This was my last NVIDIA hardware !

[axobeauvi]

Originally Posted by Gregor A.

Well, newer cards with good linux support are normally more powerful.

But why is my old GF 7950 much more powerful in KDE, Firefox, Flash, Compiz as the new 8800 GTS?

Ohh yes i forget, more powerful only with Windows drivers...

The good Linux support comes later.

And on the waiting time we lose money.

I hope this was easy enough for you to understand it ...

so ,you mean even your 2D desktop is slow or choppy?
have you tried to tweak any mobo options?
perhaps you're setting something that worked well on the old video card ,but hinders the new one.
are all the options default?

[cjcox]

Let's see... user buys Nvidia... obviously because he's a WINDOWS user. Why? Because there's NOTHING in Linux that could possibly take advantage of the 8800 GTS right now. And that's pretty well known. But hey... he didn't like his 7950 (what?). Anyhow... this is just stupid.

Person made up their mind.... thick head... go get yourself a nice onboard Intel. The only thing worse than Nvidia is ATI.... so what's the whole point of whining? Right now... if you want high speed accelerated 3D, Nvidia is the ONLY choice. If you don't need the "high speed" part, there are many options... all of which do not cost like the 8800 GTS does. And NONE of which come close to the performance of Nvidia.... oh well.

Does Nvidia need to get its act together Linux wise? Sure. Eventually, ATI drivers that perform well if not better will be developed and they'll be better supported than Nvidia.... but might be a couple of years off yet.

So... you want the best today... get an Nvidia. Want to use ATI tomorrow? Fine.... but tomorrow may be a long ways off.

[Gregor A.]

Originally Posted by cjcox

he didn't like his 7950 (what?). Anyhow... this is just stupid.

My old GF 7950 works fine now in my suns computer.
His old card was to bad for ETQW.
So I have buy a new GF 8800 GTS for my computer and swap my old card to his computer.

And I have seen the bad difference!

The OLD GF 7950 CARD was much faster in 2D !

And you can belief me, I test many optimizations, nothing brings me the old 2D performance.

[zbiggy]

Originally Posted by Gregor A.

Well, newer cards with good linux support are normally more powerful.

No. With every new Geforce family 2D block on chip is shrinking. Driver 2D acceleration shrinks too because there is no hardware to write for. So 2D is slower. Some day all 2D will be rewritten for 3D engine and 2D performance boost will happen. Or every desktop will go to OpenGL before.

Originally Posted by Gregor A.

But why is my old GF 7950 much more powerful in KDE, Firefox, Flash, Compiz as the new 8800 GTS?

Geforce8 is almost pure 3D architecture (like first 3dfx voodoos). So 2D acceleration or XvMC have to be programmed using 3D engine. This is hard and very big work to do. Needs time and HR to be finished. Now it is not present. When 3D hardware will start doing 2D it may be faster 100x times than dedicated 2D.

Originally Posted by Gregor A.

Ohh yes i forget, more powerful only with Windows drivers...

Because GPU vendors love Windows like old good wife.
Linux is young, wild, attractive, not demanding and tempting women who never have enough in another town where chip vendors occasionally jump to for a little bit of pleasure. Of course Billy like every wife is not happy when this happens.
This is of course ugly joke and is not real but explains the GPU vendors situation.

Originally Posted by Gregor A.

The good Linux support comes later.
And on the waiting time we lose money.

No. We not wait. We just do smart shopping. When Geforce8 is present and have poor speed and lacks bunch of features we buy Geforce FX,6,7 as bargains, second hand, rebates, auctions etc. We just use what Windows users and shops gives away. They loose on this - we earn. Nvidia also earns on this because if someone is selling Geforce7 he/she will buy Geforce8 ... or Radeon/Intel. Linux users are smart to take all the sweet: older hardware have all bugs fixed, there are known workarounds, drivers are optimized because companies learned about hw bugs from Windows driver programming. And price is very good for performance.

For Geforce8 for me the blockers are XvMC, 2D and XRender. I will buy Geforce8 when XvMC will appear and will be well working. Now I use Geforce 6150 and look for Geforce7.

For Creative X-FI I wait till promised OpenAL hardware Linux driver will be present. Now I'm happy with my second hand Audigy2 (4x times cheaper than the cheapest X-FI few years ago).

AMD Phenom looks also tempting but current gcc only supports sse3 so there will be no use of ssse3 or sse4/sse5 instructions so this buy is postponed.

Nvidia Nforce7 buy is postponed because of geforce8 integrated in it. If I buy discrete Geforce7 this chipset buy may come back to consideration if there will be driver support from Nvidia.

If you buy hardware it have to be better than previous buy. It is dumb to buy something because it is just new and loose.

