np.dot is faster than jnp.dot in docs

[rozeappletree]

According to docs (https://jax.readthedocs.io/en/latest/notebooks/quickstart.html)

x = np.random.normal(size=(size, size)).astype(np.float32)
%timeit np.dot(x, x.T)

Takes

223 ms ± 804 µs per loop (mean ± std. dev. of 7 runs, 1 loop each)

which is faster than

from jax import device_put
x = np.random.normal(size=(size, size)).astype(np.float32)
x = device_put(x)
%timeit jnp.dot(x, x.T).block_until_ready()

401 ms ± 928 µs per loop (mean ± std. dev. of 7 runs, 1 loop each)

Am I missing something ???

[jakevdp]

Thanks for the question. Agreed this is a bit confusing, but just before the timing you quote the docs say

If you have a GPU (or TPU!) these calls run on the accelerator and have the potential to be much faster than on CPU.

The readthedocs documentation build uses a CPU, where, for simple microbenchmarks consisting of single array operations, numpy is generally faster. Perhaps we could add some language to the doc to clarify this.

[jakevdp]

I've added a section to the FAQ addressing this, because this kind of question comes up often: #8681