ati 5770 or still a nvidia

[LinuxID10T]

Yeah, see the sacrifice I've been doing for ATI??!?

I'm leaning towards getting an older but cheaper ATI or Nvidia (aka non-Fermi) card and do a wait and see after that (for Evergreen). I haven't even bought a NEW video card before!

The only desktop video cards I've ever owned was an ATI Radeon X300SE and my current one, a Nvidia EVGA 7950GT.

Ha! I am still running my Radeon X300SE. Unfortunately, I have to run the open source drivers with it. The OSS drivers have gotten better though.

[deanjo]

I think, with a modified heatsink, a GTX 480 would definitely cook an egg. FINALLY! Something actually useful that a Radeon HD 5xxx can't do!!!

You could do that with pretty much any GPU made in the last 7 years.

[LinuxID10T]

You could do that with pretty much any GPU made in the last 7 years.

Actually, I don't think so. Really only within the past few years. Remember the 90's? You didn't even need a heatsink back then...

[Adarion]

LOL! Jen-Hsun Huang at barbecue. Great pic.
I'm happy my ATI can't do this, though.

[ungoliant]

I recently bought a 5770 from asus. (EAH5770 CUcore), and i'm really pleased with how it performs on 3D games, and its lower power consumption compared to nvidia equivalents. I'm using catalyst 10.4 test, and its perfectly stable so far (2 weeks now), and fast with kde 4.4, but still needs the maximize-fix.

Drivers are getting better, but the need of the maximize patch and not beeing able to set up vsync on my desktop, is kinda annoying. I also have some glitches due to maximize patch.

It's up to you.

[Emanem]

I recently bought a 5770 from asus. (EAH5770 CUcore), and i'm really pleased with how it performs on 3D games, and its lower power consumption compared to nvidia equivalents. I'm using catalyst 10.4 test, and its perfectly stable so far (2 weeks now), and fast with kde 4.4, but still needs the maximize-fix.

Drivers are getting better, but the need of the maximize patch and not beeing able to set up vsync on my desktop, is kinda annoying. I also have some glitches due to maximize patch.

It's up to you.

What is this maximize issue?

Cheers,

[ungoliant]

What is this maximize issue?

Cheers,

It's a known issue either in xorg or ati drivers that causes windows to resize extremely slow. Even going fullscreen on a video player, takes 2 seconds.

Ati is working on this, their new Direct2D accel for 2D makes it pretty smooth, but is still in early development. Meanwhile, we have a patch for xorg that fixes this, at the cost of some graphical glitches in the very first instants that you open a window.

Google for it, there are tons of pages discussing this (I guess even here at phoronix)

[bridgman]

In a nutshell, there was a patch included in most distro releases for years (called something like "104-no backfill") which improved performance when running under a compositor. The patch eliminated what seemed to be a redundant copy of the screen image from video ram to system ram when performing certain window operations such as unmiminizing a window.

In the run-up to Ubuntu 9.04 it was noticed that with newest drivers (a) some visual artifacts started to appear when unminimizing windows, (b) that the artifacts seemed to be old video RAM contents, possibly from an earlier session or user, (c) eliminating the no-backfill patch eliminated the corruption, and (d) with the latest Intel and NVidia drivers the performance problems which had driven use of the patch in the first place no longer seemed to be present.

The artifacts were declared a security issue, the patch was removed, and 2D performance under a compositor became lousy for some hardware and driver combinations. The performance hit turned out to be a function of the acceleration API implemented by each driver. Running the open ATI drivers under XAA showed the problem while running the same drivers under EXA did not, for example. If the acceleration architecture happened to have hooks for (and HW-accelerate) the copy from video to system memory then performance stayed good without the no-backfill patch, but if the acceleration architecture did not include that operation then unminimizing windows became very slow since the memory copy was done on CPU and took a couple of seconds.

Intel had switched over to UXA (a variant of EXA) in the open drivers so did not show the delays and NVidia's proprietary acceleration architecture seemed to hook the download OK. ATI's binary driver was using XAA rather than a proprietary acceleration architecture and so got hit by the unminimize delays.

One of the ATI/AMD devs (Felix Keuhling) wrote a "backclear" patch which provided the performance benefits of the old no-backfill patch (the "back" part of the name) but eliminated the artifacts (the "clear" part of the name). Applying the patch upstream would have been an easy solution, but unfortunately at least one program was discovered which relied on the download (something in KDE) so instead the patch was offered on a distro-by-distro basis as a patched X server.

The general consensus today seems to be something along the lines of "yeah that download sucks and it really shouldn't be in the specs, but as long as it's hardware accelerated it doesn't take too long on most systems so what the heck". As ungoliant said, we are implementing a new acceleration architecture (not actually called Direct2D, btw, it just shares some acceleration routines and registry keys) so that the "downloading of screen info which is subsequently thrown away" will be equally fast with the fglrx driver. Fixing the spec to eliminate the download would have been a better solution but there didn't seem to be much support for that solution, since everyone had already decided "it must be an ATI bug".

Presumably once we have implemented a new acceleration architecture that accelerates the wasted download it will be easier to get support for fixing the problem properly, ie getting rid of the download completely, in which case our XAA implementation would run fine. A little irony if you like that sort of thing.

[Kano]

Well is a reaction time of about 1 year supposed to be good? That was similar to an obvious opengl error in ati drivers. It seems time runs about factor 10 slower for ati then for others.

[Panix]

Yeah, any chance of any benchmarks (with Ubuntu 10.04 when it's released, perhaps?) of the FOSS radeon driver v.s. the Fglrx driver (perhaps, 10.4 and 10.6) so three various drivers tested and compared?

It would be interesting to see an actual update presented and compare the two types of drivers tested on popular applications INCLUDING games, standard video playing, 3D graphics and then a power consumption and temperature measurements.

All of this with a Radeon HD 5770, of course. I would choose one with a reference cooler (standard/reference pcb) so that everything could be as 'standard' as possible. A lof of gamers/"ATI users" tend to prefer these anyway, it seems. Other than that, there are a few with 'customized' coolers but the standard reference cooler in many XFX and Sapphire cards might make it easier to test (and acquire? :-) ).

I bet a lot of people would find such benchmarks interesting.