Considering a new GPU soon. How's the 7700 series on Linux?

[Paul Frederick]

As I see it

k6(-2/3) were very bad compared pentium 2/3. but athlon and athlon xp where good cpus at that time. the problems for amd began with intel core2 and especially witht the core2 quad. intel just put 2 duals together to create one quad core, amd needed too long time to develop a "real" quad. phenom was 1 year to late and had one stupid tlb error, phenom ii fixed that, but then came intel i series with an additional performance boost. you can forget the new fx series, the speed/core even degraded because compared to 6 real cores you get a curious 8/4 core combo. now amd wants to sell cpu+gpu combinations, but intel does the same. amd has the "better" gpu part when you look at win drivers for mainstream, but i dont think the target user group really wants to play games which would barly run with intel gfx and 25% faster with amd gfx. All that matter is the price in that market segment, there raw power is not important for office/internet use.

Right now I'm getting a feeling we may all be rooting for our favorite dinosaur. The days of the big CPU may be drawing to a close, or perhaps the days of the small CPU being big enough might just be dawning. Keep your eye on ARM. It could be the wave of the future none here are seeing yet.

[Kano]

@bridgman

The legacy definition is the same as the amd one: every chip with is not supported by a current binary driver. Every other defintion would be useless.

[bridgman]

Does your definition of "current" still mean "within the last month" ? That doesn't really hold any more now that we've changed release scheduling.

I think it's fair to complain about deficiencies in the open source drivers for 5xx/rs6xx/rs740 and earlier since there are no plans to provide fglrx updates for those parts, but lumping 6xx/7xx in just confuses things IMO.

[Kano]

There is no other driver out there with support for those chips, did you see em? I do NOT count announcements, i only count REAL drivers. For newer chips there is also no driver with xserver 1.12 support on debian 64 bit. no support for kernel > 3.3, while 3.3 is already marked as EOL on kernel.org...

[bridgman]

Ahh, OK, so you're just saying "between driver releases I assume that you're lying and that the driver I have is the last one I will ever see". I can accept that.

It's also true that less frequent driver releases will make it more difficult to use the newest Linux distros with older hardware, but I think you will find things work out better than your worst-case assumptions.

[Paul Frederick]

I doubt it was you

Hi Paul;

Do you have any specific examples of this on the graphics side ? I've heard the same statement multiple times over the years but never been able to pin down an actual example. There are a lot of cases where random people on the internet have said things which we haven't done, but that's obviously not the same.



We have all the same issues. There's a 4th solution -- spend the time and effort to cut very carefully between what you can release and what you can't release, try to negotiate agreements to release the remaining bits, and/or find other technical solutions then release what you can. It's a lot of work, it was a tough sell internally, and it goes more slowly than anyone would like, but it works.

Unless you were working for ATI over 10 years ago now. That is when the legendary backpedal occurred. Like all of the other random people on the Internet I didn't take notes, but I don't forget either.

When I'm browsing for video cards I check box Nvidia, I don't even look at AMD's offerings. I tell everyone I am in contact with to buy Nvidia too. I may be just a random person to you but I assure you, my actions are far from random. What is random is ATI/AMD's policy shifts towards Linux. Actually it isn't random at all, just the BS spewing from AMD is random.

I tend to ignore it all anymore. All Intel and Nvidia here. Pretty consistent if you ask me.

[bridgman]

Thanks Paul. I was working for ATI 10 years ago (joined in 1998) but not in the Linux area. I did watch press releases even then, however, so I'm surprised I didn't see whatever you are talking about. Has anything like that happened more recently, say in the last 5 years when I was involved ?

The only thing I can think about from 10 years ago would be moving from limited open source driver support to providing binary drivers as part of acquiring FireGL, but I don't think that would count as "backpedaling" since the only thing we said about Linux before that IIRC is "there's an open source driver and here's a link to the project(s)". I believe we occasionally also funded some 3D development work ourselves, helped with the Weather Network-funded effort for R200, and provided some programming information to the developers for subsequent generations, but don't remember any press releases or other statements about that work.

