4 ms·

Am I wrong to assume that this will forever be expensive and impractical compared to e.g. Intel's RDRAND implementation where the RNG is built using the same pr

by MertsA 6y ago

Am I wrong to assume that this will forever be expensive and impractical compared to e.g. Intel's RDRAND implementation where the RNG is built using the same process as the rest of the processor? It seems like you couldn't easily integrate this on the same die as a processor and if it's a completely separate chip, how in the world are you ever going to make it economically advantageous compared to a silicon RNG feeding a CSPRNG? The CSPRNG could be outputting as much random data as you could ever want and it's simple enough to guard your seed data to that CSPRNG, not to mention reseeding it just to make side channels that much more cumbersome to exploit, such that the output of the CSPRNG is for all practical purposes just as secure as whatever output this thing is putting out.

tus89 6y ago
Well integrating a diverse range of cores into a CPU is all the rage these days...if Intel doesn't do it Apple would.
ncmncm 6y ago
A common problem is the microcontroller that powers up and needs randomness immediately, before it can initialize its peripherals and establish communication links. There, a billion or even a million bits per second may be enough. Now that we have packaged chiplets, we should be open to stuff on a different semiconductor process than our CPU cores. Where we need random bits as fast as they can be had is when running Monte Carlo processes on GPUs. There, the sources need to be on-chip because interconnect would be too expensive; we need a separate bit source integrated into each functional unit, and producing a whole machine word's worth each cycle, so a word is always ready immediately when it could be used. Sometime we need more than one before starting a computation.
fulafel 6y ago
There have boon some major problems with rdrand, on amd side at least (first the bulldozer/jaguar incident, and now the ryzen one).

by MertsA 6y ago

tus89 6y ago

Well integrating a diverse range of cores into a CPU is all the rage these days...if Intel doesn't do it Apple would.

ncmncm 6y ago

A common problem is the microcontroller that powers up and needs randomness immediately, before it can initialize its peripherals and establish communication links. There, a billion or even a million bits per second may be enough. Now that we have packaged chiplets, we should be open to stuff on a different semiconductor process than our CPU cores. Where we need random bits as fast as they can be had is when running Monte Carlo processes on GPUs. There, the sources need to be on-chip because interconnect would be too expensive; we need a separate bit source integrated into each functional unit, and producing a whole machine word's worth each cycle, so a word is always ready immediately when it could be used. Sometime we need more than one before starting a computation.

fulafel 6y ago

There have boon some major problems with rdrand, on amd side at least (first the bulldozer/jaguar incident, and now the ryzen one).