By not buying Nvidia you can loose. I remember Radeon 9000 and 'no driver problem' I had. Paying money for hardware and realizing it only works using vesa driver is painful and learns smart buying.

[Linuxhippy]

Hi again,

Can't left this without a comment, sorry

Now it is not present. When 3D hardware will start doing 2D it may be faster 100x times than dedicated 2D.

Its already present - just look at the recent XRender improvements. However its no where near it already should be.
By the way why should 3D hardware be 100x faster than deticated 2D hardware - this is quite unlikely. 3D means a much more generalized render-engine, which means a more complex setup for each rendered primitive.
Engine setup (seting appropriate registers, telling the chip what to do with the input) is a critical part of 2D rendering, sometimes taking more time than the rendering itself.

AMD Phenom looks also tempting but current gcc only supports sse3 so there will be no use of ssse3 or sse4/sse5 instructions so this buy is postponed.

Who cares about that - 99% of all software you'll ever run on your computer is compiled for i386/i586 - and if you are extra lucky you find something dedicated to i686 (like recent firefox builds).
If you use gentoo, sorry - my fault

lg Clemens

[zbiggy]

Originally Posted by Linuxhippy

Hi again,

Can't left this without a comment, sorry

No apologizes needed. I visit this forum for reading comments to learn something. I'm not so lazy to throw comments and run away.

Originally Posted by Linuxhippy

Its already present - just look at the recent XRender improvements. However its no where near it already should be.

Notice that recent XRender improvements are present only in driver for hardware with shaders => 2.0 (Geforce FX,6,7,8). 2D is disappearing. "An X module for wrapped software rendering /usr/X11R6/lib/modules/libnvidia-wfb.so.x.y.z and optionally, /usr/X11R6/lib/modules/libwfb.so); this module is used by the X driver to perform software rendering on GeForce 8 series GPUs." - see Nvidia docs. I asked Nvidia about this new file on this forum. This file is designed for doing software 2D on Geforce8 because this is 3D only chip. Nouveau driver developers say that on Geforce8 everything has to be done via 3D engine because there is no other way. That is why there is big problem for Nouveau to provide support for Geforce8 - programming 3D engine is much more complicated than simple fixed function 2D.

Originally Posted by Linuxhippy

By the way why should 3D hardware be 100x faster than deticated 2D hardware - this is quite unlikely. 3D means a much more generalized render-engine, which means a more complex setup for each rendered primitive.
Engine setup (seting appropriate registers, telling the chip what to do with the input) is a critical part of 2D rendering, sometimes taking more time than the rendering itself.

Look at growing clock speed, faster memory clock and technology, wider buses, more speedy interfaces. Look at how present desktop look like and how it looked in the past. Before almost everything was drawn from lines and one colour fills. Now textures, multiple video textures, non rectangular windows - clipping and alpha, transformations (like saturation, colour->BW live conversion, YUV->RGB) are in heavy use. The old 2D core can not keep up with shader speed in such tasks - it is too much limited.
New GPUs are built for MS VISTA which is pure 3D so 2D is redundant. 2D block is cut to make space for more 3D units. Motherboards with PCI-E has Athlon64 as minimum due to architecture and socket. With such speed and sse2 instructions 2D is silently done using software fallback on CPU.

Originally Posted by Linuxhippy

Who cares about that - 99% of all software you'll ever run on your computer is compiled for i386/i586 - and if you are extra lucky you find something dedicated to i686 (like recent firefox builds).
If you use gentoo, sorry - my fault

I use Lunar since long time and contribute to it very often (especially Nvidia driver updates and build scripts). Source based distros list do not start and end on Gentoo -> http://distrowatch.com/source.php
And everything on my box is built with -msse3 -march=athlon64 without -mtune which means trying to use CPU lower than athlon64 will cause crash on boot or simple use due to "unknown instruction" CPU hard error.

I know this because I tried to move my hdd to athlonXP box for a moment to fix other Linux and I failed to boot.

But I like it very much. After rebuilding all system for athlon64 in 32bit mode it goes fast.

[AaronP]

Originally Posted by zbiggy

2D is disappearing. "An X module for wrapped software rendering /usr/X11R6/lib/modules/libnvidia-wfb.so.x.y.z and optionally, /usr/X11R6/lib/modules/libwfb.so); this module is used by the X driver to perform software rendering on GeForce 8 series GPUs." - see Nvidia docs. I asked Nvidia about this new file on this forum. This file is designed for doing software 2D on Geforce8 because this is 3D only chip. Nouveau driver developers say that on Geforce8 everything has to be done via 3D engine because there is no other way. That is why there is big problem for Nouveau to provide support for Geforce8 - programming 3D engine is much more complicated than simple fixed function 2D.

