Nvidia officials want to bring GPU acceleration to hyperscale environments to help drive the development of machine learning and artificial intelligence.
The company today is unveiling two new Tesla GPU accelerators that enable researchers to more quickly create deep neural networks used in deep learning—also known as “machine learning”—and to deploy these networks across data centers. Nvidia also is rolling out software for hyperscale environments to support the workloads that are processed through the GPUs.
The new products are part of a larger push by Nvidia to become a significant player in the development of neural networks, machine learning and artificial intelligence. At the company’s GPU Technology Conference 2015 in March, Nvidia founder and CEO Jen-Hsun Huang introduced a number of products for developers and researchers working on deep learning technologies, and told attendees that deep learning is a central part of the vendor’s strategy going forward.
“The topic of deep learning is probably as exciting an issue as any in this industry,” Huang said at the show.
The CEO and other executives will be in San Francisco today talking about the company’s strategy around artificial intelligence and machine learning. Machine learning—essentially giving compute systems the ability to learn over time by using a growing database of information and neural networks—is being used in a broad range of scenarios, from making voice recognition more accurate to identifying objects in videos and photos to enable faster tagging for searches, driving services that can identify individual interests and responding to voice commands made in a conversational tone.
Nvidia for many years has been making inroads into the high-performance computing (HPC) and scientific computing spaces through the development of its Tesla GPU accelerators, which enable organizations to increase the performance of their systems while holding down the power consumption. This is made possible by offloading some of the workloads from the CPUs to the GPUs, which offer parallel computing capabilities and run hundreds of more cores.
An increasing number of supercomputers on the twice-yearly Top500 list of the world’s fastest systems use GPU accelerators from Nvidia or Advanced Micro Devices or x86-based Xeon Phi co-processors from Intel. In the most recent list released in June, there were 90 systems using such accelerators, up from 75 in November 2014. Of those, 52 use Nvidia GPUs. The next Top500 list is due out at the SC 15 supercomputing show, which begins Nov. 15 in Austin, Texas.
Now Nvidia is bringing GPU acceleration to hyperscale data centers, such as those run by Web-scale companies like Facebook, Google and Baidu. The challenge for these environments is the massive amounts of data being created, according to Ian Buck, vice president of accelerated computing at Nvidia.
“Today, there are exabytes of data being created by [hyperscale] companies daily,” Buck told eWEEK, noting that much of it is driven by user-created content.
For example, 4 billion videos are viewed every day on Facebook, a 400 percent increase in six months, he said. Baidu handles 6 billion queries a day—10 percent of which use speech—while YouTube users watch 300 hours of videos per minute, half the time on their mobile devices. The challenge is making sense of the data—sharing the millions of images and videos with millions of people, tagging and enhancing each second of video as close to real time as possible, discovering and analyzing the content, and instantly creating relevant advertising.
“This is a huge computational challenge,” Buck said.
It also increasingly will involve artificial intelligence, and the new Tesla GPU accelerators are designed to help organizations drive innovation in the area. The Tesla HyperScale Accelerator line will give users a 10-fold improvement in performance, according to CEO Huang.
“Machine learning is unquestionably one of the most important developments in computing today, on the scale of the PC, the internet and cloud computing,” he said in a statement. “Industries ranging from consumer cloud services, automotive and health care are being revolutionized as we speak. Machine learning is the grand computational challenge of our generation. [With the new Tesla hyperscale GPUs] the time and cost savings to data centers will be significant.”
Nvidia Unveils GPU Accelerators for Hyperscale Data Centers
Included in the Tesla Hyperscale Accelerator line is the Tesla M40 GPU, which is optimized for machine learning and reduces training time for systems by eight times compared with systems running only CPUs. A typical AlexNet training process takes 10 days for a CPU-only system, but 1.2 days for one that is accelerated, Buck said. It offers scale-out performance through its support of Nvidia’s GPUDirect technology that helps speed up multi-node neural network training. The Tesla M40 offers 3,072 cores, 12GB of GDDR5 memory and 288 Gb/s bandwidth, all in a 250 watt power envelope. It offers a peak performance of 7 teraflops.
The Tesla M4 is a low-power GPU built for hyperscale environments that helps run the trained models in the data center and is optimized for Web service applications, such as video trans-coding, image and video processing and machine learning inference. It can trans-code, enhance and analyze up to five times more simultaneous video streams than CPUs, consumes 50 to 75 watts of power, and offers up to 10 times the power efficiency of a CPU for video processing and machine learning algorithms, according to Nvidia officials. Its small size fits into the enclosure designs for hyperscale data center systems.
The Tesla M4 holds 1,024 cores, 4GB of GDDR5 memory and 88 Gb/s bandwidth, with a peak performance of 2.2 TFLOPs.
The Nvidia Hyperscale Suite of software tools include cuDNN, a popular algorithm software for processing deep neural networks used for artificial intelligence applications, GPU-accelerated FFmpeg multimedia software to accelerate video trans-coding and processing, GPU REST Engine to easily create and deploy high-throughput, low-latency Web services, and Image Compute Engine service with REST APIs that enables image re-sizing five times faster than a CPU.
The Tesla M40 and the software suite will be available later this year, with the Tesla M4 coming in the first quarter 2016.