2 ms·
Of course GPUs are MIMD. They are also SIMD (or SIMT if you prefer). The same is true for CPUs. Parallelism is balanced along multiple axes, and if your prob
by moonchild 4y ago
Of course GPUs are MIMD. They are also SIMD (or SIMT if you prefer). The same is true for CPUs. Parallelism is balanced along multiple axes, and if your problem is sufficiently heterogeneous, it seems completely plausible that the ideas in the linked paper would be useful for extracting additional parallelism from the hardware.