Sorry, I have to chime in here. Back in the day, a "graphics accelerator" accelerated a certain set of operations such as solid fills, copies (a.k.a. "blits"), pattern fills, lines, ROPs, etc. Anything more complicated was relegated to either software rendering or a 3D accelerator, which was often a completely separate card. All of that 2D functionality is still there in modern NVIDIA GPUs.

The real issue is that what is generally considered "2D" rendering has changed over the years. As you mention, modern applications prefer to use Porter-Duff compositing, antialiased text, etc.: things that the 2D engine just plain never could do. For these things, we need the 3D engine.

You mention libwfb. I should point out that libwfb for the GeForce 8 series and higher and libfb for older chips perform exactly the same function (and indeed are built from the exact same source code). libwfb just has some extra helper routines to deal with different memory layouts. Pretty much all X drivers use lib[w]fb in one form or another.

[Linuxhippy]

@AaronP:

1.) I am not really up-to-date when it comes to latest render improvements, but are antialiased trapezoids or trapezoids with alpha (w/wo AA) accalerated?
Is the hardware capable enough to meet the render specification for AA trapezoids, I've read at least for opengl there are some problems with the accuracy of the gpu. (thats why java in opengl mode sends down 8-bit alpha masks). What about gradients?
I expect that the use of such non-image functionalities will grow a lot, especially with cairo and the trend to SVG everywhere.

2.) If so much is accalerated, why do people still complain so much about 2D performance of the geforce 8000 series? Do you have any ideas where this could come from?

The answer to 2. could be secret, but 1. would be really interesting...

lg Clemens

Can I ask a simple, possibly silly question as i'm still more newbie than anything?

Exactly what 2d rendering issues are people having and with what apps? I'm not trying to be obtuse, I've just never experienced and issues with 2d from my 6800->7800GT SLI->7900GTX->8800 GTS across quite a few driver versions, and lately with Compiz.

Just for my own knowledge what are you seeing? Thanks.

[zbiggy]

Originally Posted by AaronP

Sorry, I have to chime in here. Back in the day, a "graphics accelerator" accelerated a certain set of operations such as solid fills, copies (a.k.a. "blits"), pattern fills, lines, ROPs, etc. Anything more complicated was relegated to either software rendering or a 3D accelerator, which was often a completely separate card. All of that 2D functionality is still there in modern NVIDIA GPUs.

The real issue is that what is generally considered "2D" rendering has changed over the years. As you mention, modern applications prefer to use Porter-Duff compositing, antialiased text, etc.: things that the 2D engine just plain never could do. For these things, we need the 3D engine.

You mention libwfb. I should point out that libwfb for the GeForce 8 series and higher and libfb for older chips perform exactly the same function (and indeed are built from the exact same source code). libwfb just has some extra helper routines to deal with different memory layouts. Pretty much all X drivers use lib[w]fb in one form or another.

Thank you @AaronP for clarification . We all here know only Geforce as black box, as benchmark results and from user experience so any light on architecture from 'inside' always help us understand what's going on for real in Geforce world. Seeing "software rendering" in Nvidia docs always raises bad user thinking and when nouveau developers say on www there is no YUV blitter and everything must be done via 3D engine for video stream may make false conclusion there is no 2D on GF8. After checking libwfb library topic I confirm this is part of xorg-server 1.4. Nvidia probably needed this file before it was added to xorg-server package so that is why we can find it in Nvidia driver package.

[AaronP]

@Linuxhippy:

Trapezoids are rendered in a two-step process by the X server. First, it allocates a scratch pixmap and draws the trapezoids into that. Then, it uses the result as the mask for a composite operation. The second step is accelerated. The first step is still done in software right now because the Render extension makes it very complicated to accelerate correctly. Modern GPUs ought to be able to do it, and I plan to look into what it would take soon, but I can't make any promises.

[Edit: I should clarify this. The second step is accelerated except in cases that would make a normal Composite operation fall back to software, such as using the disjoint and conjoint blending operations]

@zbiggy:

That's right. We build and ship libnvidia-wfb for use with X servers that are too old to have their own libwfb. libnvidia-wfb is built from a slightly modified version of the X server.

[Gregor A.]

@all:
OK, please excuse my outburst of emotions.
Please understand me, I fell me a little disappointed.

@NVIDIA:
If I buy a NVIDIA Card, i pay for good Linux support to.
Maybe the driver is good and the graphic concept of Linux is the problem.

I don't know. I am not professional enough to appraise it!

But I hope, NVIDIA solve this problems, maybe together with X.org developers.