Which (high end) graphics card to buy?

[Kano]

In order to test a gfx card driver you can simply get a cheaper one from each company and compare yourself. I see no reason to buy a high end product first to just try differnt drivers. I am sure not 100% of the users will be fine with nvidia, most likely those who want to use lots of monitors with 1 or 2 cards. Some features for that use case are only enabled with Quadro cards with nvidia chipset which is not really nice. Cheap (best: passively cooled) cards do not really matter if you run em with binary or oss drivers for lots of use cases - well you will never be able to run very demanding demos/games with em. Those are really rare on Linux anyway, maybe Unigine Heaven/Oilrush will not run correctly on those cards. If you have got a few bucks left just try it if you never used cards from the "other" company - it would not hurt... Highend cards are a bit harder to justify to buy just for testing, therefore test cheaper ones first, problems with the drivers are often the same or at least very similar - maybe you get additional problems with powermanagement on oss drivers with high performance cards, but i would say you wont be using oss drivers for those cards for permanent use anyway - that would be more or less false spent money.

[pingufunkybeat]

I am convinced that deanjo is getting a nice paycheck for this. Nobody can spew so much FUD for entertainment alone.

The fact is that 99% of all games ever released on Linux work great with AMD OSS drivers. I dare you to find one that does not work. I run a GTX 560 at work (+blob) and an HD4550 (+oss drivers) at home and honestly, the Radeon offers a better desktop experience.

That's not to say that the nvidia blob is bad, it just suffers from the standard problems these blobs have. The nvidia blob certainly offers fast and complete OpenGL 4+, but struggles with simple things like randr and kernel upgrades. Recently, it has struggled with resizing in KDE, totally broke gvim, and before that it fried your graphics card. It's software and all software has some bugs. The Mesa + kernel drivers are the proper way to support hardware under Linux, and here AMD is well ahead. The OSS drivers are slower, but perfectly sufficient for most people and most tasks. Especially if you plan to game on Windows. They support most of OpenGL 3.0, are at 75% of Catalyst in many benchmarks (and optimisation work is not yet finished), have decent powersaving, excellent 2d and randr support, and are really stable. Not quite as fast as the blob, and no GPGPU atm, but everyone should decide on their own where their priorities lie.

In general, it would be a good idea to simply ignore anything deanjo says about OSS drivers, or any OSS project, because it tends to be 90% false information. It's nauseating how much of what he wrote in here is wrong.

[deanjo]

I am convinced that deanjo is getting a nice paycheck for this. Nobody can spew so much FUD for entertainment alone.

The fact is that 99% of all games ever released on Linux work great with AMD OSS drivers. I dare you to find one that does not work.

You mean like Solder of Fortune which gives you a screen full of garbage and NWN with random artifacts all over the place?

The nvidia blob certainly offers fast and complete OpenGL 4+, but struggles with simple things like randr and kernel upgrades. Recently, it has struggled with resizing in KDE, totally broke gvim, and before that it fried your graphics card.

Resizing was an issue on some cards which was fixed, gvim seemed to be isolated to you or your distro (I've shown you screen shots of it running fine in previous threads). The frying of cards was to a limited a few skus as well and was fixed promptly.

It's software and all software has some bugs. The Mesa + kernel drivers are the proper way to support hardware under Linux, and here AMD is well ahead.

Mesa is generic band aid solution. While it may be the "linux way" it is not necessarily the "proper" way of supporting the hardware itself given the fact that effectively neuters the cards capability on features that are present in the hardware.

The OSS drivers are slower, but perfectly sufficient for most people and most tasks.

But not sufficient enough to do what he is wanting to do if you read his planned uses.

They support most of OpenGL 3.0, are at 75% of Catalyst in many benchmarks (and optimisation work is not yet finished), have decent powersaving,

Still isn't up to par with the blobs, which is what I said.

excellent 2d and randr support,

Nvidias control panel works fine here for desktop use. 2D is perfectly fine here on the blobs as well.

and are really stable.

Not one nvidia blob related crash hear in years (excluding the early releases of vdpau support)

Not quite as fast as the blob, and no GPGPU atm,

Again more stuff that I said.

In general, it would be a good idea to simply ignore anything deanjo says about OSS drivers, or any OSS project, because it tends to be 90% false information. It's nauseating how much of what he wrote in here is wrong.

I'm wrong you say but most of what you said repeats the same things as what I say just with sugar coating to FOSS in a better light.

[mirv]

"did too"
"did not"
"did too"
"did not"
"did too"
"did not"

Business as usual in the phoronix threads. Now I'd get some popcorn and watch, but I don't like popcorn, so I'll just watch and laugh at everyone.

[pingufunkybeat]

Nah, dean, it was the regular FUD and blatantly incorrect claims, followed by backpedaling and creative interpretation of what you've said.

I've just installed KDE 4.6.4. No blacklists
I've just played all the HiB games. No problems.
I've finished Prey, Doom3, Quake4 and Amnesia on a 3-year old low-end passive card. In HD. No problems.
I've used powersaving for years. Works perfectly.
NWN works: http://kdekorte.blogspot.com/2011/03...-on-r600g.html
Soldier of Fortune has been certified Platinum since Mesa 7.6: http://wiki.x.org/wiki/RadeonProgram . If it doesn't work on your HW, file a bug
Id's Rage uses OpenGL 2.1. I don't know if it will run on Mesa, but my guess is that it will not take more than a week or two, even if it doesn't.

