We are thrilled to announce the availability of a robust infrastructure for our cloud desktop service.
For individuals without native access to a modern NVIDIA GPU, their best bet is often to execute GPU-intensive training tasks in the cloud. Paperspace offers a cloud-based service that provides instant access to a preconfigured Ubuntu 16.04 Linux desktop environment, fully equipped with a dedicated graphics processing unit (GPU). With the addition of this new capability, you can quickly and easily provision a prepared-to-use RStudio TensorFlow with GPU workstation in mere clicks. Preconfigured software program contains:
-
RStudio Desktop and RStudio Server
-
NVIDIA’s GPU libraries, including CUDA 8.0 and cuDNN 6.0.
-
TensorFlow v1.4 w/ GPU
-
The r, stringr, and tidyr packages.
-
The suite of packages (comprising ggplot2, dplyr, tidyr, readr, and numerous others).
Getting Began
To get started, first utilize the RSTUDIO Sign up and receive a $5 account credit with this exclusive promo code!
Newly designed Paperspace event: What’s Next for Innovative Architecture?
Select a Paperspace GPU case over the CPU case among various options. By selecting the P4000 machine type, we opt for a configuration that features an NVIDIA Quadro P4000 graphics processing unit.
For a comprehensive overview of how to get started, please visit the TensorFlow for R website.
Coaching a Convolutional MNIST Mannequin
The potential benefits of optimizing coaching processes for convolutional and recurrent neural networks on Graphics Processing Units (GPUs) could be significant. Let’s focus on coaching employees for our new Paperspace initiative.
Trained to optimize its performance over a period of 12 iterations, the process requires approximately one minute to complete, with each iteration taking around five seconds. Alternatively, training the same model on CPU alone on a high-end MacBook Pro takes around 15 minutes. (~ 75 seconds per epoch). By leveraging a Paperspace GPU, you can significantly enhance the performance of your mannequin training.
This mannequin was trained on a dataset that costs $0.40 per hour. To avoid accumulating unnecessary cloud costs when not actively using Paperspace, machines can be set up to automatically shut down after a period of inactivity.
Without access to a local NVIDIA GPU when coaching convolutional or recurrent networks, leveraging cloud-based GPU acceleration via Google Colab or AWS SageMaker can significantly boost training efficiency. You should use the RSTUDIO Get a $5 account credit when you sign up for Paperspace with this exclusive promo code!
Quotation
For attribution, please cite this work as: Smith et al. (2022). Title of the Work. Journal Name, Volume(Issue), pp. XX-YY. DOI: XXXXXXXX
Allaire (2018, April 2). Can You Train AI Models in Minutes? Discover the Power of Paperspace's GPU Cloud Workstations! Retrieved from https://blogs.rstudio.com/tensorflow/posts/2018-04-02-rstudio-gpu-paperspace/
BibTeX quotation
@misc{allaire2018gpu, author = {Allaire, J.J.}, title = {{RStudio Blog}: GPU Workstations in the Cloud with Paperspace}, doi = {}, url = {https://blogs.rstudio.com/tensorflow/posts/2018-04-02-rstudio-gpu-paperspace/}, year = {2018} } 
