Radeon R200
From Wikipedia, the free encyclopedia
The Radeon 8500 (a.k.a R200) was the core Radeon video card that put ATI on a competitive level with nVIDIA, and marked out ATI from other failing graphics card companies of the period.
Contents |
[edit] Flagship product
ATI's first DirectX 8 card was the Radeon 8500. In early 2002, ATI launched the Radeon 8500LE (re-released later as Radeon 9100), an identical chip with a lower clock speed and slower memory. Whereas the full 8500 was clocked at 275 MHz core and 275 MHz RAM, the 8500LE was clocked more conservatively at 250 MHz for the core and 200 or 250 MHz for the RAM. Both video cards were first released in 64 MB DDR SDRAM configurations; the later 128 MB Radeon 8500 boards received an extra performance boost resulting from a memory interleave mode.
[edit] Architecture
Radeon 8500 had a similar fundamental architecture as the other cards of the time; consisting of 4 pixel units, each with 2 texture units, along with DirectX 8.1 vertex and pixel shaders. The 2 vertex shaders were called Charisma Engine II, and offered excellent performance for both the new vertex shader programs and legacy DirectX 7 hardwired T&L. The "R200" core was more capable with regards to advanced pixel and vertex shading when compared to the competition at the time (GeForce 3 and 4), supporting DirectX 8.1 (pixel shader 1.4) with Pixel Tapestry II. R200 also had the most advanced memory bandwidth saving hardware of the time onboard, the next version of ATI's innovative HyperZ, HyperZ II. The chip was capable of dual-monitor display, through the Hydravision technology. Finally, R200 was equipped with Video Immersion II, ATI's advanced video decoding engine with high quality hardware deinterlacing. A DVI-I to component conversion connector was available for home theater enthusiasts.
R200 introduced pixel shader version 1.4, a significant revision to the 1.x pixel shader design. Whereas PS1.2 and 1.3 were just improvements to PS1.1, PS1.4 changed things around quite significantly. Notable instructions include "phase", "texcrd", and "texld". The phase instruction allows a shader to operate on two separate "phases" (2 passes through the hardware), effectively doubling the maximum number of texture addressing and arithmetic instructions, and potentially allowing the number of passes required for an effect to be reduced. This allowed not only more complicated effects, but can also allow a speed boost by utilizing the hardware more efficiently. The "texcrd" instruction moves the texture coordinate values of a texture into the destination register, while the "texld" instruction will load the texture at the coordinates specified in the source register to the destination register.
| DirectX 8.0 Pixel Shader 1.1 |
DirectX 8.1 Pixel Shader 1.4 |
||
|---|---|---|---|
| Max. Texture Inputs | 4 | 6 | |
| Max. Program Length | 12 instructions (up to 4 texture sampling, 8 color blending) |
22 instructions (up to 6 texture sampling, 8 texture addressing, 8 color blending) |
|
| Instruction Set | 13 address operations, 8 color operations | 12 address / color operations | |
| Texture Addressing Modes | 40 | virtually unlimited |
While R200 lost a texture unit per pipeline compared to R100's 2x3 architecture, R200's pipelines were far more robust. Each could address a total of 6 texture layers in a single pass if the application was coded to do so. The chip achieved this by a term commonly known as ‘loop-back’. Increasing the flexibility in the number of textures allowed per pass reduced the number of times the card was forced into ‘multi-pass rendering’, a performance draining scenario which increases both the geometry required for the scene (by needing to recalculate the entire scene for the number of times it is passed) and the external memory bandwidth used. Therefore, the chip's overall efficiency and performance was increased by enabling more textures to be addressed per pass. Testing using Serious Sam The Second Encounter showed a 16 % performance increase at 1600x1200 resolution with quad texturing and bilinear filtering (in comparison with dual and triple texturing).
