Why Your Cryptocurrency Mining Chips Won't Cut It for AI - Explained

Why Your Cryptocurrency Mining Chips Won’t Cut It for AI – Explained

Updated: 13 May 2023

Disclosure

Don’t invest unless prepared to lose all the money you invest. This is a high-risk investment, you shouldn’t expect to be protected if something goes wrong.

Join Our Telegram channel to stay up to date on breaking news coverage

The world of proof-of-work altcoin mining was once captivating, generously rewarding miners with astonishing profits.

Ethereum’s shift towards proof of stake marked a turning point, leaving those with many idle graphics cards searching for new opportunities. Fortunately, a burgeoning industry has captured the attention of these individuals: artificial intelligence.

The advent of ChatGPT and other similar technologies has given rise to a demand for extensive data computation comparable to the requirements of Ethereum mining. To efficiently handle such intensive tasks, the utilization of a multitude of GPUs has emerged as the optimal solution.

The Divergent Needs of Crypto Mining and AI Training for GPUs

The crux of the issue is the contrasting requirements of ideal GPUs for crypto mining and AI training, which can be likened to dating two individuals with different preferences. Proof-of-work cryptocurrencies prioritize GPUs with high hash power, determining the number of calculations a GPU can perform per second.

This hash rate increases the chances of successfully mining a block, similar to speed dating, where meeting more people (performing more calculations) improves the odds of finding a match (mining a block).

AI training relies on GPUs with ample vRAM capacity, allowing them to handle and store large volumes of data simultaneously. Furthermore, AI training needs GPUs that can process significant amounts of data concurrently. In the dating analogy, choosing vRAM would involve inviting a crowd into a room and asking them all the same questions (data points) simultaneously to find the perfect match (AI model).

This creates a paradox: GPUs optimized for hash power, ideal for mining Ethereum, need help with AI training. It’s like asking a world-class sprinter to participate in a marathon—it’s not their forte.

Reimagining GPUs from Crypto Mining for AI Tasks

Cryptocurrency mining and AI training have different requirements regarding GPU specifications, but that doesn’t mean GPUs used for mining are entirely useless for AI tasks. While crypto mining heavily relies on high hash power, it uses vRAM as part of its mining algorithm, albeit with lower requirements than AI training.

For example, a competent GPU for crypto mining may have around 4GB of vRAM, which is inadequate for running AI image generators like Stable Diffusion for training. In contrast, AI training conducted by OpenAI employs GPUs such as Nvidia’s A100 and V100 models, which offer significantly higher vRAM capacities – 32GB and up to 80GB, respectively.

However, GPUs used for crypto mining can still find utility in AI training scenarios that involve smaller models or tasks that don’t demand extensive vRAM. Scott Norris, CEO and founder of Optiminer, explains that legacy GPUs can efficiently handle many AI operations and models. Entrepreneurs can leverage these GPUs in AI farms by focusing on specific AI applications, either with custom models or slightly adjusted ones.

Some crypto mining companies have already started transitioning to AI operations. Omega AI, Hive Blockchain, and Hut8 Mining are among those exploring this avenue. While it may not be as lucrative as traditional crypto mining, incorporating GPUs from mining rigs into AI-related tasks could be a viable “plan C” if alternate proof-of-work altcoin mining appears unfeasible.

In summary, while crypto mining and AI training have distinct GPU requirements, GPUs used for mining can still be repurposed for AI tasks that don’t demand high vRAM capacities. This provides an alternative avenue for mining companies looking to diversify their operations and adapt to changing market conditions.