Benchmarks executing workloads using DPC++, oneAPI’s implementation of SYCL achieves close to native performance on Nvidia and AMD GPUs, when comparing to the same benchmarks run with CUDA® and HIP, respectively.