Keyword is Efficiency and that is why I can not argue that ASIC’s is a bad idea, there is a magnitude of factors here including fair coin distribution which is very important to me as well as total coins in circulation.
Some people in Etheruem are talking about programmable proof of work to narrow the gap betwen ASICs and GPUs. This seems like a fiar solution for everyone. ASIC manufacturers can still make ASICs, but they won’t be 100x more efficient like we’re currently seeing.
I studied that and what I understood is “A custom ASIC to implement this algorithm is still possible” but “the efficiency gains available are minimal”
But what if an ASIC manufacturer builds a multi-ASIC rigs each ASIC designed for specified case of this programmed PoW?
in other word one ASIC is not efficient but a multi ASIC machine is still efficient compare to graphic cards for a programmed PoW.
You can design whatever ASIC (or multi-ASIC) efficient than a graphic-card in any case, just not in video gaming.
the result of this technique brings bigger machines and more centralisation.
This would not be the case at all, you can think of this as POW hashed per area of chip die; this cannot be developed more efficient than what is available on GPU’s chip die if POW exercises the tera flops of processing power of the GPU’s and equivalent ASIC will have to be developed with the same hashes per die area which will be at a comparable cost and performance to GPU so no one would have an advantage. Also GPU manufacturing process nodes are more advanced (more transistors per unit area) so you would in short limit ASIC to very specific farms at tremendous cost. This is not the same today where ASIC’s have a factor of x10 minimum improvement in POW processing compared to GPU’s, this would level the playing field.
While i’am not a very good tech expert i have some questions:
- While reading on gifthub about ProgPoW i have the feeling that it’s based on older and known Asic machines,
- This being said did someone try to simulate it what effect it would have on the Bitmain Sophon AI BM1680 processor?
- I could be totally wrong here, but i believe that the new asics generation for some algos will be based on exactly that Sophon chip. I received the other day a Bitmain B3 unit for the Tensority algo and i have a feeling that the Z9 and E3 will be based on a similar setup/design than the B3.
- For example the B3 has 3 boards with 4 chips, means 12 processors on it. From what i read each processor can/could handle up to 16GB of DDR4-2666 RAM and the very same processor is used on Bitmains GPU AI cards which seem to be as well high an interesting design.
So again, did someone test the Algo would be efficient IF the new Asics are based on the Sophon processor and all it’s possible setups and limits?