After months of speculation, AMD has pressed the big red “go” button on the Radeon HD 5000 series. The reviews are pouring in–and we’ll get to those–but we want to focus on what improvements have been made to make for a superior end-user experience.
Raw horsepower
One of the big initiatives over the last few years has been the field of “stream computing.” This fledgling industry leverages the power of today’s GPUs–which are more like CPUs than ever before–to handle tasks like physics, video coding, and game AI. In essence, GPUs are now crunching the numbers that once required a CPU.
Both ATI and NVIDIA have worked to develop the stream computing initiative. Between NVIDIA’s CUDA language and ATI’s Stream Technology, both major industry players now have a solution to do stream computing. But just like the processor, heavier workloads and higher performance call for faster chips, and the HD 5000 series is no slouch in that department.
According to AMD, the new Radeon HD 5870 offers 544 double-precision GFLOPS of processing power. The FLOP stands for FLoating point Operations Per Second. A floating point operation is a basic calculation used by the CPU to process code, especially “scientific” ones like computer AI, video encoding and physics. Double-precision FLOPS ensure a high degree of accuracy in these calculations, which translates to more accurate rendering or encoding. The Radeon HD 5870 can perform 544 billion such double-precision calculations every second; in comparison the Intel Core i7 975 XE chip can only perform 70 billion of them.
ATI's TeraScale 2 is more powerful than ever.
In terms of raw numbers, the new Radeons are at least 7 times more powerful than the CPU. In other terms, that’s twice the capacity offered by the Radeon HD 4870. That is some seriously righteous throughput, and the rise of DirectX 11 will ensure that it will increasingly be put to very good use.
Image quality
The difference between filtered and unfiltered scenes.
Gamers everywhere know that the sense of immersion is dependent upon the quality of the story and the quality of the graphics. In the latter department, ATI has made strides to dramatically improve overall IQ while simultaneously improving performance.
ATI has undertaken a significant effort to improve the overall level of anisotropic texture filtering. Anisotropy is a technique by which textures viewed from a distance or at oblique angles are processed to retain their sharpness. A 3D scene–such as any game–without a degree of anisotropy can appear “blurry” or “muddy” towards the horizon, and this effect can be quite pronounced.
ATI’s effort to improve their anisotropic filtering has given rise to texture processing which produces a near-perfect circle at any angle. Older video cards cannot replicate a perfect circle when forced to render a texture at an extreme angle, such as the one found in the image below. This will have the direct effect of ensuring that game textures on things like walls and characters will remain sharp and clear no matter how you view them.
Beyond the reform of its anisotropic engine, ATI has also doubled the accuracy of all texture rendering. That means users who eventually sidle up to even the most bargain basement GPU in the 5000 series will enjoy an overall improvement in texture quality.
Perhaps best of all, this new image quality voodoo comes at virtually no performance impact over prior GPUs. Anyone who fires up one of the new Radeons and kicks on a high level of anisotropy will not suffer painfully lower frame rates.
ATI has put an emphasis on making textures sharper.
ATI also dedicated significant time to improving the Radeon’s anti-aliasing engine. Anti-aliasing is a technique whereby the video card works to reduce jagged edges, sometimes called “jaggies,” on 3D objects. Again, the difference between a game being anti-aliased and one that is not can be rather noticeable
The chain links on the right have been anti-aliased.
In particular, the most work has gone into improving Multi-Sample Anti-Aliasing performance. MSAA is today the most traditional way to anti-alias a scene. There are several implementations, but the basic mechanism reads, or “samples,” a pixel’s colors a number of times equivalent to your AA level (2x, 4x, 16x, etc.) and uses them to calculate a final color. The result is a smoother color transition along a line, and therefore a smoother image.
ATI has improved MSAA performance to such an extent that 8x MSAA now performs every bit as well as 4x MSAA might have on Radeon 4000 parts.
Work has also gone into the ATI-proprietary Custom-Filter Anti-Aliasing method. Like MSAA, CFAA randomly samples areas on a pixel a number of times equivalent to your AA level. But CFAA differs by taking samples in an area that extends beyond the boundary of each pixel. Taking samples that include neighboring pixels gives the GPU a better idea of how to create a smoother color transition, and the result is an even smoother picture.
ATI claims that the performance of CFAA has improved, and that is no doubt due to improvements in the algorithm, as well as the hardware.
A new MSAA engine offers a consistently higher level of performance.
Lastly, ATI has revived Super-Sampling Anti-Aliasing with the Radeon HD 5000. SSAA is the first and oldest AA method used by GPUs, and it offers tremendous quality in exchange for a heavy performance penalty. SSAA renders each frame at a much higher resolution than what is selected, and then scales the image down prior to putting it on the monitor. Anyone who has shrunk a picture knows that the result has softer edges, and the same is true for SSAA. The problem with SSAA is that, for example, a 1680×1050 game with 8x SSAA forces the GPU to render at 13440×8400. This is obviously unrealistic for games that are already demanding of the GPU.
Even so, SSAA is a high-quality choice for older games that won’t crush the GPU when being rendered at outrageous resolutions. ATI’s decision to return this anti-aliasing mode is largely due to consumer request, but it’s a nice perk for gamers who want to revisit older DirectX 9 titles in renewed detail.
Pages: 1 2
Articles RSS