What is the machine RStudio Cloud runs on?

keras

#1

I want to use keras in RStudio Cloud, and I need to decide which version of Tensorflow to install. What is the machine in which RStudio Cloud runs? It's clearly a 64 bit Linux machine, and I know from other threads that it has 1 Gb RAM and 3 GB HD space. However, I'd like to know:

  • is a GPU available (don't think so)?
  • which is the CPU? When training a model, I get the message
2018-10-02 22:42:38.780968: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA

Thus I was wondering if I could try to compile Tensorflow from source in order to get a speedup....the compiling environment in RStudio Cloud seems to be setup pretty well (gcc works flawlessly whenever I need to compile a package from source) thus I may attempt Tensorflow compilation through the RStudio terminal.

PS great cloud environment, it's currently one of the best possible ways to run R! You get a great IDE, up-to-date package versions and the latest (or near-latest) R version, without having to install a single file on your local machine!


#2

I think it's a T3.micro or T2.micro:

Think of it as a Chromebook...


#3

RStudio Cloud use Docker containers to run instances of the IDE. The containers are limited to 1 core, and 1GB of memory. The containers themselves currently run on AWS using c4.4xlarge instances (Xeon E5-2666 v3 CPUs).

While we currently do not support instances with GPUs, this is something that we're definitely going to add support for, likely as part of a paid offering.


#4

Wow! That's 16 virtual CPUs for free! I did notice a surprising speed. Great! :+1:


#5

Note that Andy says "The containers are limited to 1 core, and 1GB of memory."


#6

Whoops you're right! By the way, this is a textbook example of attentional bias - I knew already that the container had 1 Gb of RAM and 1 GB of storage, thus when I read Andy's answer, containing two 1, one of which was "1 Gb of RAM", my brain read the other one as "1 Gb of storage" - when it was "1 core"! Still looks pretty fast to me though - probably because usually R uses only 1 core , so I didn't see any noticeable slowdown with respect to my local machine.