If you could give me some additional clues that would be a real help.

EDIT -- I did some searching for press releases from 2001/2002 that might cover what you are hinting at - for your amusement only the first one I found was :

http://tech.slashdot.org/comments.pl?sid=45686&cid=4723848

[Kano]

Well the first fglrx driver i tested (and wrote a script to install) was

http://www2.ati.com/drivers/firegl/fglrx-glc22-4.3.0-3.2.8.i586.rpm

dated back 2003-10-09. My old script used rpm packages which have been converted using alien. It supported 21 releases, my current fglrx script for Kanotix Hellfire and Ubuntu supports 84 releases right now. No idea how many drivers have been there before, at that time i got my first ati radeon 9700 pro (for free of course). Did somebody else test (partly patch) over 100 _official_ fglrx releases? Btw. you can verify what i wrote using:

wget -qO- http://kanotix.com/files/install-radeon-debian.sh|grep ^[#]*VER= -c
wget -qO- http://kanotix.com/files/install-fglrx-debian.sh|grep ^[#]*VER= -c

Btw. if somebody wants to know how many nvidia drivers i tested:

wget -qO- http://kanotix.com/files/install-nvidia-debian.sh|grep ^[#]*VER= -c

[Dandel]

@bridgman:

I can pretty much be certain that the most appropriate method of examining the cards and using as a reasonable definition is the following...


deprecated/Outdated - These are the cards that will never see updates... This will pretty much Cover the "5xx and earlier, rs6xx/rs740 and earlier".

Legacy - These cards are older, and are not as likely to see an update, however there is a chance to see fixes on these devices. This covers the "6xx/7xx/rs7xx/rs8xx"

Current - Currently supported cards that are supported with monthly updates... This includes Evergreen and newer cards.


Given the list of cards here, there is one major fact that can be shown. It's impossible to mix and match these cards reasonably given the current driver status. The fglrx driver currently prevents developers from having a mixed configuration. I personally would love to be able to have a single system setup with a Radeon X1900GT (R580), a Radeon HD4550 ( RV710 ), and a Radeon HD 5770. I know that it is easy enough to find mother boards where this type of configuration can happen, however, the current drivers are lacking in the sense that all three are unable to be used at the same time.

Using this, I figure that most end users would actually expect linux to respond similarly to what would happen on windows. This would mean that all three cards can be used at the exact same time without failure ( Given what happens with windows 7 and Windows XP ).

Also, Please keep in mind That this is where on the given system all three displays have their own monitor attached. If anything, There should be improvements to allow the Open Source drivers and closed source drivers to coexist within the system. This would mean that improving the driver to use XRandR 1.5, and Possibly moving to instead make use of the kernel drm drivers ( This would mean deprecating all kernels released before kernel 3.1, since kernel 3.1 was when the symbols in drivers/gpu/drm where changed from EXPORT_SYMBOL_GPL to EXPORT_SYMBOL ). Also other kernel changes might make it more important to improve interoperability also...The 3.4 kernel release added DRM_prime to the DRM system. This in theory will allow much better support Crossfire, and CrossfireX.

[PsynoKhi0]

You have to pay more for the fastest intel cpus than those from amd, but they are faster. You pay more and get more. If you dont have got the money you could go with the celeron/pentium brands as well. I would say a 3ghz pentium will beat any amd cpu in terms of single/dual core speed which is important for many games. It is not my problem what cpu you can afford or not.

Completely missing the point.
The issue isn't whether intel has a performance advantage, the issue is how they got it back.
And surely anyone would enjoy getting the same performance for less money, hence the benefit of healthy competition.

Why the hell do I need to keep stating the obvious? Are most people that hopelessly narrow-minded in here?

ANYWAY:
Newfie, the HD7700 series is fine. There is no need for 9 pages of comments.
If you want to check the issues you may have with nvidia cards, try the nvnews message board.