NVIDIA's Linux Driver On Ubuntu Is Very Competitive With Windows 8

[Artemis3]

It is not clear to me.
did the nouveau driver reclocked the 9800gtx to the max before running benchmark or not?

If the 9800 GTX is anything like the 9800 GT, there is no reclocking to worry about (it always runs at full clock) ^^
Which might explain those better nouveau results with the 9800 vs the 460, which does reclock under load and its probably downclocked by default.

[Artemis3]

I simply don't understand why AMD can't deliver closed drivers that are almost as good as the Windows ones. What's so hard? What does NVIDIA do differently?

According to nvidia, the driver code base is common in the different platforms. Perhaps AMD has a completely separate code base for each?

[mendieta]

Click the minuscule boxy-box at the bottom left of the chart.

Ah, I see it, but clicking on it does nothing for me (Google Chrome). Anyways, thank you!

[Artemis3]

- Easier upgrades: Nvidia binary driver (a.k.a the "blob") supports the latest hardware at launch time. Something that neither nouveau nor the open source AMD driver do. We are still told to use the 6000 series more than one year after the 7000 series launch (and the 9000 series are coming soon), because RadeonSI is not ready yet and you are stuck with OpenGL 2.0 (an API which was released on September 7, 2004. That is nearly 10 years ago).

That is only because they don't disclose their hardware, if they did, it would be technically possible not only to offer support when the hardware is sold, but in anticipation, the way it occurs with intel CPUs for example.

Given that Nouveau (and to some extent Radeon) are reverse engineering efforts, its normal for them to lag behind. Even if you sent hardware samples to developers it still takes them a lot of time to figure 'em it out on their own.

Intel gpus on the other hand are getting better supported, because they are both spec open and Intel has put people to actually work on them.

The open drivers also can fall back to mesa for things the hardware can't do or they haven't still figured out how to make it do; that might explain those funny nouveau results being much faster than nvidia for some benchmarks; maybe some functions offloaded to the cpu ended completing faster there

[curaga]

- Easier to fix bugs: Easier for whom? Programmers know how to fix bugs in both open source and closed source drivers.

Show me how you fix a bug in a closed driver you have no source to.

And don't tell me that having the source allows you to fix bugs 10 years after the company that manufactured your gfx card has ceased its support, because you are not going to be using the same hardware.

Says who? I certainly use my hw until it dies.

We have also the issue of who (in the open source world) is supposed to fix the drivers for your 10 years old gfx card... YOU?

Yes me. It's not rocket science. I'm far from a lvl70 guru programmer even.

[mendieta]

Ah, I see it, but clicking on it does nothing for me (Google Chrome). Anyways, thank you!

Oddly enough, it works in Firefox, and not in Google Chrome!

[GreatEmerald]

Oddly enough, it works in Firefox, and not in Google Chrome!

Yeap, sure works in Firefox. I wonder if it works in Rekonq...

[DrYak]

According to nvidia, the driver code base is common in the different platforms. Perhaps AMD has a completely separate code base for each?

Nvidia drivers' are mostly just a recompile of the Windows.
Including down to the small details. They tend to do stuff "their own way".

A a consequence:
- some features are handled completely differently (multi display, for exemple, use their own way instead of being Xrandr from the begining like it is standard).
A a consequence, some features don't work very well or in a standart way in Linux.
(Multi-diplay can sometime break badly on Linux because the desktop environment doesn't expect the system to work that way).

- other features just plain don't work because the mecanism Nvidia relies on don't exist as-is in Linux. Often Linux provides a different way to do it, but Nvidia doesn't bother doing stuff the Linux way. As an exemple: sharing buffers between graphics cards. It functions in a completely different way in Linux (DMA-BUF) then in Windows. As a consequence, Optimus doesn't work on Linux - hence the whole "Fuck You" public announcement.

As an advantage:
- As the drivers are more or less the same, for basic functionality you more or less get the same performance between Windows and Linux as it's basically the same code running.

What you are testing in such a benchmark, isn't much the difference in performance between *the drivers* for Linux or Windows (there aren't much their basically the same).
But you are in fact testing the relative performance of both OS (Linux is usually at as light advantage here) and the overhead of the layer between the OS and Nvidia (Linux is at some disadvantage as this is basically a Windows drivers which was twisted to work on Linux).



