[A quick survey]
I've literally manipulated multiplication and division of hashes, which can be repeated any many times as desired. The more repeating, the lower hash rate, which also means less probable of GPU miner as I understand. Only deficiency is the less hash rate making hashing being hard. Let me know if you have any concerns or have something in mind.
I would personally suggest to continue the "repeating", to lower the probability of GPU mining being possible. Then, this would be a true "CPU-only" coin
My testing in my AMD cpu (8 core) shows 500 khps hash rate for pure M7; and it was down to half, when I add an extra fortran subroutines; and then repeats of multiplication and division lead to nearly tens of hps depressing per repeat. I feel not very good to keep repeating once we reach very small hash. Ref.: mining the scrypt coin with my AMD gives 10 khps.
Well, you should (not feel good about that) because this is just plain stupid... You are just making your coin hard to mine (but not especially on gpu...).
but it is your right to listen to moron (who will not mine the coin or with big botnet and big server) basically you are just shutting down the coin to many.
"Hard to mine" is relative to difficulty, so you will get the same mining reward regardless of how low your hashing speed is, as everyone has to do the same work, so difficulty will just be lower. So stop throwing accusations of stupidity when clearly you don't understand how mining works at all.
Of course this is also true of gpu miners. Joe, can you confirm that repeating the functions is making it DISPROPORTIONATELY harder for gpu over CPU? If so, then your target should be as many repetitions to ensure that an average gpu will always be slower than an average CPU. Thus you're not guaranteeing that this would be CPU only, but you would have ensured that mining with gpu would not adventagous. If it is not going to be disproportionately harder for gpu with every repetition, then it's pointless repeating it
It is very clear to me, that dev doesn't have a gpu miner to compare with... so all this is just guessing (based on an idea, m7, which has been already proven wrong in the first place).
So let add more divisions, more steps is (sorry to reiterate) just stupid because no comparison has been done.
So for the moment he is just decreasing cpu hashrate expecting gpu will decrease faster... (may-be you just learn a little bit more about gpu yourself)
Blind faith in action is nice to watch but it hurts any normal brains at the same time...
and some point, just stick with what have rather than changing every hours...
Hi djm, would you please take a look at the algo and then write your comments?! Please just don't attack a simple thought, sometimes simple thoughts could lead to something. I am unaware of any significant algo, please tell me which is a good one you have in mind since you don't believe in M7. I cannot compare with gpu, because no idea how to implement gpu in this case, that's why I ask you to take a look since you are more experienced than me in this regard. Simply let's see what's happening after launch, rather than wording game. If someone likes to implement gpu, let's see how he is doing.
p.s., the algo is just a portion of this coin project, not whole; no significant points being forged here, let's just move on.
I never say that I didn't believed in m7 (I am running a lot on it... with my gpus).
But the multiplication on that algo are a lot less hard than what was expected at first (and we arren't using any fast multiplication algo... just standard schoolbook multiplication algo with some part written in asm to speed up things...
(I assume nothing change since yesterday):
What I don't like in your algo is the fortran routine because calculation are made with real number and this may cause rounding problem accross platform (hopefully I am wrong) when the number are converted back to integer, meaning if the rounding is a bit different one one platform it could be impossible to validate the hash.... so it is a step I really don't like.
Anywhay the result of the spectral weight is multiplied to the 8th hash (which is the sum of the 7 previous hash).
I don't know the way it is done as you don't keep any number greater than u512 in your calculation, but that's ok. so why not
Then you multiply that by the Spectral weight then divide the result of the multiplication by the result of the previous operation. which should be roughly (I didn't check in detail) 38*64bit/512bit operation.
In that part the spectral weight seems to me useless (except as an attempt to slow down things, but by itself it has no purpose. The division between the product and hash[7] should have been enough. (and this part will slow the gpu more than the spectral stuff).
then sha256 of the resulting hash....
again the spectral stuff applied to the resulting hash (this I don't understand this at all..., it doesn't slow down anything, since it has been already calculated
then again the quotient of the fisrt division is divided again by spectral*hash of sha256 then re-sha256 and it is over...
I think it is long enough like that and I don't see the interest in bringing the spectral_weight at all (and even less using it a second time).
That's about it, and I am not sure I answered your question...