ATI: Open v. Closed Drivers

Since Stephen Shankland's article at CNET entitled New Linux look fuels old debate, we have been overwhelmed with requests to take a serious look at the frame-rate performance differences between the various open-source and proprietary contenders. Many of Stephen's thoughts had revolved around the recent push for Linux desktop eye candy through delivering 3D effects with such projects as AIGLX from the Fedora RenderingProject, and XGL. Some of his thoughts had also pertained to Fedora/Red Hat closing out proprietary kernel modules by shipping Fedora Core 5 with a flawed kernel. There also has been lots of open-source discussion since a disturbance that took place where the Free Software Foundation's Richard M Stallman was almost arrested at an event where an ATI programmer was speaking. Our first article on this topic, which will likely be the start of a series of examinations, is looking at the differences between the X.Org open-source ATI Radeon driver and that of ATI's official but proprietary fglrx display driver.

While some hardware manufacturers have been Linux-friendly with releasing the specifications for their product or pushing out open-source drivers that comply with the Free Software Foundation GPL, the GPU manufacturers have been largely resistant. The two names synonymous in this industry are of course ATI and NVIDIA. In the New Linux look fuels old debate article, there were quotes from both ATI's Matthew Tippett and NVIDIA's Andrew Fear, but what the proprietary drivers largely boil down to is protecting the respective company's intellectual property. Various driver components from each of the green and red sides are open-source, such as NVIDIA's nvidia-xconfig and ATI's fireglcontrolpanel. While Intel's integrated, graphics components are not largely competitive now for gamers and enthusiasts, they on the contrary do sport open-source Linux display drivers. Intel does claim they are working on "bringing out some interesting new stuff", but until the time comes we will remain mute.

Last year we at Phoronix had delivered very intriguing messages about XGI Technology potentially open-sourcing their Linux display drivers with a GPL-like license. However, since ATI's buyout earlier this year of Macrosynergy, an XGI Technology alliance company, our XGI sources have dried up. XGI's current Linux drivers for the Volari graphics cards are largely undeveloped with only recent support for the Linux 2.6 kernel, and other critical features continue to lack. With Linux not being a primary desktop or gaming platform around the world at this present time, NVIDIA and ATI's alternative OS focus for consumers and gamers largely are a derivative of their substantial workstation focus with the Quadro and FireGL products.

There of course been efforts to reverse-engineer the products but as Michael Larabel was quoted in the CNET article, these developments largely come too late for most users with them being months (or years) behind the proprietary competition. Many of these reverse-engineering efforts thus far have failed or produced rudimentary results. The open-source ATI Radeon drivers for the Radeon 7500 - 9250 (R100 and R200 GPUs) do have reliable open-source support as they were written with specifications provided by ATI. Among the ATI Linux driver projects is the SourceForge-based GATOS and r300, which has aimed to provide enhanced drivers for many Radeon graphics cards.

As mentioned earlier, the aim for this article is to provide quantitative results between the X.Org open-source ATI drivers (Radeon 9250 or lower) and that of ATI's proprietary fglrx display driver. For performing these tests, we dug out an ATI Radeon 9250 128MB GPU in conjunction with other computer hardware components for its time. X.Org v7.0.0 was used while the ATI fglrx v8.24.8 drivers were used -- which are the latest drivers at this time and are available from the Livna repository for Fedora Core 5. With this card being able to barely push any modern-day Linux-native game, for the frame-rate tests today we turned to Enemy Territory. Originally, we had intended on also delivering Unreal Tournament 2004 benchmarks, but with the Demo v3334, the game experiences a segmentation fault with the Radeon 9250 and v8.24.8 drivers.