This new chip is perfect for AI
Artificial intelligence is a significant challenge for conventional computing architecture. Standard models separate memory storage from computing, so data has to be moved from the storage area to the CPU or GPU.
This design has a problem: it takes too much time to move. It takes too much time. The \”memory barrier\” or \”bottleneck\” is a problem that can limit the performance of the most powerful processor unit.
Latency cannot be avoided when computing is superior to memory transfer. When dealing with huge amounts of data, essential for AI and machine learning applications, these delays can become a serious problem.