yes and no. they're a bunch of distinct simt cores. so what works badly on a few x8s SIMD cores can work great on GPUs. ex: df.groupby('x').agg(my_weird_symbol

yes and no. they're a bunch of distinct simt cores. so what works badly on a few x8s SIMD cores can work great on GPUs. ex: df.groupby('x').agg(my_weird_symbolic_func) can do well in rapids.ai and badly on AVX.