Nvidia CEO Jen-Hsun Huang has introduced the Tesla P4 and Tesla P40 new GPUs at the GPU Technology Conference currently underway in Beijing.
According to TechMeme, Nvidia Corp. has unveiled two new graphics processing unit (GPU) chips that cater to deep learning, a fast-growing branch of artificial intelligence (AI). Details about the new Nvidia GPUs are presented in a press release published on company's website. The chips are based on Nvidia's deep learning AI platform based on Pascal architecture.
According to SiliconAngle, Nvidia had first entered officially the deep learning market in April when the company announced its 15-billion transistor Tesla P100 chip catering to deep learning applications. The Nvidia Tesla P100, P4 and P40 provide the only end-to-end deep learning platform for the data center. These advanced GPUs are helping to unlock the power of AI for a wide range of industries, according to Ian Buck, general manager of accelerated computing at Nvidia.
The Tesla P4 and P40 are designed specifically for inferencing, while the Tesla P100 focuses on training tasks. The newly launched P4 and P40 GPU's are able to recognize images, speech, or text using trained deep neural networks, in response to queries from devices and users. The new GPUs are four times faster than the GPU chips unveiled a year ago and have a response rate that is 45 times faster compared to CPUs, according to Nvidia.
The Tesla P4's comes with a low-power design, starting at 50 watts. Compared to CPUs, the low-power design of P4 makes the GPU 40 times more efficient in conducting inferencing in production workloads. The Tesla P40 offers maximum performance with 47 tera-operations per second (TOPS) of inference performance. Only eight Tesla P40 accelerators can replace 140 CPU servers, according to Nvidia.
According to Forbes, the two new GPUs unveiled by Nvidia come to support the company's efforts to grow its share in the artificial intelligence market. Already, Nvidia's GPUs have become the standard for training deep learning AI algorithms. Now Nvidia is focusing on the so-called "inference" part of deep learning.