I'm not talking about hashes/s, but specifically, the amount of time required to perform a single SHA256 on the fastest computers known. How long does a very fast ASIC take to perform one SHA2? Is this evolving as time passes, like the hashrate, or is that value more or less constant, and only the hash/s increases with time?
1 Answer
Simply head over to https://en.bitcoin.it/wiki/Non-specialized_hardware_comparison#CPUs.2FAPUs and look it up.
For AMD, the columns nprocs
and Mhash/s
are important. Divide the latter by the former, sort by that value, and you find that A10-5800K
is the fastest with 26.25 Mhash/s/core
.
ARM is obviously not relevant. (So much for mining on a raspi.)
For Intel, the first number in the column p/t
or the first number times the number before the slash if there is multiplication and the column Mhash/s
is relevant. If I didn't mess it up, it's the Core i7 3930k
with 11.1 Mhash/s/core
(almost twice as fast as the second place).
So your answer is that a sha256 hash takes 3.8*10^-8 s
on the best-suited non-specialized CPU. What you want to do with this value is rather unclear, though. You might have to account for pipelining, instruction fetching, or even a memory access, as with all instructions.
The fastest miner is the AntMiner S9 with 14 Thash/s. ASICs do their work in parallel in a highly optimized way and computing a single sha256 hash is just not something they do, so it's not possible to state how long 1 sha256 hash takes them.
The hash rate of CPUs per core stays pretty much the same and therefore the time a single sha256 hash takes. Individual cores can't really get much faster so manufacturers integrate more cores into a single chip, increase the cache sizes, etc., to make CPUs better for tasks they're usually used for.
Note that every time I talked about a hash or a sha256 hash, I meant a sha256 of the amount of data a Bitcoin block header has. Obviously, sha256summing the downloaded English Wikipedia or an entire HDD takes a lot longer than this.
-
Good. I should've stated I'm specifically wondering how fast one could build a specialized repeated SHA3 (i.e.,
SHA3^N(x)
, for big Ns), in order to estimate a lower bound of time in which nobody could compute it for some N. This post is a great start. Thanks! Mar 5, 2017 at 14:59