The Nvidia CUDA platform has been on the market for almost 15 years. This is a software package, which is an extension of the C programming language, and is designed for non-graphical computing on the GPU.
At the same time, CUDA has always been exclusive to Nvidia cards, because even the computing cores in the company’s GPU are called CUDA cores. Tools for porting CUDA applications to widely available programming languages have existed before, but even semi-automated tools required active developer intervention.
The ZLUDA tool, which has recently become available, supposedly allows you to run CUDA applications on Intel GPUs without any changes. Unless you need to keep in mind that processors not older than the Skylake generation are supported.
According to the assurances of the developers, ZLUDA should provide almost the level of native performance, but the possibilities are limited. In particular, only Geekbench is currently running. In this benchmark, the developers got 6333 points on the Core i7-8700K versus 6482 points on OpenCL. That is, the drop in performance is really small.
Moreover, if you look in more detail, you can see that in a number of subtests ZLUDA provides significantly higher performance.
Of course, now this is just a blank, but later ZLUDA can become a very interesting tool for developers. Especially considering the release of powerful discrete Intel graphics cards this year.