[marshray]

The idea that a hardware RNG would ever need to consume a noticeable amount of power on a 20W CPU seems strange to me. Lots of low power chips have them. A quick search returns a paper titled "A 2.92μW Hardware Random Number Generator".

[jbri]

Having a look at that paper, the actual rate at which it produces bits is very low - for 2.92μW, they're producing only 500 bits per second.

For an embedded device that might be suitable, but for a consumer or server machine you need a much higher bitrate.

[tedunangst]

You only need a little "grade-A" random to seed an appropriate high bitrate algorithm from time to time.

[marshray]

No one has ever broken a properly designed CSPRNG properly seeded with more than 100 or 200 bits of entropy, total.