Out of the ashes of my controversial thoughts on AMD in the post highlighting their new mini gaming box comes an editorial about AMD's graphics drivers.
I am no stranger to calling AMD out on bad drivers, I am not saying they are bad right now as I haven't personally used them for around a year now, but in the past they have been bad.
With AMD shaking up the industry with their Mantle API it does beg the question, where will AMD go with their drivers on Linux?
We have Mantle which they haven't even confirmed if they will support it on Linux yet. Mantle is the API that's supposed to give performance gains over traditional graphics drivers. Now that's a great idea in theory, but it has already been shot down somewhat by the fact that AMD themselves along with Intel and Nvidia all said you can unlock many multiples of performance gains like that with OpenGL.
We then have Catalyst their closed-source graphics drivers that probably have some sort of middle-ware stopping them from open-sourcing it. I doubt that is the only reason they probably have certain techniques they wouldn't want Nvidia seeing which is just business.
Then we come to their open-source drivers which according to reports from nearly everyone who has stated they have used them that they have improved dramatically. I've seen benchmarks over time from Phoronix which backs this this up too, sadly I can't link any as trying to find an AMD benchmark of old vs new drivers wasn't easy.
Will AMD drop their Catalyst drivers for general desktop Linux in favour of their open-source drivers on Linux and keep Catalyst for things like professional work stations?
It sure would make a lot of people happy, there are masses of upsides to having fully FOSS (Free and open-source software) graphics drivers like not having to install anything yourself, it just comes with the system and works right away. This mostly already happens though, all we need is better day 1 support of newer graphics cards in the open-source drivers which brings me to my next point.
What would be great to see is some sort of driver updater for open-source drivers like Intel have, that would be great for AMD open-source users. Especially if you use a distribution on a 6-month or longer release cycle, if you picked yourself up a shiny new AMD card that wasn't currently supported, but gained support in a release a month before you don't want to wait another couple months for the new distro, having that software to just update the open-source drivers without fiddling around manually with repositories or PPA's and the like would be fantastic.
I have to hand it to AMD (and Intel) for providing good open-source support, Nvidia is lacking heavily in that category.
So, what do you the readers think AMD will do in future? Article taken from GamingOnLinux.com.
I am no stranger to calling AMD out on bad drivers, I am not saying they are bad right now as I haven't personally used them for around a year now, but in the past they have been bad.
With AMD shaking up the industry with their Mantle API it does beg the question, where will AMD go with their drivers on Linux?
We have Mantle which they haven't even confirmed if they will support it on Linux yet. Mantle is the API that's supposed to give performance gains over traditional graphics drivers. Now that's a great idea in theory, but it has already been shot down somewhat by the fact that AMD themselves along with Intel and Nvidia all said you can unlock many multiples of performance gains like that with OpenGL.
We then have Catalyst their closed-source graphics drivers that probably have some sort of middle-ware stopping them from open-sourcing it. I doubt that is the only reason they probably have certain techniques they wouldn't want Nvidia seeing which is just business.
Then we come to their open-source drivers which according to reports from nearly everyone who has stated they have used them that they have improved dramatically. I've seen benchmarks over time from Phoronix which backs this this up too, sadly I can't link any as trying to find an AMD benchmark of old vs new drivers wasn't easy.
Will AMD drop their Catalyst drivers for general desktop Linux in favour of their open-source drivers on Linux and keep Catalyst for things like professional work stations?
It sure would make a lot of people happy, there are masses of upsides to having fully FOSS (Free and open-source software) graphics drivers like not having to install anything yourself, it just comes with the system and works right away. This mostly already happens though, all we need is better day 1 support of newer graphics cards in the open-source drivers which brings me to my next point.
What would be great to see is some sort of driver updater for open-source drivers like Intel have, that would be great for AMD open-source users. Especially if you use a distribution on a 6-month or longer release cycle, if you picked yourself up a shiny new AMD card that wasn't currently supported, but gained support in a release a month before you don't want to wait another couple months for the new distro, having that software to just update the open-source drivers without fiddling around manually with repositories or PPA's and the like would be fantastic.
I have to hand it to AMD (and Intel) for providing good open-source support, Nvidia is lacking heavily in that category.
