Nvidia is one of the giants of artificial intelligence, but the competition is getting tighter and tighter, and resting on your laurels can be a loss of leadership when you least expect it. Cambricon Technologies claims to have a Cambricon-1A chip that is capable of standing up to the all-powerful Tesla V100.
Cambricon Technologies has developed the new Cambricon-1A chip focused on artificial intelligence. It is a silicon that is capable of reaching 166.4 TOPS of power when making simple calculations of 8-bit integers, sufficient for the needs of artificial intelligence. Best of all, it does so with a power consumption of only 110W. As for the Nvidia Tesla V100, it is capable of delivering 120 TFLOPS power in 8-bit decimal operations. In addition, the Cambricon-1A is capable of performing 83.2 TFLOPS in medium precision operations, an important advantage over the 30 TFLOPs achieved by Nvidia’s proposal.
This Cambricon-1A processor is manufactured by TSMC using 16 nm lithography and reaches a maximum clock frequency of 1.3 GHz. The core is accompanied by 16-32 GB of DDR4 memory at a frequency of 1600 MHz, much cheaper to produce than the HBM2 memory used by the Nvidia Tesla V100, which will have a direct impact on the final retail price.
The Cambricon-1A comes out in a PCI Express card format with a turbine heatsink, the company already has development kits that support Caffee, TensorFlow and MXNet, the most common development languages for artificial intelligence applications. This Cambricon-1A promises to be Nvidia’s biggest rival in the AI field.