For those interested in playing with the latest open-source ATI code but don't want to dive into kernel mode-setting or the new AtomBIOS parser, you may want to test out the CS branch. The CS branch was started last week by Luc Verhaegen as it turns all calls made by the driver into macros. This was done due to GCC not optimizing the inline functions within the xf86-video-radeonhd driver on OpenSuSE, but now the performance of these calls is much more efficient. This driver branch also introduces a command submission infrastructure for R500 MMIO (Memory Mapped I/O) to allow the use of the GPU's CP (Command Processor) without depending upon DRM (Direct Rendering Manager).
Many Linux users will be celebrating the Christmas holiday in five months, but it seems there's a holiday worth celebrating today for open-source ATI Linux users.
There are already separate branches of the xf86-video-radeonhd driver for AtomBIOS support and another one for improved 2D acceleration, but today we have yet another branch. This new branch is called "CS" and what it does is change all calls within the driver for the command sequences (the "CS") into macros.
Since last month's release of Catalyst 8.6 for Linux we've seen the introduction of the Radeon HD 4850 and 4870 graphics cards and we've been allowed to share with you that CrossFire is coming to Linux along with other yet to be announced features. Today AMD has released the Catalyst 8.7 Linux driver and it doesn't deliver any new ground-breaking features, but it does bring a few improvements.
While AMD's financial outlook has been bleak with it closing down 12% today, if you're a Linux user -- particularly one with a quad-core Phenom processor -- there is good news to report from the AMD camp.
For those of you that have been using the open-source xf86-video-ati driver, need we remind you of its rapidly-improving state and feature set? One of the latest additions to this open-source ATI driver that supports the old ATI R100 graphics cards up through the new Radeon HD 4800 series (RV770) is tear-free acceleration. The current implementation of this tear-free acceleration is for EXA and Textured Video (X-Video) and should eliminate any "tearing" issues that some users experience. This code isn't yet found in the master branch of its development git repository, but can be acquired through the vsync_accel branch. Alex Deucher, the mastermind behind this latest code, has described this improvement on his blog.
This morning an embargo expired covering AMD's newest high-end graphics card, the ATI Radeon HD 4870 X2 (codenamed the "R700"). Like the Radeon HD 3870 X2, this new X2 variant just combines two of AMD's high-end GPUs on a single PCB and connected via CrossFire. The Radeon HD 4870 X2 with its dual RV770 cores aren't available today, but it will start shipping later this summer.
Earlier this week we reported on RadeonHD driver support for the RV770 with the Radeon HD 4850 and Radeon HD 4870 graphics cards. This support arrived within the RadeonHD driver's new AtomBIOS branch that relies upon ATI's video BIOS abstraction layer as opposed to "banging the registers" and interfacing with the hardware directly. However, the lead developer of the xf86-video-radeonhd driver has back-ported the RV770 support using these hard-coded paths.
The xf86-video-ati 6.8.0 driver was released back in February, and today version 6.9.0 has been released, which is coming just a day after its 6.9.0-rc2 release. Version 6.9.0 of this ATI driver adds improved EXA render support for R100/200 graphics cards, EXA render support for R300/400/500 graphics processors, and Textured Video support through X-Video for R100-500 graphics processors. This open-source driver supports all generations of Radeon graphics processors (aside from official support for the just-released HD 4850 and HD 4870). The complete change-log for xf86-video-ati 6.9.0 and source download links can be found on the X.Org mailing list.
While AMD still has yet to release any R600 programming documentation or the source-code to their KGrids or TCore simulators (though the documentation may finally just be days away), Alex Deucher and David Airlie have been working on R600 DRM (Direct Rendering Manager) support. Today the first bits of this DRI component for the Radeon HD 2000/3000 series is now available. Within the Mesa/DRM git tree at FreeDesktop.org is a new r6xx-support branch. This R600 DRM uses the CP (Command Processor) for communication, but before checking out this branch, be forewarned that 3D acceleration isn't ready yet. While there is open-source R500 3D support and it's working quite well, Alex believes it will be at least another month or two until the Mesa and DDX code is in place for this R600 3D hardware acceleration.
