The CoESRA virtual desktop can be launched powered by either CPU (Central Processing Unit) or GPU (Graphics Processing Unit) nodes. Your virtual desktop account folder as well as the data and analysis that you have stored in it will persist regardless of what settings you use or how many times you change the settings of your virtual desktop.
How do I know if I need to customise the settings?
A simple rule of thumb is to choose the smallest set of resources that will effectively and efficiently perform your tasks. In general, attempting to launch a desktop that requests more resources (ie. a GPU with 64 cores and 512GB of memory) will take longer and may result in your request being delayed as the task cannot be run until sufficient node resources become available.
To determine the optimal settings for your task, first launch a desktop using the default settings. If the tools you are using or your analysis takes an unreasonable amount of time to run, you should stop that desktop using the “Stop Desktop” button. Then depending on the task that you wish to run, customise your settings before launching the desktop again. It may take you a number of tries to work out if you need to change the settings and what settings will work the best for the analysis that you wish to run.
I think I need to customise my settings - which node type should I choose?
CPU nodes perform sequential processing and are best suited for general tasks. This setting will allow users to perform various tasks, such as complex calculations and running applications. CPU will utilise several cores with lower latency (delays in processing) making it ideal for executing single-threaded tasks at a faster rate. Using the Change Settings button to reveal the Setting options allows users to configure their desktops with more cores.
CPU node is most useful for:
Data Preparation
Feature Extraction
Small Scale Models
Complex Calculations with smaller data sets (ie. data collected using traditional techniques by one or two researchers)
GPU nodes deliver more performance and make use of parallel processing. GPU utilises a high number of cores however it has a higher latency (delays in processing). It is ideal for parallel processing making use of multi-threaded tasks to perform more efficiently. This makes the GPU node beneficial for machine learning or scientific computations because it can parallelise and run repetitive tasks quickly.
GPU Node is most useful for:
Machine Learning with Python packages
Data analysis with big data sets (ie. data collected by multiple remote sensors in a longitudinal study)
Processing large-scale images, including drone images
Run analysis on acoustic data
Training AI models
Image processing
Remember that the CoESRA nodes are a shared resource! There are potentially numerous other researchers needing to access the computing resources at the same time, so only request the resources that you need.
0 Comments