The lucrative world of proof-of-work altcoin mining has been eye-opening, yielding staggering returns for miners…until Ethereum took a detour towards proof-of-stake. But one emerging industry that has caught the attention of those who have found hundreds of graphics cards unused after consolidation is AI.
This newfound technology darling, facilitated by the launch of ChatGPT, requires computations on vast amounts of data similar to Ethereum mining, and the best way to do such heavy lifting is GPUs. to use it in large quantities.
However, this may not be the panacea many expected. Because GPUs, the credible workhorses of cryptocurrency mining, are not necessarily MVPs in the AI training arena. why? Let’s try to extract this complex problem to its essence.
A tale of two needs
Here is the crux. Cryptocurrency mining and AI training have different demands for a perfect GPU. It’s like dating two people with completely different interests. Proof-of-Work cryptocurrencies are good for cards with a lot of hash power, but vRAM (Video Random Access Memory) cannot be ignored. AI, by contrast, has a penchant for cards with plenty of vRAM, so it disdains hash power.
Consider hash power to be an important part of the computation. It all has to do with how many computations the GPU can do per second for him. The higher the hashrate, the higher the chances of unlocking that elusive block and immersing yourself in cryptocurrency riches. This is like a speed dating scenario. The more people you meet (calculations are done), the more likely you are to find a match (blocks are mined).
Conversely, vRAM is the capacity to process and store huge amounts of data simultaneously. It’s the difference between him juggling 2 balls or 20 balls. AI training is notoriously data overload, requiring GPUs to simultaneously process and process vast amounts of data. Using the dating example, going the vRAM route is like inviting a crowd into a room, asking the same question (data points) at the same time, and finding an exact match (AI model).
And here is the contradiction. A powerful GPU that loves hash power is great for mining Ethereum, but it’s breathtaking when it comes to training AI. It’s like asking a world-class sprinter to run a marathon. It’s not their game.
Do you have a good backup plan?
Now, oversimplification aside, cryptocurrency mining certainly requires high hashpower, but it doesn’t completely leave vRAM aside. Crypto mining utilizes vRAM as part of the mining algorithm, but vRAM requirements are typically lower than those required for AI training. For example, a good GPU for crypto mining he might have 4GB of vRAM, which runs AI image generators like Stable Diffusion properly (don’t even think about training them) Not suitable for
To sort things out, one of the best GPUs for ETH mining was the Nvidia RTX 3060Ti with 60MH/s hash power and 8GB vRAM. In contrast, OpenAI used his Nvidia’s A100 and V100 models. The V100 comes with 32GB of vRAM and the A100 model handles up to 80GB.
However, this does not mean that GPUs used for cryptocurrency mining are completely redundant for AI training. GPUs with relatively little vRAM can still be used for training small AI models and tasks that don’t require a lot of vRAM.
“There are many AI operations and models that work quickly and efficiently on traditional cards,” said Scott Norris, CEO and founder of Optiminer. Decryption, added that entrepreneurs should properly research which AI applications to focus on. The GPUs used for cryptocurrency mining can run “AI farms with either custom models or slightly tuned models,” he said.
Norris said some cryptocurrency mining companies have already switched to AI operations. Omega AI is one such example. Hive Blockchain and Hut8 Mining are also trying their luck.
So while this solution may not be as lucrative as good old cryptocurrency mining, if the “plan B” of mining alternative proof-of-work altcoins seems bleak, it’s worth doing. It could act as a possible “Plan C”.
Want to learn more about AI? Check out our latest Decrypt U course, Getting Started with AI. We cover everything from the history of AI to machine learning, ChatGPT, ChainGPT, and more. Learn more about.
