<!DOCTYPE html><html><head><title></title><style type="text/css">
p.MsoNormal,p.MsoNoSpacing{margin:0}</style></head><body><div>This is part of the problem with these things; Linux /dev/random has changed and improved a lot over the last 10 years or so, and it might not make much sense to use that as a positive use-case for any of these devices. General OS security (ASLR etc) needs very little entropy; TLS/SSH/VPN implementations might need more, but you can't get valid entropy estimates out of the kernel any more (and arguably never could anyway).<br></div><div><br></div><div>For Windows, we never really figured out how to introduce entropy to the OS - given that later versions seem to require TPM, I expect they get 100% of their entropy from the PC hardware. Not sure how that scales through VM hypervisors of course.</div><div><br></div><div>Then for stats and maths, PRNG is probably better (if you can control the seeding, you can provide reproducibility in your models).<br></div><div><br></div><div>So, what's left? What do people really want to use hardware RNGs for?</div><div id="sig67082176"><div class="signature">-- <br></div><div class="signature"> Jim Cheetham<br></div><div class="signature"> <a href="mailto:jim@cheetham.nz">jim@cheetham.nz</a><br></div><div class="signature"><br></div></div><div><br></div><div>On Sat, 3 Feb 2024, at 8:06 AM, Ethan Spoelstra wrote:<br></div><blockquote type="cite" id="qt" style=""><div>I was actually going to mention that one as I've picked up a few, but I've discovered I don't do enough that needs TRNG to actually put them to use.<br></div></blockquote><div><br></div></body></html>