Variation in specifications and performance of computers of course means that the choice of hardware can make a difference to the efficiency of your workflow.
Author: EIVA CEO Jeppe Nielsen
We decided to look into just how much a difference this is via the performance statistics gathered within the NaviSuite software product for data modelling and visualisation, NaviModel. The statistics proved to hold valuable information, and the answer is: a huge difference.
(Click to enlarge) Scatter plot showing S-CAN computation time (in seconds) versus size of data being processed (in points) where each plot is a single run of S-CAN
Comparing computer performance in connection with advanced subsea point cloud cleaning algorithm
When we released NaviModel 4.1, it came with an improved, much faster S-CAN algorithm. S-CAN is an advanced tool for automatic point cloud cleaning, developed in collaboration with the Center for Massive Data Algorithmics (MADALGO) at Aarhus University in Denmark.
This posed the perfect opportunity to dive into the effect computer specifications (especially in terms of RAM and disc size) have on the efficiency of software tools.
The above data was gathered from our own tests as well as those customers who have given consent to us using the data, during a period of 8 months. It covers more than 40,000 runs of S-CAN, and numerous different computers, covering the entire scale of specifications from poor to high performance.
When analysing the data, it becomes clear that:
- There are big variations in computing time. For example, with 10 million points (10E6), we see performance ranging from 25 seconds to 600 seconds. This means that there are computers out there that are 24 times faster than the slowest computer used.
- The performance of S-CAN is quite linear.
- Most usages of the S-CAN tools are carried out with less than 30 million points, and the largest we have measured is 750 million points in a single S-CAN run.
- S-CAN SCORE (blue plots – typically used for data sets with a single surface) is used more often than S-CAN COMPONENTS (green plots – typically used for data sets with multiple surfaces) among our users.
So what can we learn from this?
If you run S-CAN on a data set with for example 10 million points and measure the processing time, you can see from the graph how other machines are performing. If your results come in closer to the 600 seconds than 25, it is an indication that it may be worth your while buying a faster machine.