[edit] Features
With Radeon 256 ATI had introduced their take on anisotropic filtering, using a method called "RIP" mapping. RIP mapping is not technically accurate anisotropic filtering, but it offers a very similar effect while significantly reducing performance impact. Unfortunately, their technique was only functional in combination with bilinear filtering. The implementation was also quite angle dependent, meaning that the texture had to be at a certain angle to the viewport or it would not be sharpened. Of course this reduced the computational load, but it also made it possible to see textures that were merely bilinear filtered right beside nearby sharpened textures. Radeon 8500 uses the same technique, but with some refinements to improve quality while still maintaining excellent performance. When anisotropic was enabled on 8500, the boundaries between mip maps, normally easily seen when using bilinear filtering instead of trilinear, were fairly well hidden and the anisotropic mode greatly improves texture sharpness at medium to long distances. Still, the remaining angle dependency and forced-bilinear were controversial from a quality standpoint. NVIDIA's GeForce 4 Ti series offered a far more accurate anisotropic implementation, but it also carried with it a heavy performance hit.
The 8500 had the first and only ATI hardware accelerated tessellation engine called "Truform" which could add polygons to 3D models and smooth them out. The technology required developer support though and was not well supported; unfortunately "Truform" was not ideal for all 3D rendering scenarios and had a tendency to round out objects undesirably. TruForm hardware wasn't included in future ATI GPUs, but was still supported. Future chips either performed Truform in software on the host system's CPU, or as a vertex shader program, neither of which were as efficient as 8500's Truform unit.
[edit] Performance
Radeon 8500's biggest disappointment was its initial driver release. At launch, the card's performance was poor to say the least. Initial drivers were very buggy and weren't feature complete, not even supporting everything the board was capable of. The card's anti-aliasing support was only functional in Direct3D, and was very slow at that. In many games the early drivers did not function perfectly, displaying strange issues such as polygon misplacement and flashing textures. nVidia released their Detonator4 driver pack on the same day as most web sites previewed the Radeon 8500, and nVidia's drivers at the time were of substantially better quality, and they also significantly boosted the GeForce 3's performance.
Worse still, several benchmark sites discovered that the real performance of the Radeon 8500 was much lower than benchmarks reflected. Reviewers discovered that ATI was detecting the executable Quake3.exe and forcing the texture filtering quality to much lower than normally produced by the card. Kyle Bennett of HardOCP was the first to bring the issue to the community, and he proved its existence by renaming all instances of "Quake" in the executable to "Quack." The result was much improved image quality, but lower performance.
Still, even with the Detonator4 drivers, the Radeon 8500 was able to outperform the GeForce 3 (which the 8500 was intended to compete against) and in some circumstances its faster revision, the Ti500, which Nvidia had rolled out in response to the R200 project. In addition, updated drivers helped to close the performance gap between the 8500 and the Ti500, while the 8500 was also significantly less expensive and offered additional multimedia features such as dual-monitor support. Though the GeForce 3 Ti200 did become the first DirectX 8.0 card to offer 128 MB of video memory, instead of the common 64 MB norm for high-end cards of the time, it turned out that the GeForce 3's limitations prevented it from taking full advantage of it, while the Radeon 8500 was able to more successfully exploit that potential.
In early 2002, to compete with the cheaper GeForce 3 Ti200, ATI launched the slower-clocked 8500LE which became popular with OEMs and enthusiasts due to its lower price, and overclockability to 8500 levels. The delayed release of the potentially competitive GeForce 4 Ti 4200, plus ATI's initiative in rolling out 128 MB versions of the 8500/LE kept the line popular among the mid-high performance niche market. The greater features of the All-In-Wonder (AIW) Radeon 8500 DV and the AIW Radeon 8500 128 MB proved superior to Nvidia's Personal Cinema equivalents which used the faster GeForce 4 Ti 4200.
