Amazon Web Services Unveils Nvidia GPU Instance

The new G2 instances will be aimed at organizations that need parallel computing capabilities for their cloud workloads.

Amazon Web Services is offering new instances to its public cloud that will leverage Nvidia's graphics technology and will be aimed at workloads that need high levels of parallel processing capabilities, such as video creation, visualization and streaming graphics-intensive applications.

The new Amazon Elastic Compute Cloud (EC2) G2 instances will use Nvidia's Grid GPUs, which come with 1,536 parallel processing cores and will enable developers to create applications and services leveraging OpenGL, CUDA, DirectX and OpenCL.

The G2 instances will complement Amazon Web Services' (AWS) compute clouds and the cluster GPU instances, which the company launched about two years ago. However, since then, organizations using AWS have been looking for more GPU capabilities, according to Matt Garman, vice president of Amazon EC2 for AWS.

"Many customers have asked for expanded functionality to extend the power of our GPU instances beyond HPC [high-performance computing] applications to graphics-intensive workloads such as video creation services, 3D visualizations and game streaming,” Garman said in a statement. “By enabling the use of DirectX and OpenGL, G2 instances allow developers to cost-effectively build scalable, fast 3D applications on Amazon EC2 and deliver high-performance 3D graphics using the cloud.”

Organizations running HPC environments have been turning to GPU accelerators from Nvidia and Advanced Micro Devices to increase the performance of their systems without ramping up the power consumption. Certain workloads, particularly those needing parallel processing capabilities, can be offloaded from the CPUs and onto the GPUs, which offer many more processing cores and better power efficiency. In parallel processing, workloads can be broken up and the various pieces can be processed at the same time before being brought back together. CPUs tend to complete one job before going onto the next.

Intel last year introduced its Xeon Phi coprocessors to compete with GPUs in the accelerator space. The x86-based chips, which have more than 60 cores, offer organizations the capabilities of GPU accelerators while also bringing the familiar x86 architecture and tools to programmers.

Officials with AMD and Nvidia have argued that any recoding needed to be done to run applications on the GPU accelerators is minimal.

Intel in June introduced two new Xeon Phi coprocessors and also talked about the upcoming Knights Landing, a 14-nanometer chip that will be able to be used as either a coprocessor or as a primary processor.

More organizations are using accelerators and coprocessors in their HPC environments. On the Top500 list of the world's fastest supercomputers, 54 use the technologies, including 39 that use GPU accelerators from Nvidia, 11 that use Xeon Phis and three that use AMD’s ATI Radeon GPU accelerators.

System OEMs are looking to leverage the growing interest in accelerators and coprocessors. Supercomputer maker Cray in October said it is bringing Intel Xeon Phi chips and Nvidia GPU accelerators to its XC30 systems, part of Cray's larger Adaptive Supercomputing initiative.

Now organizations looking to leverage Amazon's cloud will be able to take advantage of Nvidia's GPU technology. G2 instances at first will be available in the United States (in AWS facilities in Virginia, Northern California and Oregon) as well as in Ireland, according to the company. More G2 instances will be available in other regions in the coming months.

Customers can launch G2 instances with the AWS console, EC2 command-line interface, AWS software-development kits and third-party libraries, officials said.