Ati driver status?

Have the ati drivers improved over the months? Last time I read here people still placed nvidia above ati so just wondering if the tides have turned. If ati drivers are good then I’ll probably get 9500pro or 9700np as higher end cards are out of my price range. I’m skipping the gffx for obvious reasons and I don’t want to wait until nv35 unless I hear a confirmed shipping date which might sway me to wait a bit longer, not too long though. Anyone else in the same situation?

ATi has improved alot over the months, but I think that nVidia is still better in the OpenGL point of view.

I would agree. The only bug I’ve found in NVIDIA’s drivers for as long as I can remember involved using 150 MB of textures in a single frame on a 64 MB card.

I’ve seen quite a few problems in ATI’s implementation of ARB_vertex_program, to give a specific example. They’re getting better all the time, but I’m still a little too paranoid to buy one just yet.

– Tom

Thanks for the replies so far. Ok, how about this: on a scale from 1 to 10, one being poor and ten being great what score would you give the ati drivers?

My personnal evaluation:

  • Nvidia: 9
  • ATI: 7
  • Matrox: 3

Y.

I’d also like to chime in. I’ve been using a 9700 Pro at work for about 3 months now and have found about 3 bugs. In contrast, that’s about the same number of bugs I ran into in NVIDIA’s drivers over a 3 year period spanning 3 generations of hardware…and I haven’t encountered any in the past year.

One interesting difference in the consequences of the bugs - ATI’s OpenGL bugs have a tendency to lock the entire system, requiring a reboot. This makes them very time consuming to find. I have never had a system lock or bluescreen from NVIDIA’s drivers.

You may also want to know that there are still ATI driver issues that plague certain popular games, such as CounterStrike, and some people experience severe studdering. My home system stutters really bad in 3dmark2k3, but I don’t really like that program anyway so I don’t care.

So, after a large amount of programming on both, I would rate the drivers like this:

NVIDIA 9.5 (fewer bugs than I’ve ever seen with any software, consistent improvement too)
ATI 6 (It’s really annoying to crash my system when I’m trying to debug, or break things that used to work)

However, right now, I’d rate the top of the line hardware like this:

NVIDIA 7 (lots of shader instructions)
ATI 9 (more general floating point textures, great AA/AF performance, quiet cooling solution, availability)

So choose wisely for your needs

– Zeno

Thanks Zeno and others. Three bugs with ati? Are they major showstoppers? Any workaround for them? How does ati respond to bug reports? I switched from d3d9 to gl and I got used to hard crashing my gl apps. D3d is very forgiving in this respect but I find gl better overall. I am a bit surprised to find out that both ati and nvidia share so many EXT extensions, I did a compare and contrast and wow! I was expecting lot less commonality. This is great, with d3d I never knew who supports what but in gl this is spelled out for me.

There used to be less commonality, NVIDIA, ATI and others have been working hard at cooperating on some key extensions. In terms of commonality, OpenGL is experiencing a bit of a renaissance having emerged from a bit of a dark ages. We’ll see how far it goes, this metaphor has already gone far enough :-).

Three bugs with ati? Are they major showstoppers? Any workaround for them? How does ati respond to bug reports?

The bugs I found were:

  1. System freeze when binding a DXTC3 texture with size 2048x2048 (I submitted a demo app and ATI fixed it between Cat 3.0 and Cat 3.2) http://www.opengl.org/discussion_boards/ubb/Forum3/HTML/008993.html

  2. Issue with vertex programs and invariant position. I haven’t had time to narrow it down to “This is bug x and only happens when you do x and y” but you can read my thread about it here: http://www.opengl.org/discussion_boards/ubb/Forum3/HTML/009006.html

  3. Unknown bug that broke our WaterFX program between Cat 3.0 and 3.2. I’m assuming that at least one of those two driver revisions has a bug…this is another issue I haven’t had time to deal with. http://www.opengl.org/discussion_boards/ubb/Forum3/HTML/009011.html

I also get system freezes with ATI’s stuff from time to time when I make mistakes. Granted, these are my mistakes, but a driver should never crash the system.

If you have the time to make them an example app demonstrating the problem, I have found that both companies are very fast in fixing bugs. No complaints there.

– Zeno

Thank you Zeno for that detailed explanation and links, much appreciated Your waterfx looks amazing!

I think Zeno is about right regarding drivers/hardware quality:

Drivers:
ATI: 7
NV: 9 - 9½

Hardware:
ATI: 9
NV: 8

ATI needs to make sure that bugs that are fixed, stay fixed. Generally, I won’t report the same bug/issue twice. My list includes,

(1) Texture corruption bug in 3.0 and 3.2 when using EXT_texture_rectangle/CTT ( looks like it was fixed in 3.1 and re-introduced in 3.2 ).

(2) Shared program objects still not possible.

I reported an occlusion query bug a while back, this was fixed in 3.2 - I’ll be very annoyed if it doesn’t stay fixed with the following driver releases.

Other than that, ATI is doing a good job with the drivers and their hardware is great.

PS:

ATI devrel: 10

Originally posted by PH:
[b]PS:

ATI devrel: 10 [/b]

You know, I’ve never actually got an email back from nVidia’s devrel… The only time I ever get anything back from nVidia is if I email someone directly.

I’m not really happy with Nvidia’s devrel site, with the exception of the new board store, it doesn’t really have anything useful except a few things. And I’m still waiting to hear from ATI, I registered about 3 weeks and still haven’t received a response.

I did emailed someone directly from nvidia and got fast feedback. I think that’s the way to go for any company.

Before you give NVIDIA bad marks for developer support, please email me. The reason I participate on this board is to help developers - both with NVIDIA-specific issues and OpenGL/graphics issues in general.

Thanks -
Cass

Cass,

We can’t give NVIDIA devrel any marks as there aren’t any driver bugs to report . Seriously, though, you and Matt have always been extremely helpful on OpenGL.org and the NVIDIA specific bugs/issues I’ve had were confirmed and fixed by posting here.
I’ve never e-mailed NVIDIA devrel, so I can’t say whether I would get a response or not.

I would actually prefer having discussions here, as others would benefit as well. I do miss some of the good discussions on OpenGL.org - interesting topics are, unfortunately, too far apart.

ATIs drivers were nothing but problems before my Radeon9500 died.

A couple that I had were:

  • Writing to the alpha buffer didnt work properly (some writes would work, others wouldnt). I submitted a demo of it, and it was fixed in Cat3.1.

  • Severe texture corruption. Occurred in the same demo I submitted in the previous point, though wasnt fixed. I cant test Cat3.2 as my Radeon9500 crapped out… still waiting on ATI devrel to ship my ordered 9700pro… been waiting around 3 weeks now.

  • ARB_v_p is very unstable on the radeons… to the point that it takes the entire system down.

  • The vga signal from the card to the monitor died for no apparent reason most days… sometimes several times per day. Had to reboot to get a picture back.

Problem with nVidia drivers (cards?) that Ive had:

  • GL_CLAMP behaved like GL_CLAMP_TO_EDGE

GL_CLAMP behaved like GL_CLAMP_TO_EDGE

I remember a discussion about that issue on this board a long time ago. If I recall correctly it was a bug a long time ago, but there were a lot of apps that used it (incorrectly) and so they left it. In recent driver versions, though, I’ve been seeing a checkbox under the OpenGL settings to enable conformant GL_CLAMP behavior.

– Zeno

[This message has been edited by Zeno (edited 04-20-2003).]

I remember those posts too. If I recall, the clamp mode causes the tex. filter to sample the border values as well but if you have zero border then it works like clamp to edge as I understand it. I also have that clamp compliance in the nvidia driver panel.