Another Look At The Latest Nouveau Gallium3D Driver

[zoomblab]

Drivers made by people other than the hardware maker will never match the official ones, either in quality or speed. Even with hardware documentation available it will take years for development to reach a state of good-enough but by then your hardware is already obsolete.

If I go and spend some good cash on a brand new geforce / radeon card tomorrow I'll want to get the maximum juice out of the card on that same day. That will never happen with open source drivers.

[pingufunkybeat]

Drivers made by people other than the hardware maker will never match the official ones, either in quality or speed. Even with hardware documentation available it will take years for development to reach a state of good-enough but by then your hardware is already obsolete.

If I go and spend some good cash on a brand new geforce / radeon card tomorrow I'll want to get the maximum juice out of the card on that same day. That will never happen with open source drivers.

What you are saying is not completely wrong, but if you follow that logic, you have to stop using Linux altogether, because many drivers are not written by the hardware manufacturers.

Also, if you use Fedora or Debian, then your argument is hypocritical because these distributions are compiled by GCC and not the Intel compiler, which is much more optimised. If you buy a new i7 tomorrow, then you are right to demand more juice and 100% optimised peformance, but you don't do that, you ignore that and use GCC happily, and you use Linux happily, and you use ext3/4 happily, although there are more optimised closed-source alternatives out there.

The same goes for Windows, which is compiled using Visual Studio, which is good, but inferior to the Intel compiler.

I don't see why people are so religious about getting 5% or 10% more performance from a graphics card, unless they happen to be teenager game kiddiez. Most operating systems out there and most drivers are written by people other than the manufacturers, and usually don't deliver 100.00% performance, but somewhere near 90 or 95. And gain some maintainability and stability in the process. Same goes for graphics drivers. If they can match 75% of the performance of the blob (and it looks like they can), and bring all the advantages of open source, then damn right, I will use them.

If I want 100% optimizerzed performans, I'll buy a console.

[lolren]

same here...

monitor black , nouveau 8400gs, nv rocks

[marek]

Drivers made by people other than the hardware maker will never match the official ones, either in quality or speed. Even with hardware documentation available it will take years for development to reach a state of good-enough but by then your hardware is already obsolete.

We've recently proven that it is possible to match the speed of official drivers. It doesn't matter who develops the OSS drivers if a hardware manufacturer at least provides technical support and if there is enough manpower.

[crazycheese]

You pay AMD, they get more money, they invest more in R&D including Linux. What's hard to understand?

It is hard to understand how they figure out Im using it (and bought it) only due to opensource driver and I want my money going in direction "opensource". But no such way, up to this moment, exists. I pay AMD, yes. They develop windows drivers. Not opensource. Where is my money going?

Basically, there are important distributions out there (RedHat) who want all GPUs to have working support out of the box. Since Nvidia is boycotting this like the plague, they fund developers to reverse-engineer it.

I wonder why RH does this. They are no way "fun" or "hobby", they are making money with opensource solutions driving opensource forward. If they need tesla in corporate there is so much research to be done on foreign(as in not manufactured by them) hardware, same as everything else.

Last time I installed Linux Mint on my GF parents PC it refused to boot due to KMS in NV driver, which I disabled for VESA and all went fine(including installing proprietary for 8300 IGP, which is more than enough to run almost all titles).

So why do they put money there, it wont be functional and it wont be complete. For fun?

[crazycheese]

We've recently proven that it is possible to match the speed of official drivers. It doesn't matter who develops the OSS drivers if a hardware manufacturer at least provides technical support and if there is enough manpower.

And the manufacturer STANDS behind this. Provides access to hardware simulation, even via NDA. Checks for actual real world usage and demand (&tendencies). Provides tight work with hardware designers themself. Prints in large letters that this OS is official supported - on the box. Four developers is GREAT. Really. But it is by far not enough for efficient use of hardware.

[DeepDayze]

Basically, there are important distributions out there (RedHat) who want all GPUs to have working support out of the box. Since Nvidia is boycotting this like the plague, they fund developers to reverse-engineer it.

I don't think nVidia is providing ANY funding OR docs to devs at all, not even under NDA AFAIK...so you must mean Red Hat most likely funds the nouveau devs.

[Remco]

It is hard to understand how they figure out Im using it (and bought it) only due to opensource driver and I want my money going in direction "opensource". But no such way, up to this moment, exists. I pay AMD, yes. They develop windows drivers. Not opensource. Where is my money going?

Your money is going into lawyers who are clearing hardware documentation for release. And your money is going into developers who are writing driver code. Say AMD spends 0.1% of their budget on open source Linux drivers. That 0.1% is going to grow in absolute terms when AMD earns more money. They probably don't know how exactly how many customers are using their cards with an open driver, but they must have estimates.

[bridgman]

And the manufacturer STANDS behind this.

I think so, depending on exactly what you mean by this.

Provides access to hardware simulation, even via NDA.

Yep, that's how we got the r600 3D driver working in the first place. The work was done on a hardware simulation of what became the HD4770 (rv740), before the product was even released.

Checks for actual real world usage and demand (&tendencies).

Yep, unless you mean going out and asking every individual user which OSes and drivers they are running. The only reliable way to do that is with a mandatory reporting mechanism that sends user information back to AMD for all OSes, which would be extremely unpopular.

Voluntary reporting is not particularly useful since you have no way to determine whether the group that *did* respond is representative of the entire user base, or whether they have "self-selected" based on particular interests or preferences.

Provides tight work with hardware designers themself.

Yep. If we didn't do that the whole effort would move a lot more slowly.

Prints in large letters that this OS is official supported - on the box.

We don't choose which OSes the manufacturers choose to support, and right now most (all ?) board mfgs choose to only support some or all of the Windows OSes. Hopefully this will change with time.

Four developers is GREAT. Really. But it is by far not enough for efficient use of hardware.

Perhaps, but there's also the proprietary driver which shares code across multiple OSes in order to provide efficient use of the hardware, at least on the 3D side.

[bridgman]

When I respond to posts like this I'm never sure how much to repeat from previous posts, but I think one thing is worth repeating.

We are neither claiming to nor attempting to provide enough development resources to fully implement the open source drivers ourselves. That was not the plan in the past, is not the plan today, and I don't expect it to be the plan in the future.

We *are*, however, trying to make sure that we get enough information, support and sample code out to the driver development community to let the community make good use of the hardware (other than specific blocks like UVD, which we said up front would probably not be exposed), and I believe we are doing that today.

In some cases that means we write the initial driver code for new hardware enablement; in others we provide documentation. Over time, we try to provide both (typically working code first, then documentation later, since we need to write the code in order to know what to put in the documentation).