ATI Radeon HD 2000 Series

AMD-ATI today announced the availability of it’s next generation of video graphics cards: the Radeon HD 2000 series. Encompassing a complete line of graphics cards from mobile to mainstream to enthusiast, ATI brings it’s new main guns online. PlanetX64 was fortunate enough to have participated in the press events leading up to this launch and have a preview for your reading pleasure.

ATI will be sending us review samples for our full in-depth review later this week.

ATI’s long rumored RD600 graphics chipset has finally surfaced in the form of the Radeon HD 2000 series. This launch is not just a single board priced somewhere in the stratosphere, this is a complete graphics lineup bringing DX10 and HD viewing to all aspects of computing.

Products being announced today are:

ATI Radeon HD 2900 Series for the Enthusiast
ATI Radeon HD 2600 Series for the Mainstream
ATI Radeon HD 2400 Series for the Value segment

ATI Mobility Radeon HD 2600 Series for gaming/performance notebooks
ATI Mobility Radeon HD 2400 Series for mainstream thin & light notebooks
ATI Radeon HD 2300 for entry level notebooks

ATI Radeon HD 2000 Series Highlights

• Technology Leadership
  • Highest clock speeds – up to 800 MHz
  • Highest transistor density – up to 700M transistors
  • Lowest power for mobile
• 2nd generation unified architecture
  • Superscalar design with up to 320 stream processing units
  • Optimized for Dynamic Game Computing and Accelerated Stream Processing
• DirectX 10
  • Massive shader and geometry processing performance
  • Enabling the next generation of visual effects
• Cutting-edge image quality features
  • Advanced anti-aliasing and texture filtering capabilities
  • Fast High Dynamic Range rendering
  • Programmable Tessellation Unit
• ATI Avivo HD technology
  • Delivering the ultimate HD video experience
  • HD display and audio connectivity
• Native Crossfire technology
  • Superior multi-GPU support

We will now discuss each of these items in depth.

{mospagebreak title=2nd Gen Superscalar}

With the Xbox360, ATI introduced it’s first unified archtecture. Up until that point all ATI designs were based around a fixed number of discrete vector and scalar processing units. The Xenos graphics ship that powers the Xbox360 brought forth the notion of the stream processor, with a dispatch processor capable of processing hundreds of simultaneous threads.

Nvidia embraced this idea with their 8 series GPU bringing 96 shader units to bear against the raging game industry and sweeping through all of the benchmarks.

With Radeon HD, ATI drops a whopping 320 stream processors coupled with a new ultra threaded dispatch processor capable of handling thousands of simultaneous threads! The stream processors are capable of processing vector, scalar and non-graphics tasks and can switch on-the-fly.

And here is a shot of the stream processor layouts for the 2600 & 2400 models:

As you can see the main differences between the various models (architectually speaking) is the number of stream processors available and the attendant support hardware to drive them.

Here is a shot of the internal schematics of a shader processor:

This next picture shows (at a high level) the architectual differences between the Radeon X series and the Radeon HD series. Note that the unified shader model allows for more efficient use of the card hardware per clock cycle.

This next slide gives you an idea of the amount of computing horsepower in one of the cards.

Yes, they claim this card approaches a teraFLOP of performance! That’s a lot of OPs to grind.

{mospagebreak title=Other hardware}

The Radeon HD 2900 employs a brand new 512-bit memory controller with the following specs:

• Over 100GB/sec memory bandwidth
• Eight 64-bit memory channels
• Kilobit Ring bus
• Fully distributed design – no central hub
• Simplified layout, highly scalable

Here is an image to put it in perspective:

Here is a close up of the render backends listed in the Radeon HD 2900 layout.

As you can see, these little beauties handle the alpha/fog, depth/stencil, blending, and MSAA resolve functionality. This last one is important as it allows Custom AA filters to be applied via the GPU instead of the main processor.

Another interesting feature of the support hardware is the tesselation unit. Tesselation is the process in which a low-polygon count 3D object is subdivided to make it smoother and more realistic looking. This process is well-known to 3D artists and VFX modellers but has the drawback of increasing the polygon count of an object exponentionally, casing it to occupy more RAM and require more processing horespower. The Radeon HD’s built-in tesselation unit does this function on-the-fly allowing games to load faster by using low-poly count models that are tesselated in the GPU and then displacement mapped for high-levels of realism.

{mospagebreak title=Anti-aliasing and HDR}

These next slides are aimed at telling us all of the cool new features that will improve the quality of what we see on the screen. Most notable is the fact that AA functionality can be improved and new features added via driver updates – very cool.

The second slide show us how badly HDR performance on the new card whomps the previous generation.

Here are some slides explaining the benefits of CFAA and how well the Radeon HD performs:

So what is the point of all this? Funny you should ask….

