Performance in Synthetic Benchmarks
With the ongoing transition to DirectX 10, 3DMark06 loses its significance as an accurate tool for benchmarking modern graphics cards with unified architecture. However, the Radeon HD 4850 scores almost the same amount of points as the Nvidia GeForce 9800 GTX whereas the Radeon HD 4870 scores over 12,000 points, being only inferior to the Radeon HD 3870 X2.
It’s all clear when we check out the groups of tests. Despite the improvements in the texture-mapping section, the ATI Radeon HD 4850 is inferior to Nvidia’s solution in the Shader Model 2.0 tests but wins the SM3.0/HDR group. It is quite logical considering that the RV770 has inherited its main architectural features from ATI’s previous GPU. The Radeon HD 4870 on its part is slightly faster than the GeForce 9800 GTX even in the SM2.0 tests.
The first test result has improved considerably over the Radeon HD 3870 but the Radeon HD 4850 it is still 39% slower than the GeForce 9800 GTX. The two cards don’t differ much in terms of GPU clock rate, so it can’t be the reason. We can’t find an explanation because the theoretical tests suggest that the Radeon HD 4850 doesn’t feel a lack of texture processors. The Radeon HD 4870 can’t beat Nvidia on its turf, either, at least in the first test.
The second test is more favorable towards ATI’s cards which are glad to show their ability to process vertex shaders. The Radeon HD 4850 is inferior to the GeForce 9800 GTX by only 7-8% whereas the Radeon HD 4870 is 15% ahead of Nvidia’s card because transferring arrays of vertex data requires fast graphics memory.
The new cards have no rivals in the SM3.0/HDR tests, except for the Radeon HD 3870 X2 whose processors are working in AFR mode and rendering two sequential frames during the time the Radeon HD 4850 and 4870 do only one. Despite the use of SM3.0 and HDR these tests are not complex enough to load modern solutions by 100%.