Google has announced their new supercomputer lineup called the A3 Supercomputers. The A3 Supercomputer can easily take care of any activity that a Artificial Intelligence model can carry out, however demanding. Google pushes its focus more and more towards the AI models applications while keeping in mind the sheer number of users the have. Their new computers boast enough power to carry out commands of millions of user.
During their Google I/O event, Google mainly talked about various AI models and applications. Their showed their primary focus on their AI model, Bard and how it will feature a new language model for machine learning artificial intelligence. This will be an important base for other Google services like Maps or Photos.
Google AI and Nvidia
Google has also reported that these supercomputers use NVIDIA H100 GPUs for the main processing activities. The A3 will actually make use of 8 of these. The GPU features 4th Gen Intel Xeon Scalable memory boasting clocked 4800 DDR5 tech. The GPU also features Random Access memory of 2 terabytes. These GPUs also have 3.6 TBps bandwidth using NVLink 4.0 and NVSwitch. The Cards allows sharing of data at upto 200 GBps using DMA. The new Supercomputers are expected to be 10 times faster than the old A2 supercomputers.
Google reports that these GPUs are way more powerful than any other computer or processor in the market, boasting 26 exaflops of processing capability. The company claims that Computer is optimized basic and heavy machine learning activities by millions of users worldwide. The tech is mostly gonna be used by general public however, other companies are also interested in the tech. NVIDIA H100 GPUs are really expensive, costing nearly $10,000 USD. However Google claims they are still cheaper and faster than the competition. Other Business can apply for usage of A3 for their AI training in a preview. But the hype regarding will make it harder for everyone to get a stop for the preview.