Twelve days after the first xf86-video-ati 6.9.0 Release Candidate, the second RC release is now available for testing. Since xf86-video-ati 6.9.0-rc1, the man page has been updated, a ShadowFB R600 fix, PLL tweaks, a possible fix for VGA on ATI IGP chipsets, warning fixes, cleanups, and other work. If you're interested in testing out xf86-video-ati 6.9.0-rc2 on your Radeon graphics card (up to and including the just-announced Radeon HD 4850/4870), this is a git-only release and can be cloned from git://anongit.freedesktop.org/git/xorg/driver/xf86-video-ati.
Now that there is a Stream SDK for Linux (related reading: AMD Stream Linux Q&A), if you're looking for a new solution for accelerating CAL and Brook+ development / stream computing, AMD has a possible answer for you. This morning in Dresden, Germany they have announced the FireStream 9250. The FireStream 9250 is also the first unit to break the tera-flop barrier! The first FireStream part introduced earlier was the FireStream 9170, which provides up to 500 GFLOPS of computing power. The FireStream 9250 will be available later this year at a price of $999 USD. More information is available in their press release.
AMD's Alex Deucher has today announced the availability of the documentation covering the R600 Family Instruction Set Architecture. This ISA (Instruction Set Architecture) documentation covers the unified shader block found on the Radeon HD 2000/3000 series and newer. This PDF document is 342 pages long and does go into detail surrounding R600 vertex and geometry shaders.
AMD has yet to publicly release the R600 (Radeon HD 2000 / Radeon HD 3000) documentation, but the xf86-video-radeonhd developers at Novell have had this documentation under NDA. This documentation, however, is expected to be cleared for release soon.
Yesterday was NVIDIA's turn in the spotlight with the introduction of the GeForce 9 Mobile GPUs and Hybrid SLI. Today the attention turns to AMD with their new announcements coming out of Computex. AMD has introduced their next-generation notebook "Puma" platform, its fastest notebook graphics processor ever, an external graphics solution for notebooks, and PowerXpress improvements.
NVIDIA has long supported their CUDA (Compute Unified Device Architecture) technology on Linux for allowing general-purpose code algorithms to be executed on the graphics processor, while AMD and their Stream Computing support has been absent on Linux. AMD has only been supporting their Stream SDK on Windows XP, but this morning we have confirmation that the Software Development Kit will be released for Linux in the coming days. According to AMD's Michael Chu on the AMD Developer Forums, an SDK v1.1 Beta is expected within the next two weeks (this message appeared a week ago) and that testing has been done with Red Hat Enterprise Linux and SuSE Enterprise. This SDK will make it possible to use CAL and Brook+ on Linux, permitting of course you're using an R600 GPU.
For those of you using an RS480 or RS690 IGP with the open-source xf86-video-ati, there is great news coming out of the Airlie camp. Compiz is now working for the RS480 and RS690! David Airlie has found a bug in the Mesa R300 DRI driver and has committed a patch (containing just three new lines of code and one line removal) correcting this issue. So that non-git users can experience the joys of Compiz on their ATI IGPs with the open-source driver, David will be pushing this fix into the Mesa 7.0.x (the next release should be Mesa 7.0.4) as well as releasing updated Mesa packages for Fedora 8 and Fedora 9. David Airlie mentioned this on his blog.
AMD's Alex Deucher has taken the Mesa r500test branch maintained by David Airlie for the open-source ATI R500 work and has pushed it into Mesa's r500-support branch. This branch is for those looking to play with the initial open-source R500 DRI (Direct Rendering Infrastructure) support. Alex has also added in the remaining R500 PCI IDs so that all of you with Radeon X1000 hardware should be able to more easily experience the joys of hardware-accelerated glxgears (Okay, it's not quite as exciting as ET:QW or Doom 3 but still a major milestone).
The RadeonHD DRM isn't yet in a usable state to end-users, but Matthias Hopf has pushed his latest RadeonHD work into his personal git tree. DRM is now able to initialize on an RS690 GPU, but DRI isn't working yet. Nor does Matthias's code work with 2D acceleration currently. However, he does hope to have this code working within a few weeks. The git URL and his message can be read on the RadeonHD mailing list. Luc Verhaegen has also made a very interesting RadeonHD enhancement without DRM, which could be available for public showing soon.
Just four days ago the RadeonHD 1.2.0 driver was released as the first xf86-video-radeonhd driver update in several months. The RadeonHD 1.2.0 driver added support for several new Radeon HD 3XXX GPUs, R500 2D XAA/EXA acceleration, and many other underlying changes. Today, however, that release has been succeeded by RadeonHD 1.2.1.