So, what do you the readers think AMD will do in future? Article taken from GamingOnLinux.com.
0 Likes
Some you may have missed, popular articles from the last month:
The comments on this article are closed.
They have improved very much with their drivers, but...
In their latest driver for Linux, 14.4, they removed option in CCC to scale image manualy.
My GPU is connected with monitor via HDMI, and when I install their proprietary driver, image on monitor has black borders. I was able to solve it (with 14.3 and earlier drivers) by going to "Scaling Options" and putting the slider at 0% instead of around 7 or 8 where it was. Now that slider bar is gone and non of other "options" does work and so the driver is useless. That same driver for Windows still has that same slider bar... BTW, my GPU is AMD Radeon 7790 1Gb and I had only problems since I've bought it. Doesn't work very well with opensource drivers, and with proprietary one thing works, the other doesn't and so on.
Previous ATi 5750 worked fine with both drivers.
In their latest driver for Linux, 14.4, they removed option in CCC to scale image manualy.
My GPU is connected with monitor via HDMI, and when I install their proprietary driver, image on monitor has black borders. I was able to solve it (with 14.3 and earlier drivers) by going to "Scaling Options" and putting the slider at 0% instead of around 7 or 8 where it was. Now that slider bar is gone and non of other "options" does work and so the driver is useless. That same driver for Windows still has that same slider bar... BTW, my GPU is AMD Radeon 7790 1Gb and I had only problems since I've bought it. Doesn't work very well with opensource drivers, and with proprietary one thing works, the other doesn't and so on.
Previous ATi 5750 worked fine with both drivers.
0 Likes
The open source support of AMD is non constant. It comes and goes.
It seems they struggle to get enough competent people on board (5 open source devs for all GPUs...)
And AMD should drop their close source driver for good, intel is right on that one.
Regarding open source drivers, well opengl is a piece of sh**. When you have a look at the steam dev days, basically, if you want to code a 3D engine with no opengl overhead, you have to ditch 99% of the API, and most of the "efficient" API functions are recent ARB additions, not even in core (what a joke!).
Look at mesa, the open source implementation of opengl: it's an awful pile of sh**, and that is due to opengl inherent complexity (and that c++ which makes developers code object orientish brain f**kages, cf llvm for shader compile).
I do code an alternative low level linux radeonsi driver primarely for myself, then, I'm quite well aware of hw programming AMD GPUs: it's *DEAD simple*... but opengl takes the technical cost to insane levels.
The only reasonable way out of this mess, would be an opengl 5 with 99% of the API removed (which will look like mantle API a lot).
Khronos must eat the bullet for good.
Regarding APUs, they are very bad compared to discret GPUs: you cannot beat a GPU with specialized RAM and memory bus (that's why the PS4 has gddr mem, they decided to favor big time the GPU and kill CPU perfs, xboxone is all wrong...). But AMD GPUs are hell a lot faster than intel GPUs, then their APUs could become a far better alternative than intel ones.
It seems they struggle to get enough competent people on board (5 open source devs for all GPUs...)
And AMD should drop their close source driver for good, intel is right on that one.
Regarding open source drivers, well opengl is a piece of sh**. When you have a look at the steam dev days, basically, if you want to code a 3D engine with no opengl overhead, you have to ditch 99% of the API, and most of the "efficient" API functions are recent ARB additions, not even in core (what a joke!).
Look at mesa, the open source implementation of opengl: it's an awful pile of sh**, and that is due to opengl inherent complexity (and that c++ which makes developers code object orientish brain f**kages, cf llvm for shader compile).
I do code an alternative low level linux radeonsi driver primarely for myself, then, I'm quite well aware of hw programming AMD GPUs: it's *DEAD simple*... but opengl takes the technical cost to insane levels.
The only reasonable way out of this mess, would be an opengl 5 with 99% of the API removed (which will look like mantle API a lot).
Khronos must eat the bullet for good.
Regarding APUs, they are very bad compared to discret GPUs: you cannot beat a GPU with specialized RAM and memory bus (that's why the PS4 has gddr mem, they decided to favor big time the GPU and kill CPU perfs, xboxone is all wrong...). But AMD GPUs are hell a lot faster than intel GPUs, then their APUs could become a far better alternative than intel ones.