So many falsehoods in such a short thread, that's quite an achievement.

And then you have the misleading stuff like:

"you sacrifice A LOT in terms of performance" (=20% on new hardware, 0% on old)
"you sacrifice A LOT in terms of selection of software you can run" (what exactly?)
"you sacrifice A LOT in terms of power consumption" (no. just no.)
"NVIDA Excellent blob > featureless FOSS + higher power consumption + limited api support" (they both support 99% of what actual applications use. They BOTH have very limited API support)

Then you define "A LOT!!!!" to mean "5-15%" and rationalise it away. That's a pathetic abuse of the moderator status, you are using it to intentionally mislead people instead of trying to offer a level-headed explanation to users who are asking for facts.

Resizing was an issue on some cards which was fixed, gvim seemed to be isolated to you or your distro (I've shown you screen shots of it running fine in previous threads). The frying of cards was to a limited a few skus as well and was fixed promptly.

I guess these are not the droids I am looking for.

KMS, xrandr (= DE integration) and kernel integration are also not the droids I am looking for.

Only deanjo knows what we are all looking for.

[a7v-user]

2D on a Gtx 560 (Nvidia blob) is a bit slow, but I have a sinking feeling that's more an issue with the Gtx 560 and newer cards since the Gtx 260 was really fast. I guess 3D compositing desktops are the way to go whether we like it or not.

[deanjo]

"you sacrifice A LOT in terms of performance" (=20% on new hardware, 0% on old)

20% is huge and optimistic.

http://www.phoronix.com/scan.php?page=article&item=amd_r600g_june10&num=1

Lets look at the the common 1080 resolution using the best foss scores

Nexuiz ~160 fps vs ~40 fps = 25% the speed of blobs
Open Arena ~300 fps vs ~180 fps (catalyst is very CPU restrained here it looks like) 60% the speed of blobs
Padman ~250 fps vs ~100 fps 40% the speed of blobs
Warsow ~240 fps vs ~60 fps 25% the speed of blobs
Urban Terror ~106 fps vs ~80 fps 75% the speed of blobs
vdrift ~49 fps vs 82 fps 167% the speed of blobs
lightmark 427 fps vs 87 fps = 20% the speed of blobs

Overall that equates to about 41% the speed of blobs. That is a far cry from your claims.


That's is the difference that you often see migrating from one generation of card to the next. Old hardware is irrelevant as he is not buying old hardware, he is looking at purchasing new.

"you sacrifice A LOT in terms of power consumption" (no. just no.)

Last time I benched a Dell 15" with a 4570 ( back in March) there was a huge difference in consumption. Running a looped test of Nexiuz the Dell would run just under two hours using the catalyst drivers, on the foss drivers it would run for about 80 minutes. Not only was the power consumption increased but because the foss drivers were not delivering the same in performance this means that it was also suffering in efficiency.

"NVIDA Excellent blob > featureless FOSS + higher power consumption + limited api support" (they both support 99% of what actual applications use. They BOTH have very limited API support)

The foss drivers have no higher GL support, they are just starting some type of video decode acceleration on modern codecs and there is no gpgpu support. The blobs have these.

Then you define "A LOT!!!!" to mean "5-15%" and rationalise it away. That's a pathetic abuse of the moderator status, you are using it to intentionally mislead people instead of trying to offer a level-headed explanation to users who are asking for facts.

First of all I'm not abusing my status as mod, have I censored, banned, or even warned any one? Nope. Again I'm not sure what math you are using but the difference is far from just "5-15%".

I guess these are not the droids I am looking for.

KMS, xrandr (= DE integration) and kernel integration are also not the droids I am looking for.

Only deanjo knows what we are all looking for.

For you those are the things YOU ARE LOOKING FOR. It however is NOT anything stipulated by the OP. There are however items that he does stipulate that the FOSS drivers cannot do and you seem to ignore that fact.

[pingufunkybeat]

Haha, way to pick the ONE resolution where the FLOSS performance drops off significantly and ignore that the range is within 60-80% for all the others. You totally picked that one randomly, kudos

Do the same calculation with the common 1024x768 resolution and you end up with 80%. Average all of them and you'll be somewhere around 70%. And keep in mind that texture tiling and 2d tiling, as well as Hyper-Z are all being worked on and are still to come.

And this stopped being about what the OP wants as soon as you started arguing with droidhacker and discussing all sorts of things that the OP is not interested in.

In his case, the nvidia blob is likely the best option. This doesn't excuse the usual FUD you came up with. "Sacrifice A LOT in terms of available software" my ass.

[deanjo]

Haha, way to pick the ONE resolution where the FLOSS performance drops off significantly and ignore that the range is within 60-80% for all the others. You totally picked that one randomly, kudos

Do the same calculation with the common 1024x768 resolution and you end up with 80%. Average all of them and you'll be somewhere around 70%. And keep in mind that texture tiling and 2d tiling, as well as Hyper-Z are all being worked on and are still to come.

LOL, so you really think he is buying a high end card to run at decade old resolutions?

[pingufunkybeat]

You were giving genral statements, and you know it. You post the same thing in every single thread.