Nvidia has announced its two new Tesla GPU accelerators, the Tesla M40 and Tesla M4, as part of the hyperscale accelerator line of offerings that is targeted at accelerating the massive machine learning workloads of web services companies. The hyperscale line is accompanied by the Hyperscale Suite, which consists of GPU-accelerated libraries.
The Tesla M40 GPU accelerator, which is capable of up to 7 teraflops of single precision performance using its 3,072 Cuda cores, 12GB GDDR5 VRAM and 288 Gbps bandwidth, is aimed at enabling researchers to innovate and design "new deep neural networks" of applications that can make use of artificial intelligence.
On the other hand, the Tesla M4 GPU, which is fitted with 1,024 Nvidia Cuda cores, 4GB GGR5 memory and 88 Gbps bandwidth, is capable of up to 2.2 teraflops. It also only needs 50 to 75 watts to function, compared to the M40's 250 watts. The lower power consumption makes it ideal for deploying the neural networks in data centers.
"Machine learning is unquestionably one of the most important developments in computing today, on the scale of the PC, the Internet and cloud computing," says Nvidia's CEO and cofounder Jen-Hsun Huang. "Industries ranging from consumer cloud services, automotive and healthcare are being revolutionized as we speak."
Huang says deep learning is the "grand computational challenge" of the next generation, and the Tesla hyperscale GPU accelerator line is designed to give machine learning a tenfold boost. A graph by Nvidia shows that the Tesla M40 reduces training time from a CPU's eight days down to the accelerator's 1.2 days.
The Hyperscale Suite included in the hyperscale line up lists cuDNN, which is a popular algorithm for accelerating video processing and transcoding, a GPU-accelerated FFmpeg multimedia software, Nvidia GPU REST Engine for easier deployment and creation of web services and Nvidia's Image Compute Engine, which resizes images five times quicker than a CPU.
Both the Hyperscale Suite and the Tesla M40 GPU accelerator will come out before 2015 ends, while the Tesla M4 GPU accelerator will be available in Q1 2016.