Hmmmm, the blog here doesn't really sell it right IMO.

CPUs are actually hugely parallel today. Every CPU, even the "small cores" (E-Cores from Intel or little cores from ARM) are out-of-order superscalar cores. This means that every CPU can execute 2, 3, 4, 6, 8 or more instructions per clock tick.

What a CPU is, fundamentally, is a device that takes a classically written sequential program and *DISCOVERS* parallelism for you. And then executes that discovered parallelism.

A GPU, and SIMD in general, is the opposite. SIMD requires that the programmer writes the program in a grossly parallel manner before it even executes on the GPU. Naturally, GPUs lead to more parallelism, not due to magic but simply because the programmer thought about parallelism in a more scalable way.