This summer, Data Systems unveiled a large chip of wafer-grade. The world’s largest chip, with 1.2 trillion transistors, has found its meaning on a new system that accelerates deep learning. The CS-1 of Data Systems, Inc., made its debut at this year’s Supercomputing Conference. It is approximately 26 inches high and can be mounted in three in one rack. The machine has 20kW of power, of which 4kW is used for cooling.
(Instagram via TechSpot)
The new CS-1 provides 15kW of power to wafer-level chips (1kW lost due to inefficient power supplies).
It supports the Cerebras Wafer Scale engine, which is 56 times the largest GPU ever, 78 times the core, 3000 times the on-chip memory, and 33,000 times the bandwidth.
In other words, CS-1 delivers a computing experience that is too fast. In addition, it works with open source machine learning frameworks such as PyTorch and TensorFlow to increase application flexibility.
For more details about the hardware, such as clock rate, the company said it would share it with you in the near future.
(The size of a wafer chip, the same as an iPad tablet)
Such an amazing set of equipment, the price is obviously quite expensive. A company spokesman said the cost was in the millions of dollars.
But it’s not a big deal for customers who can afford it. The Agung National Laboratory, for example, has one that specializes in basic scientific experiments such as drugs that can be used for cancer research and treatment.