From the beginning, AMD and nVidia provide support OpenCL, thanks to which video cards of different vendors provide approximately the same opportunities GPU. However, according to developer forum AMD, the owners of video cards with chip series Radeon HD 4000 may be very unhappy when it comes to computing on the GPU.

The developer, who described himself as Matt Taylor (Matt Taylor), wrote: "We use OpenCL in the design and have one infernal machine with nVidia GTX 260, and the other with ATI 4870... It's unpleasant to acknowledge that we received from the nVidia graphics card in about 5x more performance than with the ATI ". Productivity is so low that, according to Taylor, the processor Intel Core 2 Quad processor with a frequency of 2.4GHz outperforms Radeon "twice".

AMD engineers Mick Willow (Micah Villmow) replied: "That depends entirely on how you encode the kernel and which functions OpenCL you use. There are no known performance issues with the video series HD4XXX on OpenCL, and currently no plans to focus exclusively on improved performance for this family. HD4XXX series was not intended for OpenCL, while the series was HD5XXX. will improve performance in this series because of the improvement for the series HD5XXX, so they will be better, but it is not our goal. "

Later, Willow supplement: "Radeon HD 4870 just has to be programmed differently than the series 5xxx, to get a performance boost, due to lack of adequate support to local equipment. You can get a good performance, but not directly with CUDA". He also stressed that the compiler, AMD will include more optimization under a more specific device when it comes out. In any case, this is not a good omen for the Radeon without the support of DirectX 11 in anticipation of a wave of applications OpenCL.

It should, of course, be noted that not all current nVidia card based on the same architecture, GT200, that of the GeForce GTX 260, which supposedly has shown himself so well. Models based on the G92, such as GeForce GTS 250, GeForce 9800 GT and a series of GeForce GTX 200M constitute a sufficiently large part of the current proposals nVidia, and they were all derived from the old design G80. G80 is the first DirectX 10 architecture from nVidia, and it may be the same architectural limitations as a series Radeon HD 4000, when it comes to calculations on the GPU.