You still don't need the unwieldy blog post to explain that. The SIMD section is unnecessary. Comparing the naive dot product on CPU vs GPU (+round-trip transfer) would have sufficed.