Fastest computing challenge, NVIDIA released its fastest GPU, Tesla T4
Today, the artificial intelligence computing company “NVIDIA” released the latest GPU product – Tesla T4 (T stands for Nvidia’s new Turing architecture).
“NVIDIA” Tesla T4″
The new product introduced this time is mainly used for machine learning and data reasoning, and has revolutionised the widely used P4 graphics processor.
Currently, almost all large cloud providers are able to offer P4 GPUs, and NVIDIA wants to push technology to the next generation. The Tesla T4 is configured with a total of 320 Turing tensor cores and 2,560 CUDA cores.
“NVIDIA” said that Google will be the first company to use T4 GPU, which will be pushed to Google’s cloud platform. The NVIDIA Tesla T4 is significantly faster than the P4.
For example, for language reasoning, T4 is 34 times faster than using a CPU and 3.5 times faster than using a P4.
Most importantly, Tesla T4 is designed by NVIDIA specifically for artificial intelligence reasoning . Ian Buck, vice president of the company and vice president of data center operations, said that Tesla T4 is able to perform artificial intelligence reasoning so efficiently because of its new Turing tensor core.
“NVIDIA” CEO Jen-Hsun Huang talked about tensor core, said it can not only serve the game, rendering, also can be used for computer reasoning.
In addition to this new chip, NVIDIA has introduced a TensorRT software for optimizing deep learning modes. TensorRT Inference Server, a fully integrated data reasoning service that seamlessly plugs into existing Kubernetes facilities.
Tesla T4 computing power
“NVIDIA” is working with Microsoft on the use of “NVIDIA” chips to develop AI features such as voice and face recognition in Microsoft’s Cortana and Bing.
According to foreign media reports, Huawei may compete with Nvidia. Microsoft is discussing cooperation with Huawei and considering using Huawei’s newly developed artificial intelligence chips in Microsoft China Data Center.
D?NN device identification text
Looking beyond, an interesting technological advance is that the future technology that might replace the GPU is also coming. Recently, scientific researchers have developed a technique for deep learning using the speed of light, the Diffraction Depth Neural Network (D?NN). The technology was invented by researchers at the University of California, Los Angeles.
They used 3D printing to create an “all-optical” artificial neural network that can analyse large amounts of data and identify targets at the speed of light with high accuracy and low network construction costs. 50 U.S. dollars. This technology was released in the important scientific journal Science.