[edit] Refresh
A new high-end refresh part, the 8500XT (R250) was supposedly in the works, ready to compete against the GeForce 4 Ti line, particularly the top line Ti 4600. Prerelease information touted a 300 MHz core and RAM clock speed for the "R250" chip. ATI, perhaps mindful of what had happened to 3dfx when they took focus off their "Rampage" processor, abandoned it in favor of finishing off their next-generation DirectX 9.0 card which was released as the Radeon 9700. A Radeon 8500 running at 300 MHz clock speeds would have hardly defeated the GeForce 4 Ti4600, let alone a newer card from NVIDIA. At best it could have been a better performing mid-range solution than the lower-complexity Radeon 9000, but it would also have cost more to produce and would have been poorly suited to the Radeon 9000's dual laptop/desktop roles due to die size and power draw. Notably, overclockers found that Radeon 8500 and Radeon 9000 could not reliably overclock to 300 MHz without additional voltage, so undoubtedly R250 would have had similar issues because of its greater complexity and equivalent manufacturing technology, and this would have resulted in poor chip yields, and thus, higher costs. [1] [2]
[edit] Mainstream Line
The Radeon 9000 (RV250) was launched alongside the 9700 (the new flagship part for ATI). This chip dropped one of the two texture units, the "TruForm" unit, Hierarchical-Z, and one of the two vertex units, bringing the configuration down to a 4x1 pixel/texture pipeline layout. It was not just cut down, however, but was actually refined as well. The texture cache was doubled in size to 4KB, improving a serious inefficiency in R200. Because of this, performance was still quite competitive, considering that the "R200" chips were more expensive and much larger and power consuming. In games, it performed around the same as the highly-refined "NV17" used on GeForce 4MX440. Its main advantage over the GeForce4 MX440 was that it had a full DirectX 8.1 vertex and pixel shader implementation. The Radeon 9000 replaced the uncompetitive Radeon 7500 (RV200) in the mainstream market segment.
A later version of the 9000 was the 9200 (RV280), which, aside from supporting AGP-8X, was identical. However, there was a cheaper version, the 9200SE, which only had a 64-bit memory bus. Another board, called the Radeon 9250, launched in summer 2004. It was simply a lower-clocked variant of Radeon 9200. It, in fact, used the same "RV280" GPU. It was usually equipped with more RAM than the Radeon 9200 cards though (128 MB or even 256 MB), taking advantage of the low-cost of slow-clock high-density DDR SDRAM, a popular trend at the time.
[edit] Mobility
The derivative Mobility Radeon 9000 was launched in early summer 2002 and generated more press than its desktop counterpart since it was the first DirectX 8 laptop chip. It far outperformed the competition at the time (nVidia GeForce 2 Go), was more feature-rich than the GeForce 4 Go, and had advanced power-saving technologies incorporated. It helped to maintain ATI's lead in the mobile market established by the previous Mobility Radeons. As another first ATI had the new Mobility 9000 ready to ship in several OEM models within days of its announcement.
A Mobility Radeon 9200 later followed as well, derived from the desktop 9200.
[edit] Models
(Sorted by model)
| Desktop Graphics Boards | |||||||||
| Board Name | Core Type | Die Process | Clocks (MHz) Core/RAM | Core Config1 | Fillrate2 (Mpx/s:Mtx/s) |
Memory Interface | Memory Bandwidth | Memory Size | Notes |
|---|---|---|---|---|---|---|---|---|---|
| 8500 | R200 | 150 nm | 275/275 | 4x2:2 | 1100/2200 | 128-bit | 8.8 GB/s | 64/128 MB | 128 MB variant has memory interleave. |
| 8500 LE | R200 | 150 nm | 250/250 | 4x2:2 | 1000/2000 | 128-bit | 8.0 GB/s | 64/128 MB | 128 MB variant has memory interleave. |
| 8500 XT | R200 | 150 nm | 300/300 | 4x2:2 | 1200/2400 | 128-bit | 9.6 GB/s | 128 MB | unreleased, only Gigabyte offered 300/300MHz version |
