NVIDIA Tesla P100 GPUs Now Available on Rescale’s ScaleX Platform

Rescale just added NVIDIA’s most advanced GPU yet, the Tesla P100, to the ScaleX big compute platform. Beginning today (September 13, 2017), all platform users are able to select the P100s as part of our standard batch workflow.

Fastest GPUs, Fastest Interconnect Always on Rescale

Rescale’s GPU platform now offers unique NVIDIA GPU-based systems with up to 8 NVLink-connected P100s and up to 8 PCIe-connected P100s. Furthermore, our latest P100 nodes are connected with EDR InfiniBand interconnect, facilitating high performance multi-node clusters. The specs for the new GPU core types are shown in the table below:
NVIDIA’s announced next-generation architecture, Volta, will be available early 2018 on Rescale.

Shorter Design Iterations for Accelerated Simulation

To allow Rescale users to take advantage of the new P100 core types, GPU-accelerated simulation software applications Sim4Life is available and pre-configured to run with flexible, on-demand licensing on Rescale’s big compute platform with ANSYS Mechanical, ANSYS HFSS, ANSYS Fluent and Abaqus 2017 Standard becoming available shortly after. Users can also use their own existing licenses on the new P100 GPUs on Rescale.

High-Performance Training for Deep Learning

The new P100 core types are also optimized to run the latest versions of CUDA-enabled deep learning frameworks like TensorFlow, Keras, PyTorch, and Caffe. Optimize your search for the best-performing models by training faster and in parallel on more P100s. Our new NVLinked P100 systems offer performance comparable to the DGX-1 on frameworks like TensorFlow, as shown below:

NVIDIA P100 benchmark Inception V3
P100 benchmark against DGX-1 on Inception V3 model in TensorFlow
nvidia p100 benchmark on resnet-152 model in tensorflow
P100 benchmark against DGX-1 on ResNet-152 model in TensorFlow

Start Running on P100s Today

To get you started running NVIDIA P100s today, below are a couple sample jobs you can clone and run to test our new hardware. For instructions on how to clone a job, click here. You will of course need a Rescale account to do so—you can sign up for one here.

TensorFlow InceptionV3 benchmark

TensorFlow is known for providing high-performance model training. Here are some benchmark results comparing Rescale’s NVLinked P100 system with the high-performance DGX-1 deep learning server. We see that Rescale can achieve comparable performance to high-end on-premise GPU servers.
Click here to clone the InceptionV3 4xP100 job.

PyTorch DCGAN example

The link below is an example of training a Deep Convolutional Generative Adversarial Network (DCGAN), which generates realistic fake images that are similar to the input training images. This example is trained using Rescale’s NVLinked P100 systems on the LSUN bedroom image dataset.
Click here to clone the PyTorch DCGAN job.

Deep Learning Kickstart Program

In conjunction with the P100 general availability, we are launching the Deep Learning Kickstart Program, which will provide $1,000-5,000 in Rescale hardware credits for P100 use to approved applicants. All Rescale users representing a company will be eligible to apply after attending a kickoff webinar that will give additional details on Rescale’s Deep Learning Kickstart Program and demonstrate how to use the NVIDIA P100s to accelerate deep learning on the cloud. The webinar will be held on September 21, 2017 at 8:00 am PDT /  11:00 am EDT.