https://hgpu.org/?p=15764
To Co-Run, or Not To Co-Run: A Performance Study on Integrated Architectures