Are ASIC Miners the Future of Machine Learning?
Artificial intelligence has gradually and subtly become part of our everyday lives. Like many forms of computerized tech, it began its genesis in films and then gradually slipped into our living rooms, cars and pockets. The world as we know it would literally cease to exist if AI were taken away, and even some of the most remote parts of the earth are at least somewhat dependent on AI.
At the same time, currency is undergoing a dissimilar, yet somewhat parallel shift. The same way AI is transforming what a computer means, cryptocurrencies are transforming what money means. In a world of spiraling inflation, bank bailouts paid for by taxpayers and battles over regulation that have no end in sight, cryptocurrencies have poised themselves to swoop in and save the day. The powerful computers used to solve the blockchains of cryptocurrencies are being touted as possibly being the right tools to further advance the AI universe. One company, Bitmain, is considering this step simply in order to continue it’s healthy expansion. Along the way, boundaries are going to be pushed and horizons are going to have to shift.
In some ways, the ship has already sailed. Bitmain has been dipping its toe in the AI waters since 2015, using their Sophon division, which focuses on the development of AI chips. The declared mission of the Sophon division is to “solve all the puzzles in the universe.” It’s a bold claim, but one that’s based in relatively realizable concepts. They seek to do this using application-specific integrated circuits, or ASICs, a new kind of computer chip.
ASICs vs. CPUs and GPUs
Your average computer chip is an extremely versatile workhorse. It can process a vast variety of computations for a wide swath of applications. These chips are typically CPUs and GPUs. ASIC silicon chips are different. They are designed for a specific task, not general use. One of these specific uses is the computation load of solving extremely complex mathematical problems in order to mine cryptocurrency.
To use a car analogy, ASICs are like Formula 1 cars and CPUs and GPUs are like your average sedan. You’re not going to take a Formula 1 car on a rough and bumpy dirt road because it isn’t designed for that. At the same time, your average sedan could do a great job on that bumpy dirt road but would be completely ineffective on an F1 racetrack. ASICs can do an amazing job with special applications that they are designed to run, and this results in not only time savings but energy savings as well because they can be fine-tuned to use less energy while performing the same specific task again and again.
Bitmain’s First Move
Bitmain has already release an ASIC designed to target AI. Its aim is to enhance tensor computing acceleration. It’s called the Sophon BM1680. Its job is to tackle deep learning and interact with neural networks. Deep learning is a type of AI that involves a computer going deeper and deeper into studying details of various things in order to categorize them and either make decisions or provide decision makers with previously inaccessible options.
The Sophon BM1680 is similar to Google’s TPU, which stands for tensor processing unit. It’s designed to work within Google’s open-source machine learning environment called TensorFlow.
Can Bitmain Do It?
But in order to be the future of AI, ASICs will have to compete with some pretty big dogs. Google may be grabbing a lot of the spotlight, but NVIDIA and Intel are quietly making impressive forays into the development of ASICs for AI as well. The advantage these firms have is their long history of success in the computer chip sector as well as deep pockets that power impressive recruitment machines. The talent needed to develop an ASIC for AI is not easy to come by. Companies with history and leverage similar to NVIDIA and Intel are going to have a significant advantage over newer players like Bitmain.
If Not Bitmain, Who?
Given the power of ASICs, the AI world is a natural next step. While Bitmain has extensive experience developing ASICs for mining cryptocurrencies, the jump to being a major player in the AI world may be too big of a leap. But even if that is true, that doesn’t mean others are not going to be able to develop ASICs that can take over the AI development space. In addition to Intel and NVIDIA, other players like Ebang and Canaan Creative are knocking on the AI doors as well.
In the end, it’s going to be a question of acquiring the right talent and directing that talent appropriately. At this point, with its fins already wet by the waters of ASIC for AI, Bitmain may end up being a bigger fish than the naysayers expect.