pull down to refresh
61 sats \ 13 replies \ @south_korea_ln 17 Jul \ parent \ on: Struggling bitcoin miners seek deals with AI companies bitcoin
I'm not sure I understand the details.
At some point, they talk about GPUs. GPUs are worthless for Bitcoin mining, right? And what about the usual ASICs? Can they be used to train AI models?
My understanding is that while the ASICs are worthless for AI, what can be reused is the rest of the infrastructure - the warehouse facilities, the wiring, cooling systems etc. and the maintenance staff.
reply
reply
Oh got it, that makes sense.
reply
Here is a published paper over the area of AI and ASICS
https://www.sciencedirect.com/science/article/abs/pii/S0141933122000163
An here are a couple of companies that deploy ASIC AI accelerators
reply
Yes, but those are not SHA-256 ASICs.
reply
Sure maybe it isnt exactly hashed out but considering SHA-256 is for validating data authenticity, identity, and integrity this type of security could easily be incorporated in the AI space. Areas like healthcare for example need strong encryption so while some tweaks of course would be needed it isn't going to mean the whole machine just needs to be tossed away
reply
The SHA-256 ASICs do hashing, they don't encrypt.
The energy-intensive part of AI is linear algebra, things like matrix multiplication, convolution, normalization etc. So I don't see much use for SHA-256 ASICs there.
reply
Considering one of the top reasons for SHA-256 is data integrity right there you have a core, fundamental, and critical issue facing AI. Not to mention we don't want the facts or history of the world to be tampered with as we are starting to live in a world where the truth is hard to come by. Hashing addresses it and is a reason why putting AI data and data sets on blockchain is being explored right now.
reply
I could write an essay on this because there seems to be a divide between BTC advocates and the AI industry. It all boils down to the AI application. With instances where memory is important like Deep Learning also known as Nural Network architecture ASICs could perform up to 10x better than GPUs.
AI itself is undergoing a huge change right now in the accelerator area and GPUs are going to move from being a basic thing you need for AI to an application-specific thing just like ASICs are.
reply
Please do write an essay on this topic ;)
For now, have a few sats already.
reply
I just threw some links up above that cover it! But here it is as well!
Here is a published paper on the area of AI and ASICS
And here are a couple of companies that deploy ASIC AI accelerators
reply
Great stuff, thanks! I feel like this was bound to happen
reply
I think of it as while some of the hardware might not be perfect or optimal you can make software changes and adapt the AI to still leverage the technology at hand. Semiconductors will always be in need so while miners may before less efficient every few years the hardware within them can be adapted for other uses or just straight up recycled.
reply