The profitable world of proof-of-work altcoin mining was a spectacle to behold, rewarding miners with staggering earnings… till Ethereum took a detour in direction of proof of stake. However an rising business is grabbing the attention of those that discovered themselves with a whole lot of idle graphics playing cards after the merge: AI.
This newfound tech darling, spurred by the inception of ChatGPT, requires the computation of huge portions of information, akin to Ethereum mining, and one of the best ways to do such heavy lifting is with GPUs—numerous them.
Nonetheless, this won’t be the panacea many had hoped for as a result of GPUs, the trusty workhorses of cryptocurrency mining, aren’t precisely the MVPs within the AI coaching enviornment. Why? Let’s distill this complicated subject to its essence.
A story of two wants
Here is the crux: Crypto mining and AI coaching have divergent appetites for his or her excellent GPUs. It is like courting two individuals with wildly completely different tastes. Proof-of-work cryptocurrencies have a factor for playing cards with tons of hash energy, however could not care much less about vRAM (Video Random Entry Reminiscence). In distinction, AI is a sucker for playing cards with plentiful vRAM and offers a chilly shoulder to hash energy.
Image hash energy because the muscle of the operation—it is all in regards to the variety of computations your GPU can crunch per second. The upper the hash charge, the higher your odds of unlocking that elusive block and basking within the crypto fortune. It is like a velocity courting situation: the extra individuals you meet (calculations you make), the upper your probabilities of discovering a match (mining a block).
Conversely, vRAM is the capability for dealing with and storing huge volumes of information concurrently. It is the distinction between juggling two balls or twenty. AI coaching is a infamous knowledge glutton, demanding GPUs to deal with and course of colossal quantities of information concurrently. Utilizing the courting instance, going the vRAM route could be like inviting a crowd right into a room and posing the identical questions (knowledge factors) concurrently to seek out the right match (AI mannequin).
And right here lies the paradox. Your buff, hash-power-loving GPU, excellent for mining Ethereum, is left gasping in terms of coaching AI. It is like asking a world-class sprinter to compete in a marathon—it is simply not their recreation.
A great backup plan?
Now, setting oversimplifications apart, crypto mining does certainly require excessive hash energy, however it would not solely sideline vRAM. Crypto mining makes use of vRAM as a part of its mining algorithm, although the vRAM necessities are sometimes decrease in comparison with these wanted for AI coaching. As an example, a reliable GPU for crypto mining would possibly sport 4GB of vRAM, which is unsuitable to even run (not even take into consideration practice) an AI picture generator like Steady Diffusion correctly.
To place issues in context, one of many best GPUs for ETH mining was the Nvidia RTX 3060Ti with 60MH/s of hashpower and 8GB of vRAM. In distinction, OpenAI used Nvidia’s A100 and V100 fashions: A V100 comes with 32GB of vRAM and the A100 fashions deal with as much as 80GB.
However this does not indicate that GPUs used for crypto mining are solely redundant for AI coaching. GPUs with comparatively decrease vRAM can nonetheless be employed for coaching smaller AI fashions or duties that do not require important vRAM.
“There are various AI operations and fashions that operate shortly and effectively on legacy playing cards,” Scott Norris, CEO and founding father of Optiminer, informed Decrypt, including that entrepreneurs must correctly examine which AI software they concentrate on. GPUs used for crypto mining can work in “AI farms with both customized fashions or barely adjusted fashions,” he mentioned.
Norris talked about that some crypto mining firms have already made the change to AI operations. Omega AI is one instance. Hive Blockchain and Hut8 Mining are additionally testing their luck.
So, whereas this resolution is probably not as profitable as your good, outdated crypto mining, it might function a viable “plan C”—in case your “plan B” of mining alternate proof-of-work altcoins appears bleak.
Curious about studying extra about AI? Try our newest Decrypt U course, “Getting Began with AI.” It covers every thing from the historical past of AI to machine studying, ChatGPT, ChainGPT, and extra. Discover out extra here.