An the AMD side:
- Catalyst share some code between windows and linux. But not everything.
- Most notably: the windows driver relies much more on the standard stack (its a DRI stack, it uses Xrandr for multi-display, etc.)

That requires some resource (mostly programmer's working hours) to adapt.
And as (currently) Linux is a much smaller market, they can't pour that much (from their limited resources) into this specific task.

Don't also forget that AMD collaborates a lot for the development of opensource drivers (whereas Nvidia doesn't give a fuck about Nouveau, with a few exceptions regarding Tegra).
This also needs resource (mostly legal department's working hours to grant clearance before public release of internal documentation, but they also contribute [paid-for] code to Radeon).
Well, I'm not complaining about this, it's actually great. Just keep in mind that this means that when AMD announce they helped Radeon on Linux, it might also mean "we didn't touch Catalyst at all, we only spent time reviewing docs for Radeon 9x00 serie and sending patches for radeonSI".



So in short :
- Nvidia's performance on linux is better, because the only concentrate all their (much abundant) resources into writing a single Windows driver and getting the same to work under linux (and to hell for the bits that can't translate easily between Linux and Windows).
- AMD spread all their (much smaller) resources among a binary Windows drivers, a separate port of that one on Linux but which actually uses different set of low-level interface to better play along with Linux, and last but not least also contributing to the opensource driver.


So they are spread

[YAFU]

I remember many years ago when I read from fanboys how great was ATI and its support for Open Source and why I should use an ATI. Meanwhile I saw how nVidia users could run fluids desktop effects with Beryl and AIGLX (install and enjoy) while I with my ATI I had to create odiousodious XGL session, having a very poor performance and full of crashes.
I wonder why AMD did not bet completely for OpenSource drivers and it still maintains Catalyst for GNU/Linux. Are the OpenSource drivers mainly designed for cards that AMD left unsupported? (leave unsupported ATI models that are not very old, something AMD does very often). Currently, Could have the best performance all ATI users only using radeon OpenSource driver? (all Steam games, for example).

[brosis]

I remember many years ago when I read from fanboys how great was ATI and its support for Open Source and why I should use an ATI. Meanwhile I saw how nVidia users could run fluids desktop effects with Beryl and AIGLX (install and enjoy) while I with my ATI I had to create odiousodious XGL session, having a very poor performance and full of crashes.
I wonder why AMD did not bet completely for OpenSource drivers and it still maintains Catalyst for GNU/Linux. Are the OpenSource drivers mainly designed for cards that AMD left unsupported? (leave unsupported ATI models that are not very old, something AMD does very often). Currently, Could have the best performance all ATI users only using radeon OpenSource driver? (all Steam games, for example).

Both AMD and Nvidia create proprietary driver,
for both the end customer is Quadro/FireGL, *not* desktop,
but Nvidia puts a lot more money and does a lot more testing - thats why usually nvidia driver works near everywhere and is good.

Still, there *is* no promise from Nvidia, that their Linux driver is for Geforce or that its for gaming. It *was* however very good financed and supported. But not *windows-like* good.

So, Nvidia never thought twice about disabling or removing a feature from Linux driver, if end-user uses Geforce instead of Quadro. Nvidia green and OpenGL extension crippling come to mind. No Optimus support, no SLI support. So, non-workstation Linux users were simply able to work nearly flawlessly with nvidia due to their awesome engineering team and good funding, even if they never were targeted as user group.

Also, remember that Nvidia strongly dislikes opensource and does this only due to workstation/Tesla market, and also both AMD and Nvidia workstation drivers have a lot of 3rd-party code, that required to be mixed up and thus makes opening up the driver impossible.

So, what AMD did, is to adapt and fund a universal graphics stack (Gallium), to work on it, to help MESA tremendously with OpenGL (together with Intel), to open some specifications and recently to reveal the code that really allows opensource driver to take full advantage of hardware. It is also a driver that is targeted at private customers, not workstation. Nothing this similar exists on Nvidia side.

To my knowledge, nouveau also grew up with some help from AMD opensource driver and documentation. Officially nvidia has only one closed driver. Ofc it took enormously long time to complete and even today is incomplete, but the progress is undeniable and their powerful hardware paired together with open driver offers them exceptional place on the shelf.

IMHO I would migrate to AMD if one cares about opensource. At least, it gained missing part - it has become quite usable and can justify hardware costs. Otherwise, nothing really prevents to stay with Nvidia if one is comfortable with it (I am less than so).