But while the CUDA moat is certainly a reality for developers looking to expand support for alternative hardware platforms, the number of devs writing code at a kernel level is relatively few ...

Over at the Parallel for All blog, Mark Harris writes that Shared memory is a powerful feature for writing well optimized CUDA code. Access to shared memory is much faster than global memory access ...

I have been hearing about all the neat processing opportunities with nvidia CUDA recently. On the nvidia website there seems to be a list mostly of custom code apps and libraries for programmers ...

I've been writing a bit of code just to learn and explore the features, and I have managed to successfully compile, both for Visual C, nVidia CUDA, Intel C and Intel Fortran. So far so good, I guess.

Intel releases an open source tool to migrate code to SYCL1 through SYCLomatic, which helps developers more easily port CUDA code to SYCL and C++.

The Argonne Leadership Computing Facility has announced a webinar covering the process of porting CUDA code to SYCL, with a focus on high-performance math libraries like cuBLAS and cuFFT, to be held ...

Nvidia will release the source code for its new LLVM-based CUDA compiler, allowing developers to extend its capabilities to other programming languages and non-Nvidia processor architectures.