The RadeonHD crew yesterday released xf86-video-radeonhd 1.2.0 while the xf86-video-ati developers (mainly Alex Deucher) have been working on a few commits for their competing driver. There have been about a dozen commits to this driver tree in the past 24 hours. The changes mostly come down to EXA/TexturedVideo improvements as well as some code cleaning. If you use this driver, be sure to check out the latest development code.
Committed to the xf86-video-radeonhd git tree today have been 17 changes to this open-source ATI R500/600 graphics driver. AMong the changes are adding full support for the RS690 IGP family, support for interlaced modes, and better support for LVTMA TMDS macro control values. All of the changes with today's drivers can be found in their public git tree at FreeDesktop.org. IRC logs from the RadeonHD channel are available at RadeonHD.org.
When it comes to ATI open-source power management, so far AMD's open-source kindness has just yielded a list of registers but no proper documentation or how PowerPlay exactly works. However, AMD's Alex Deucher has just committed "Dynamic Clocks" support for the Radeon R500 and R600 graphics cards. Dynamic Clocks just provides dynamic clock gating and static power management, permitting the graphics card supports this capability. With some graphics cards it's already been setup through the video BIOS, so today's work isn't relevant to everyone. Dynamic Clocks is not nearly as elaborate as PowerPlay, but it's a good start for the xf86-video-ati driver.
In late February AMD had released the much anticipated R500 3D programming guide followed my two minor revisions to it over the past month. Today AMD has released another revision to the R500 3D programming guide, which now includes expanded coverage of the Command Processor (CP) found on the R500 graphics processors. Version 1.3 of this documentation can be downloaded from X.Org.
For those using motherboards with the AMD RS690 IGP, the DDIA Digital Block is now supported by the RadeonHD (xf86-video-radeonhd) driver. DDIA is the second digital block on this IGP that came as a mystery to both AMD and the RadeonHD developers as they believed no motherboard vendors were actually using this block. With this latest code addition, two displays should now work with the RS690 in an independent fashion. Check out the latest xf86-video-radeonhd git code if you're one of these RS690 owners. A few other code fixes were also committed this afternoon.
Last Thursday, David Airlie achieved hardware-accelerated glxgears on an open-source R500 Mesa implementation based upon the earlier R300 code. This is a big step forward for open-source 3D on these newer ATI Radeon graphics cards, but it's still a work in progress. However, this morning David Airlie has merged his Mesa/DRM work to the mainline DRM branch. This work, previously housed in his personal git tree under the branch "r500-fp", can now be found in the master Mesa/DRM git. His Mesa work (in the "r500test" branch) still hasn't been merged to master as there is more work to be done on that side.
Alex Deucher has been working on delivering full EXA Composite support for the Radeon driver and today the first bits of this work are available via his personal git repository. The full EXA Composite not only covers the R300 and R400 generations, but it already supports the Radeon X1000 (R500) graphics cards! In Alex's blog post he mentions that some blend combinations still need to be debugged, but the R300/400 support in general is pretty solid. For the R500 support, full EXA compositing is working for some graphics cards but not others. For those interested in trying out this latest Radeon work, check out the r3xx-render branch (well, until it gets merged to master). On the other side of the table, currently the RadeonHD driver lacks full EXA Composite support.
In the Phoronix Forums following Friday's release of the (NDA-free) ATI R300 3D register information, AMD's John Bridgman has confirmed that he is looking to release documentation going back to the ATI Radeon 8500 era graphics processors. Years ago ATI had released this R200 documentation to the open-source driver developers at the time, but it was encumbered by legal restrictions.
At CeBIT 2008 in Germany, AMD today announced the 780 Series Chipset. This budget-minded motherboard chipset, which is compatible with Quad-Core Phenom CPUs but a step-down from the 790FX Chipset, takes on gaming and high-definition computing for mainstream PC users on both the desktop and mobile platforms. What is special though about this chipset is its support for AMD's Hybrid Graphics Technology.
Less than a week after AMD introduced its R300-500 3D programming guide, they have today pushed out a revised 3D programming guide. This updated documentation covers more vertex program formats than the v1.1 draft that had come out just before FOSDEM 2008. This addition adds four pages onto the 3D documentation, making it now 266 pages long. This documentation can be downloaded from the AMD developer website.
1051 AMD news articles published on Phoronix.