AWS Releases P2 GPU Instance Type for AI, CAE and Big Data
Shawn Wasserman posted on October 13, 2016 |
Amazon Elastic Compute Cloud P2 GPU instance on the cloud.
Up to 16 GPU accelerators are available on the cloud for high computational engineering problems.

Up to 16 GPU accelerators are available on the cloud for high computational engineering problems.

Amazon Web Services (AWS) has just released a GPU instance type for its Amazon Elastic Computer Cloud (Amazon EC2). As a result, engineers can have access to up to 16 NVIDIA Tesla K80 GPU Accelerators over the cloud.

This should be great news for engineers working on computationally intensive applications such as computational fluid dynamics (CFD), finite element analysis (FEA), artificial intelligence and big data computations for the Internet of Things (IoT).

What makes this a big deal is that one of the biggest challenges for mid to small engineering businesses working with computer-aided engineering (CAE) tools and big data IoT is paying for the computational power they will need to crunch numbers.

Traditionally, this would fall on expensive in-house high performance computing (HPC) hardware that would price out many organizations. As a result, engineers in these smaller businesses have started moving to the cloud. By giving these cloud-based engineers access to GPUs, their computations should speed up considerably.

Using a P2 instance, engineers can deploy a compute cycle within the CUDA parallel computing platform or OpenCL without putting any money down on physical hardware. The biggest P2 instance available includes:

  • 16 NVIDIA Tesla K80 GPU Accelerators
  • Combined 192 GB of video memory
  • 40 000 parallel processing cores
  • 70 teraflops of single precision floating point performance
  • 23 teraflops of double precision floating point performance
  • GPUDirect technology to increase the bandwidth and low latency peer-to-peer communication
  • 732 GB of host memory
  • 64 Intel Xeon E5-2686 v4 CPUs
  • Network capacity for I/O
  • Amazon EC2 Elastic Network Adaptor

Engineers can also access three instance sizes if eight GPUs or one GPU will fit their application. The GPUs can also be launched using AWS’s Management Console, Command Line Interface (CLI) and software development kits (SDKs).

“Two years ago, we launched G2 instances to support customers running graphics and compute-intensive applications,” said Matt Garman, vice president of Amazon EC2. “Today, as customers embrace heavier GPU compute workloads such as artificial intelligence, high-performance computing and big data processing, they need even higher GPU performance than what was previously available.”

 “P2 instances offer seven times the computational capacity for single precision floating point calculations and 60 times more for double precision floating point calculations than the largest G2 instance,” added Garman.

Engineers utilizing the AWS cloud for simulations will definitely appreciate the added computational horsepower. “Simulation technology is at the core of Altair’s business,” said Stephen Cosgrove, director of CFD at Altair. “With our GPU solver partner FluiDyna GmbH, we’ve made significant investments in domain decomposition to optimize our computational fluid dynamics (CFD) software, nanoFluidX, for multi-GPU scaling for increased performance and reduced cost.”

He added, “We’re able to leverage the massive amount of aggregate GPU memory and double precision floating point performance in Amazon EC2 P2 instances to fit more simulations into a single node, significantly reduce customer simulation times, and reduce the cost of running large simulations.”

Access to these GPUs can also be quite beneficial to engineers using mathematical computation software like MATLAB. Silvina Grad-Freilich, senior product manager at MathWorks, explains, “MATLAB users moving their analytics and simulation workloads onto the AWS cloud require their analyses to be processed quickly. The massive parallel floating point performance of Amazon EC2 P2 instances, combined with up to 64 vCPUs and 732 GB host memory, will enable customers to realize results faster and process larger datasets than was previously possible.”

For more on GPU-optimized simulations, read: Simulation Sees Linear Processing Scalability to 55,000 Cores.

Recommended For You