| AIW 8500 DV | R200 | 150 nm | 230/190 | 4x2:2 | 920/1840 | 128-bit | 6.1 GB/s | 64/128 MB | 128 MB variant has memory interleave. |
| AIW 8500 | R200 | 150 nm | 275/275 | 4x2:2 | 1100/2200 | 128-bit | 8.8 GB/s | 128 MB | memory interleave. more features/speed than "DV" |
| 9000 | RV250 | 150 nm | 250/200 | 4x1:1 | 1000/1000 | 128-bit | 6.4 GB/s | 64/128 MB | no interleave after 8500. |
| 9000 PRO | RV250 | 150 nm | 275/275 | 4x1:1 | 1100/1100 | 128-bit | 8.8 GB/s | 64/128 MB | |
| 9100 | R200 | 150 nm | 250/250 | 4x2:2 | 1000/2000 | 128-bit | 8.0 GB/s | 64/128 MB | same as 8500LE, minus interleave |
| 9200 | RV280 | 150 nm | 250/200 | 4x1:1 | 1000/1000 | 128-bit | 6.4 GB/s | 64/128/256 MB | RV280 = RV250 + AGP 8X |
| 9200 SE | RV280 | 150 nm | 200/166 | 4x1:1 | 800/800 | 64-bit | 2.7 GB/s | 64/128 MB | |
| 9250 | RV280 | 150 nm | 240/200 | 4x1:1 | 960/960 | 128-bit | 6.4 GB/s | 64/128/256 MB | |
| Mobility Radeons and Integrated Graphics Processors | |||||||||
| MR9000 | M9 | 150 nm | ~250/~200 | 4x1:1 | 1000/1000 | 64-bit | 3.2 GB/s | 32 MB | Mobile RV250. Powerplay power management. |
| MR9000 | M9 | 150 nm | ~250/~200 | 4x1:1 | 1000/1000 | 128-bit | 6.4 GB/s | 64/128 MB | |
| MR9200 | M9+ | 150 nm | ~250/~200 | 4x1:1 | 1000/1000 | 64-bit | 3.2 GB/s | 32 MB | Mobile RV280. Powerplay. |
| MR9200 | M9+ | 150 nm | ~250/~200 | 4x1:1 | 1000/1000 | 128-bit | 6.4 GB/s | 64/128 MB | |
| 9100 IGP | RS300 | 150 nm | 300/NA | 2x1:0 | 600/600 | 128-bit | NA | NA | Chipset for Pentium 4. Based on RV250. Host-based vertex processing. Uses system RAM for IGP. |
| 9100 PRO IGP | RS350 | 150 nm | 300/NA | 2x1:0 | 600/600 | 128-bit | NA | NA | Tweaked 9100 IGP. Improved AGP 8X, better RAM timings and compatibility. |
| 9000 IGP | RC350 | 150 nm | 300/NA | 2x1:0 | 600/600 | 64-bit | NA | NA | "Value" version of 9100 IGP. Single channel RAM interface. |
- ¹ Core Config: pixel pipelines/texture mapping units : vertex shaders
[edit] Drivers
[edit] UNIX-Related Operating Systems
ATI has released no driver support for BSD based operating systems, however they do provide drivers for the X Window System running on Linux. The PowerPC-based Mac mini and iBook G4, which run on Mac OS X, were supplied with Radeon 9200 cards.
Some segments of the Linux user community, which prefer to avoid the IP-encumbered ATI drivers due to stability and long term maintainability reasons, still prefer the R200-based chips, as they are among the fastest modern video cards with stable open source drivers.
[edit] Windows Drivers
The Radeon cards based upon R200 were supported by all versions of ATI's Catalyst Drivers, except under Windows XP x64, until the release of Catalyst 6.6. ATI maintains an archive of previous Catalyst drivers for users of such cards.However ATI skipped and released also 6.11 Catalyst for 9250 and lower Radeon cards.
[edit] References
- "ATi Radeon 8500 64 MB Review (Part 1)" by Dave Baumann, Beyond3D.Com, March 29, 2002, retrieved January 14, 2006
- "ATi Radeon 8500 64 MB Review (Part 2)" by Dave Baumann, Beyond3D.Com, April 4, 2002, retrieved January 14, 2006
- "ATI RADEON 9100 Based Graphics Cards Review: Gigabyte and PowerColor Solutions" by Tim Tscheblockov, X-Bit Labs, February 5, 2003, retrieved January 9, 2006
- "ATI's Radeon 8500 & 7500: A Preview" by Anand Lal Shimpi, Anandtech, August 14, 2001, retrieved January 9, 2006
- "ATI's Radeon 8500: She's got potential" by Anand Lal Shimpi, Anandtech, October 17, 2001, retrieved January 9, 2006
- "ATI R200 Chip Details" by Beyond3D, retrieved January 9, 2006
- "ATI RV250 Chip Details" by Beyond3D, retrieved January 9, 2006
- "ATI RV280 Chip Details" by Beyond3D, retrieved January 9, 2006
- "How ATI's Radeon 8500 drivers 'optimize' Quake III" by Scott Wasson, The Tech Report, November 6, 2001, retrieved June 1, 2006
- "Optimizing or Cheating Radeon 8500 Drivers" by Kyle Bennett, HardOCP, October 23, 2001, retrieved June 1, 2006
|
|
