough. No company is going to want to release such precise documentation on exactly how their chips work.

But that is all there is to it: the cost involved in releasing programming documentation is just the cost of sanitizing these internal documents which is far less than the cost of writing documentation from scratch.

We did write external programming documents for some of our chips; particularly the ones sold into embedded markets where they typically wrote their own drivers. Since the only people who really understood how the chips worked were the designers, the VHDL programmers and the people writing the drivers, that basically meant that many of those people had to be taken off the work they were doing for a few days in order to rough out documentation that the technical writer could turn into a book that external developers could understand. You're talking about 100+ man-days lost at the 'front line' where the next chip is being prepared; even ignoring the $100,000+ cost in salaries for that work, delaying the next chip by a week or more could cost millions of dollars.

[mtippett]

Thanks to mtippett
...

But that is all there is to it: the cost involved in releasing programming documentation is just the cost of sanitizing these internal documents which is far less than the cost of writing documentation from scratch.

...

One thing I learnt a while ago is to look out for a few keywords that usually mean bad things.

JUST - Implies a trivialization of effort. As movieman points out above the JUST usually costs hundreds of thousands of dollars.

SOMEONE - Implies that it is thought to be important, but too amorphous or too low a priority to be handled by the person making the commoent. Basically it means that NO-ONE will do it.

I have become so sensitive to those words that I don't think I ever use them, and automatically translate them as

You JUST need to go from A to B -> You will spend a large amount of time, effort and money from A to B.

SOMEONE should stand up and do this -> I don't think it is my job, and it isn't important enough for me to do it anyway. I'll just complain and accept the status quo.

Not quite on thread, but I thought I'd just add one more post .

[V!NCENT]

Patents cover SW, HW, process, programming techniques, etc. *lots* of places.

I should get a patent on patent trolling and then troll every patent troll with my patent troll patent

PS: And then earn a loads of $$$$$$$ and give it partially back to companies that got patent trolled xD

[myxal]

I should get a patent on patent trolling and then troll every patent troll with my patent troll patent

*chuckle*
*tags the post "yodawg"*

[rah_]

Disappointing interview

This was a pretty disappointing interview. NVidia isn't thought very highly of in the free software and Linux communities. The questions that were asked in the original thread reflected that. Phoronix had an opportunity to address the issues head-on and pass the difficult questions on. Instead, it seems they pandered to NVidia. The questions were all soft-balls; sliced and diced, de-toothed versions of the questions the community wanted answered. Very disappointing indeed.

And on the content of the interview, quite a lot of interesting admissions of inactivity there. They say they can't be bothered to go through the internal documentation and edit it for release. OK, fine. Are they going to change their documentation process so that future documentation will be written to be suitable for release without editing? I'd imagine they would have said as much if they were. Which begs the question, why aren't they doing this? Alas, I don't think this question is one Phoronix will ask.

[niniendowarrior]

You seem to think this interview was conducted face-to-face.

I enjoyed this interview. Thanks Michael. I think the responses have been rather more open and honest than I expected.

[BhaKi]

Things are more clear now

Thanks to mtippett for explaining.

[rah_]

You seem to think this interview was conducted face-to-face.

What makes you say that?