How Many Atoms Are in the World? Part 2)4 Feb 2024 16:31
Part 2)
So. In Part 1) we blew the budget on £25Million of storage as just under a tonne of 4TB 990 PRO PCIe 4.0 NVMe M.2 SSDs which gives us 400,000TB of storage. It takes us 10 minutes to fill up each one but let's say we have an aditional budget to buy 1,000 Bitmain T21s. Shut up at the back with your why didn't we do that and mine our own bitcoin and stop ticking up your own posts!! A T21 costs $3,185.00 or £2,500. That's only another £2.5Million so no worries. Now we can generate our user database in 1,000 minutes. Wow, that's just under 16 Hours. That's quite reasonable.
Great. Now we have our database. All quite possible after all. Shut up at the back. We would have rented some big buildings and landfill plus solved the employment problems for a small country!!!.
Where was I? Oh yes we have our 400,000TB database. Thats 200,000TB for inputs and 200,000TB for outputs. Each input/output requires 256 bits or 32 bytes, I thought it was 64 but the numbers would still be silly. So we can store 6.25 x 10^16 solutions and train our neural networks on them. Ooops, we only have one nearal network to write to and the time it takes for them to swallow and resolve an input/output solution, I believe the term here is an epoch, will be quite slow? Let's say they can do it at the same speed as our SSD's. That will only take 9 x 10^6 seconds or 17 years.
Our database now has 6.25 x 10^16 solutions of the possible 1 x 10^77 solutions. In SNR terms that's 20log(6.25E16/1E77) or -1204dB. For the non-technical it's quite hard to get a handle on what that number means but let's throw the x10 improvement in in a non scientific way and call it -1021dB.
Interlude...
https://www.youtube.com/watch?v=38mE6ba3qj8
Elsewhere to get a better idea,
https://lisa.nasa.gov/
"LISA is Extremely High Precision
These signals are extremely small and require a very sensitive instrument to detect. For example, LISA aims to measure relative shifts in position that are less than the diameter of a helium nucleus over a distance of a million miles, or in technical terms: a strain of 1 part in 10 x 10^20 at frequencies of about a millihertz."
That's golden cubes, not balls, in spacecraft at three Lagrange Points One Million Miles apart operating with signals at -400dB. That is 621dB worse, easier, than what QBT is claiming to do. That's 1.12 x 10^31 times easier.
Of course if you want to take a punt on those numbers you might want to ask yourself "Do you feel lucky Punk?"