AMD Radeon HD 5750/5770

[bridgman]

Well lets put it this way, because of that fiasco, we had to replace 2 computer labs at the community college that had RS690/SB600 that were running linux. They were "useless pieces of hardware".

With some people I wonder, but I'm 100% sure you know better than that

You know that you could have kept the machines running either by switching distros or by doing some manual patch management, but that was a pain in the butt so someone *declared* the machines to be "useless pieces of hardware". That made it possible to replace them and not have to take on the extra admin work (which may have been the right decision, I don't know the specifics).

Needless to say, when we put out the bid to tender specified no "AMD chipsets.".

Yep, it's unfortunate but that probably had to be done, even though you and probably a bunch of others knew that statistically the next problem is just as likely to come from the vendor you chose instead.

The sad reality is that if you didn't mark the bid that way, and if something *had* gone wrong the next time, all the people who didn't know better would treat you like fools. If you do mark the tender that way and something still goes wrong, at least your decision is easy to defend. It sucks, and I wish you hadn't ended up in that situation, but it does happen despite everyone's best efforts.

[bridgman]

OK, this has been bugging me. What was the college doing with 690s on Linux that the open source driver couldn't handle by mid-late 2009 ?

Maybe Blender work ? The 690s don't have vertex shaders so they wouldn't normally be a first choice for CAD work...

Thanks

[Kano]

That begins with youtube, as flash is only "accellerated" with binary drivers currently. Well glsl is the requirement, maybe that works with drivers in the work. Interestingly absolutely no final driver was out for ati cards for the last 5 month. And even that was not 6.13 but only a 6.12.x stable minor update.

[rohcQaH]

and everyone knows that the primary reason computers exist in college is youtube. That's where the real work gets done

[deanjo]

OK, this has been bugging me. What was the college doing with 690s on Linux that the open source driver couldn't handle by mid-late 2009 ?

Maybe Blender work ? The 690s don't have vertex shaders so they wouldn't normally be a first choice for CAD work...

Thanks

Multi-use as they were used for many adult classes, from "Computers for the Paranoid", to introductory classes in programming which included introduction to openGL, video tutorials, even some basic 3d animation utilizing Maya PE, linux instruction, etc. The boards were all replaced with boards with nvidia IGP's that were more then capable of handling those tasks.

[bingel]

Ok, I think I've gathered enough information so now I can do a quick summary.

I've realized that if today you buy an ATI video card, use it under Linux means having to face the following problems:
- Proprietary drivers still quite unstable and lacking some features
- Lack of interest by ATI in the development of linux proprietary drivers which largely share the code with Windows drivers and then depend more than anything else on the vicissitudes of this operating system rather than those of xorg or linux-kernel.

Consequences:

- Support for these drivers unpredictable but in any case, limited to 3, 4 or max 5 years.

- Low speed in resolving bugs and in implementing missing features.

- Impossible to know when the missing features will be added: there is no clear plan.

- Need to install an LTS distro or a professional distro so that you can continue to use the no longer supported proprietary drivers for an extended period of time.
(but we all know that installing a distribution of that type will prevent us to get the latest software upgrades and we also know that in Linux world, where growth is constant and software is rapidly implemented with new features (often significant), to be able to upgrade to new releases could be very important).
- At the end of support, need to fall back on the open drivers that although they are more robust than proprietary drivers, they are still in development, lack of features and especially poors with regard to 3D graphics. This probably means a drastic drop in performance.
- Lack of forecasts on the state of the open driver between 2/3 years as in open source is obviously not possible to realize a long-range plan. Therefore currently we cannot know how much the 3D performances of these drivers will be good in future.
- From what I've understood, ATI does not want (at least currently) over-invest in the open project but rather prefers to wait and see how things will develop and how the community will respond. ATI is therefore limited to the following:

- It has made public the specifications of its processors

- It provides direct support to implement the open driver but just enough to make the product usable even if with poor performance.

I came to the conclusion that if currently you buy an ATI video card, you better do so only if the platform you are working on is MS Windows or a dual-boot system.

[mirv]

Your conclusions are wrong in quite a few (most) places...but again I'll leave it someone else to properly correct.

[rohcQaH]

Some of those points are true, but apply similary to any other linux GPU vendor out there.

Does any other vendor guarantee a minimum-time for driver support? Does any other vendor publish feature roadmaps with target dates? Does any other vendor guarantee that there'll be no performance regressions?

- Proprietary drivers still quite unstable

I think that's a myth. I've had these machines on linux:
- notebook ATI 9700
- desktop GT 6600
- desktop GT 7600
- notebook quadro nvs something
- desktop HD 5770

my own experience tells me that both drivers have bugs, with both drivers you can have regressions after upgrades and with both drivers you never know what the future holds.