0 Likes
Quoting: XpanderQuoting: SkullyYes Linux is multithread and more cores win, but that doesn't apply to games yet. Most games are still using 1, sometimes 2 and rarely 4. So most games run *** due to poor single core performance and they also stink at floating point calculations due to having a shared fpu unit in each module.i dunno what you talking about really... check those videos:
Then take into account that the fx-8350 runs at 3.9ghz and boosts to 4.1, while the i7 is 3.6 with boost to 3.9. An I5 would be everybit as good for gaming as my i7 due to games using 4 or less cores.
eg. Serious sam 3 on Linux. In the performance options cpu performance, my fx8350 has to be set to lowest or framerate drops below 60 alot. My i7 can have it on ultra. same distro, same gpu
https://www.youtube.com/watch?v=eu8Sekdb-IE
https://www.youtube.com/watch?v=rIVGwj1_Qno
FPS difference in games is minimal.. and when it comes to running other things at the same time when gaming then AMD pulls ahead quite a lot compared to same priced CPUs from intel
also i dunno what issues you had then... i have SS3 on ultra and my 8320 at 4.2ghz with no such drops..
probably you had ondemand governor when those issues happened.. this is known to suck under linux with amd cpus.
edit: also theres __GL_THREADED_OPTIMIZATIONS=1 to make some singlethreaded games at least run rendering on second thread.. which boosts performance quite a bit on AMD CPUs that suck on single core stuff.. And when you play windows games in wine theres CSMT patchset that does that.. + luckly many games starting to use more than 2 cores lately.. spcialy thanks to new console generation that has the same AMD CPUs... like it or not but many games come from consoles to PC...
This is not completly correct Xpander as i said to you before when you use games with use 3 or more cores (mayor games at time) single thread performance is more important and AMD low IPC give loss perferformance
Compare your FX dont work because your cpu have many cores and stay OC, also i suggest you, you cant test with 2 cores at stock clock with similar INTEL cpu with same quantity of cores and frecuency and appear huge differences
Something more for make this test needs correct games with use no more from 3 cores, many titles of 2011 and before, this titles is mayor compared with recent titles with 4 core use
CSMT on machines with 4 core or less mainly depend higher IPC with higher frecuency (on most titles not scale up 3 cores, because this titles dont use more from 3 cores, mostly use 2)
On your machine is different because you use 1080p resolution and on this resolution VGA its more used and need in minor way cpu power
But many users use minor resolutions than 1080p and have lower machines
See your steam game list
http://steamcommunity.com/id/xpander69/games?tab=all
you can have some options for test for example (as i comment you, your machine allow select how cores as used):
alan wake (however this title have bug latern on recent wine)*
batman arkham asylum (this title is very interesting)*
bioshock 2 *
blades of time*
borderlands 2
burnout paradise (good title)*
company of heroes (good game)*
counter strike global offensive*
darksiders (good game)
dead space (good game)*
dishonored (good game)
fear 2 (good game)*
fear 3 (this game must be run on windowed mode for prevent mouse problems thanks gamersonlinux for tip)*
flatout 1 (good game)*
flatout 2 (good game)*
grand theft auto III*
grand theft auto vice city*
grans theft auto san andreas
hydrophobia*
mirrors edge (very good game)*
payday the heist (very good game)*
shadow warrior 2013 (very good game)*
skydrift*
skyrim*
lord of the rings war in north (good game)*
titan quest (good game)*
two worlds II (good game)*
*some titles i test on my channel
some of this titles from your list maybe usefull for test with before parameters mentionated
This give some example wiht this title: injustice gods among us
this with AMD A4 3300 Dual Core 2.5GHz
View video on youtube.com
and this with INTEL Pentium G3220 3.0GHz
View video on youtube.com
When i comment you in other topic with this video recompiling test for cpu world
View video on youtube.com
AMD core performance its too low compared with Intel at same frecuencies on same quantity of cores
This problem is for speeddevil cpu arquitecture used on FX and derivatives
However this situation its for wine
^_^
0 Likes
Patreon
PayPal
See more from me