{mospagebreak title=Dynamic Game Computing}

Dynamic Game Computing is the term ATi uses to describe what the Radeon HD series brings to the avid gamer. The three key features of Dynamic Game Computing are:

• Stunning Realism
• Advanced Visual Effects
• Comprehensive DirectX10 Support

Stunning Realism is accomplished by ultra-high polygon count objects (as provided by the Tesselation Unit), richer surface textures, complex facial and character animation (the audio-animatronic look is out) and sophisticated physics simulations. These elements combine to raise the gaming experience to the next level.

Advanced Visual Effects are provided by volumetric lighting and increased post-processing (motion blur, HDR, etc…).

Comprehensive DX10 Support. The DX10 API coupled with the vast array of stream processors in the Radeon HD gives developers the freedom to create games with highly realistic gameplay.

Geometry Performance

The Radeon HD 2900XT has the following geometry performance enhancements:

• Large vertex cache
  • 8x larger than Radeon X1950
• Fast, full-featured Vertex Texture Fetch
  • Uses same texture units as pixel shaders
• All shader units can perform vertex and/or geometry processing
  • Up to 10x the vertex processing power of an X1950
• Dedicated tesselation hardware
  • Incorporated into all Radeon HD 2000 models
  • Based on XBox360 technology

Dynamic Game Computing has three technical aspects:

1. Dynamic geometry acceleration
2. Complex pixel processing
3. Expanded non-graphical compute capabilities

You’ve just seen samples of the first two parts. We will now discuss the implications of the third aspect.

{mospagebreak title=Non-graphical computing}

Expanded non-graphical computing means the GPUs can do more than just push pixels.

• Faster, more complex sorting of transparent objects
• More realistic physics simulations
• Potential for things like:
  • Improves Artifical Intelligence
  • Pathfinding
  • Audio processing
  • User Interfaces

Radeon HD 2900 Highlights in this area:

• ATI Radeon HD 2900 Stream Processor with FP32 precision throughout + integer processing capabilities
• Support for DX10 & OGL 2.0 for graphics and CTM for stream computing
• 8192 x 8192 texture support
• Unlimited shader length (instruction cache)
• Unlimited register space (memory read/write cache)
• Accelerated state validation for reduced CPU overhead (command processor)

This next slide shows you what other applications can use the compute power of the Radeon HD series graphics processors:

Distributed Computing Projects

ATI has worked closely with distributed computing projects like Stanford University’s Folding@Home Project (http://folding.stanford.edu).

By working with the researchers at Stanford, Folding@Home will be able to take advantage of the power of Stream Processing to improve the performance of their client by a factor of 2.2x!

Here is a slide showing all of the benefits to this project:

{mospagebreak title=HD Improvements}

ATI has not skimped out on the HD capabilties of these boards. All of the Radeon HD boards have HDCP onboard incorporating the CryptoROM into the ASIC. This is at all board supported resolutions! I can finally start testing blu-ray/hd-dvd drives without cursing the questionable genetics of the board manufacturer when I try to display HD content on my 30" Dell!

HD Audio is also incorporated into the board so that Radeon HD is truly a single boad solution for HTPCs looking for the elusive Vista Premium Logo compliance. There is driver support built-in to Vista and XP users can download the driver.

While all of the boards have DVI connectors (even the entry level 2400), ATI provides an HDMI adaptor that provides not only video signal, but audio as well (thanks to the built-in HD audio chip). This means that HTPC designers will favor these cards for simplified cabling on top of all the other features.

Another HD feature is the Unified Video Decoder. ATI has designed this component to complete offload the processing of HD video content from the CPU and GPU! Handling VC-1 and H.264/AVC codecs, the UVD processes the streams internally, removing passes to memory (system and GPU) between decoding stages. The UVD is capable of sustaining full 40 Mbps bit-rates for HD optical disc playback.

{mospagebreak title=Pricing and Availability}

I have saved the best for last. When Nvidia introduced the 8800 GTX Ultra, it pushed the limits as to what a video card could cost and still be considered a gaming card. AMD-ATI have decided to reverse that trend.

Your eyes do not deceive you. The top of the line card is priced at $399 USD. Available now. The mainstream and entry level cards are also priced to sell and will be in the shops by the end of June.

I showed these slides today to some of the employees at our local Apple Store and they were all chomping at the bit for this card. This leads me to speculate that the House of Jobs is busy certifying this board for current and next-gen (Penryn-based) Mac Pro’s.

Conclusion

ATI has some very impressive specs and extremely aggressive pricing. If the boards perform as promised we can see them regaining the ground they lost to Nvidia. Even if they come up a little short, the pricing model still makes them the better deal in terms of bang for the buck. We eagerly await our review samples (hint, hint) to really give these babies a workout.

Discuss this article here.