My current notebook GPU is affected by the bumpgate problems. Some driver update soon after bumpgate went public started aggressively downclocking the GPU (or so it seems) and everything went noticably slower. (I'll leave it to the reader to figure out why that may have happened).
Compared to that, a ~20% performance hit from switching to OS drivers would have been a breeze.

I've also experiences several stability regressions with nvidia, always had to be careful when updating the drivers, often found myself going back to previous versions. Didn't have a single problem with fglrx in that area, yet. (To be fair, I've had less driver updates with fglrx so far, so it may not be conclusive.)

- It provides direct support to implement the open driver but just enough to make the product usable even if with poor performance.

no. ATI releases docs, pays a few people to work on the drivers and helps everyone who wants to contribute.

The performance problems will be fixed in time. ATI has a rough schedule for the OS drivers (non-public, but a lot can be derived by listening to bridgman) and the general idea is: first it's made to run at all. Then it's made to run stable. After that it's made to run fast.
Performance-improvements are on their to-do-list, they just aren't at the top.

What ATI does not do is to fix the linux gfx stack. New (and needed) technologies like KMS and G3D weren't created by ATI.
Then again, they don't fix windows bugs either, so why would anyone expect them to?

Your assumption is that ATI is only doing a half-assed job of delivering the OS drivers. I don't think so. They committed a few million bucks to the OS effort, doing all that for half a drivers would be a collossal waste of time and money. Projects of that size need to be done right or not at all, and judging by bridgman's posts, they do realize that.

The key-word here is that the OS drivers are unfinished. Work is being done, they will improve. AMD has had two years to work on the OS drivers from r300 to evergreen, that's 7 years worth of GPUs. The fglrx-team had more time for that. Expecting early miracles from the OS drivers isn't quite fair.


But I wish to note again that the OS drivers are already surpassing fglrx in some areas and in fact make a lot of people happy. Biggest missing features are related to power management (will be there soon), video decoding (will take a while) and heavy-duty 3D apps. Modesetting, 2D, xv etc is all there.
Your average joe with a composited desktop, a web browser and solitaire would be very happy with the OS drivers as they are now.


The safest long-term-bet right now is an evergreen-GPU. It's the first in a line of DX11-generations, thus will have longer fglrx-support than the last DX10-cards (HD 4xxx).
By the time fglrx-support is dropped, the OS drivers aren't "in early development" any more, but mature, feature-complete and performance-optimized. Maybe they won't reach fglrx-speeds for 3D-workloads (less manpower for optimizations), but close enough.

[mtippett]

I'll bite on some of the items.

- Support for these drivers unpredictable but in any case, limited to 3, 4 or max 5 years.

- At the end of support, need to fall back on the open drivers that although they are more robust than proprietary drivers, they are still in development, lack of features and especially poors with regard to 3D graphics. This probably means a drastic drop in performance.

How is this too different with NV? NV forks off the driver on a semi-regular basis. The Legacy drivers freeze the feature set, and get life support for the kernel and X versions only. Virtually no bug fixes, no performance increases, no feature increases.

With ATI you may trade 3D performance against he proprietary driver, but the X developers are keeping these legacy cards against new features in the X server. You will get the Gallium state-trackers, you will get randr improvements at the point of the release.

Intel feature and improvements drop off very quickly as the new features come out. i810 is keeping feature parity, but all the new work is focused on the new hardware.

When product is no longer up for sale, or is out of the R&D window for development, ALL vendor development slows down. At least with Intel and AMD, the community developers are able to pick up the slack.

Choose your poison.

[Panix]

I'm looking at prices for cards in my desired budget limit of $150 and if I stretch that or wait, I could get a HD 5770. But, I have read it's not as good as Nvidia for playing video and video editing. I guess that goes ditto for 3D and ATI's drivers develop way more slowly and are buggy in Linux.

The closest comparison for Nvidia is GTS 250 and 260 GTX but these are EOL so not sure I'd want to pay 'new' price and are harder to find. They are older generation, too, but at least drivers in Linux are up to date and I don't have to worry about patching or problems when the kernel updates.

One more complaint is the power saving in the binary driver would work and you'd need it with this card. At least, the HD 5770 doesn't run as hot and has less power consumption. Is ATI ever going to catch up to the comparative CUDA features and be competitive as a HTPC card or as good with video editing? If I was ready to buy in a month (when Fermi is introduced and current cards might be reduced in price), is there any chance for advancement in ATI drivers, seriously and honestly?

I want to start using Linux more but if I have to deal with constant patching and 'ATI issues' compared to 'standard binary install' of Nvidia drivers, I don't know which I would choose. But, since watching videos and 3D is important to me, I would probably consider a Nvidia card if it moves down in price enough. I'd pick the poison that is the least headache although I'd want to choose the ATI card because of the potential benefits but not if I have to google every time xorg or the kernel changes. Is that reasonable? I'm not a Linux/XORG expert and my 2nd computer, an old laptop, sucks so I don't like my desktop to 'be down' for too long a time.