Introduction to CPU, GPU, TPU, DPU, NPU, and BPU

Https://www.sohu.com/a/191538165_777155

 

The 5 trillion computation per second advertised by A12 is described as 5 tops in computer languages.

Qilin 970 NPU, according to the information is 1.92 tops.

Qilin 980 NPU, increased by 970 of 120%, that is, 1.92topsx2.2 = 4.23 tops.

A11 is about 0.6 tops. The official claim that 8 cores, that is, 0.6x8 = 4.8, is about 5 tops. There is no false publicity.

No NPU for other SOC is used for neural network computing using other GPUs, so no statistics are collected.

Gtx1080 is the computing power of fp32, that is, 32-bit floating point. NPU is generally a 16-bit floating point. Therefore, NPU is useless for playing games and can only be used for artificial intelligence such as CNN.

In conclusion, A12 is indeed the most powerful NPU, but it is not as powerful as the second kill 980.

 

Tengteng 910 is currently the chip with the highest computing density per chip. Its computing power is far greater than that of Google and NVIDIA. The computation capability of 910 semi-precision (fp16) is 256 tflops, the integer precision (int8) is 512 tops, and supports 128-channel full-HD video decoding (H.264) with a maximum power consumption of 264/265 W.

 

Https://www.leiphone.com/news/201704/bR0Ifw7coWSErOpP.html? _ T = T

Introduction to CPU, GPU, TPU, DPU, NPU, and BPU