AMD is working on compatibility with NVIDIA CUDA

Closed standards in the future lose open. The situation can be different only if the creator of the standard is the actual monopolist in the market. Closed standards include the calculation technology on the GPU NVIDIA CUDA, which appeared on the market in 2007 with the announcement of the world’s first G80 unified graphics processor. Open Standard OpenCl appeared later, only in 2009, so Cuda managed to gain certain popularity in the GPGPU niche and supercomputers. But it seems that Cuda’s exclusivity comes to an end, as is the nvidia monopoly on this technology.

As you know, the Advanced Micro Devices is a zealous adherent of OpenCl and spends a lot of effort on promoting this standard, including as part of the initiative of heterogeneous calculations — HSA technology is used in AMD classes APU. Recently, the company introduced the new HCC compiler (Heterogeneous Compute Compiler), and then realized for it the compatibility Layer with NVIDIA CUDA. The corresponding announcement was published on November 16. According to AMD developers, the new HIP function (Heterogeneous Compute Interface for Portability) in some cases allows you to automatically convert up to 90 % of the code written under CUDA into a standard model of C++.

This does not mean a direct and direct conversion of CUDA code into OpenCl, or the direct launch of CUDA programs on the AMD graphic chips, since the company still does not have the necessary NVIDIA licenses, although the latter opened licensing in 2013. But the HIP technology allows the GPGPU programmer to work as part of the paradigm convenient for him, and the subsequent conversion to the C ++ standard will allow other programmers to optimize the code taking into account the needs of the customer, for example, planning a supercomputer based on AMD FirePro support with OpenCl support. The new AMD initiative will strengthen the company’s position in the super -calculation market (HPC), which in itself is quite conservative and prone to the use of CUDA, the standard that appeared the first and well mastered over the past time.



Leave a Reply

Your email address will not be published. Required fields are marked *