• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

[O/C] ATI Driver Installation in Linux

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
4770, 90 fps in OpenArena and UrbanTerror with opensource driver. Near 300/120 fps with Catalyst.

AMD proprietary lacks video accel, Xorg/kernel range is smaller, hardware(card) range coverage is way shorter - everything else is similar to nvidia blob(with notable exception of multimonitor which is lot better than nvidia).

On the other hand, nvidia opensource is similar to amd blob from 2000.
 
I don't get why ATi doesn't just work on OSS drivers so then they only have to support a single distro and kernel, and the community could easily make it work on their own distro/kernel.
Bugs would be resolved MUCH faster too.
 
They would potentially betray too much about how the cards work, perhaps?
 
I don't get why ATi doesn't just work on OSS drivers so then they only have to support a single distro and kernel, and the community could easily make it work on their own distro/kernel.
Bugs would be resolved MUCH faster too.

I think, I understand this whole heap fairly well. The reason why driver can not be opensource by nvidia, sorted by highest blockers first, is combination of the following:
- NDA on video decoding (from decoder to HDMI) for "trusted" HDCP path
- Licensing of IP for software algorithms from 3rd parties used within driver. As example are S3TC and various others.
- Unified driver architecture, meaning one could disassemble their windows implementation as well.
- Even knowing the underlying hardware language does not mean there will be a driver. Drivers are very complex and require high amount of work. This means high solds and the advancements should also be introduced at acceptable to market/hardware revisions rate.

With Nvidia you have good GPGPU support (OpenMP is copy-paste work from CUDA, where HD4xxx hardware has only one gpgpu pipeline, vs required two by OpenMP, which requires software emulation, slowing result a lot. Evergreen HD5xxx+ are not affected.).

With Nvidia you have large support window - they support even very old riva tnt2 cards by linking old driver to knew kernel/gcc/xorg ABI/API multiversions. The driver itself is not developed though, but you can use the old acceleration code which utilizes your hardware to the max.

You have VDPAU - which is video decoding - on par with that of windows. That means graphic engine supports 100% acceleration in linux. AMD REFUSES to provide any support for UVD engine ever, even in proprietary fglrx driver, which means there are no video accel on linux.

You have CUDA and OpenMP, as freeware. Only some outgoing developments by 3rd party university in Germany have developed some code to use GPGPU/OpenMP on AMD FGLRX driver.

You have driver which was worked on by windows team as well, which results in card achiving similar performance as in windows. If you head to phoronix.com and see nvidia benchmarks, you will note nvidia card have identic to windows performance(with few exceptions, mainly quake3 engine based games favoring windows kernel itself more, see unigine tests for example).

The hardware is very fast supported, for example 560Ti support was already in beta drivers.

That means, that although the driver is proprietary and forces you to use kernels/gcc/xorg versions that nvidia has headers(it is still much bigger than of amd fglrx), you have your hardware card supported for next 10 years at least.

However, if you look at situation with Optimus, you see the disadvantages - nvidia refuses to support and you cannot do anything. No question, Optimus "we have no plans" delivered huge blow to nvidia market segment in linux.

If you look an AMD, their fglrx was almost always workstation driver for FireGL cards. Not for normal home linux. It has always crashed, bugged the software, didn't support much hardware, older hardware was excluded very soon, newer hardware was supported only in one year. They have a big team which works with corporate customers on FireGL cards and only specific linux programs. Everything else is out of scope. Windows team is totally independent. AMD cards were only good for windows(where they still compete with nvidia geforce) and on workstation market(where they compete with nvidia quadro) - and nothing else.

The new situation wtih AMD however changed a bit.

AMD released programming guides for GPU themself, but the GPU are extreme complex things and driver quality(means amount of work) plays huge role in card end performance. You also have patents even in OpenGL(so much for "open") and you have new hardware emerging - which means development should succeed even faster, barely anyone want good opensource driver in TWENTY years for TWENTY year old card. Which means this code is only useful for small bug resolvents; for students interested in how hardware works; or for concurrence >:) So, the AMD opensource policy is really nothing for usable hardware or using linux as operating system, not some joke or hobby.

So unless opensource driver has BIG attention from company ITSELF there is no hope in it as production quality driver,and not a "hobby".

The FGLRX driver was renamed to Catalyst Linux, but I barely see anything change. No video accel, still breakages, still shorter than nvidia hardware support and gcc/kernel/xorg support, big driver size, SLOWER than windows opengl performance. Several bugs were corrected, new cards are sooner supported, nothing else.

And when I try to draw attention to AMD developers that they really should either do good proprietary(where they will still end up in same situation as nvidia, without benefits), or do good opensource(which will really put really good kick to nvidia), Im answered linux has very small segment. However when I mention they could easily implement backtrace, some register linux program based only on card Serial Number to get what kind of card people buy for linux opensource, roughly how much each has invested in opensource driver and when it happend - they just laugh, ignore or play deaf. And then come Apple or Microsoft trolls and call me a troll, etc.

My personal opinion stays same - you should not believe any "plans" or "promises", you should believe in market and buy (gpu) product which you support, use what you respect, pay attention to people that respect you.

I have exchanged 9800gt for HD4770 (not HD5770 because there were no opensource driver). In near two years later I've got this: low development speed, company that is NOT interested you buy their product, support in driver for 15% of GPU features, new product support (HD69x0) is lagging exactly at same rate as HD5xxx lagged.

They hire 2-3 developers and are happy. Redhat and Novell hire each one developer and guess what they do? Right - developer driver good for server(IGP).

As much as I love ATI (I still have their first Radeon 64 DDR around), their policy is just anything but acceptable.
 
Back