Hardware config to run CP




We are buying a new PC to run Cell Profiler on our lab.

I am not sure if buying a better GPU would make my analyses run faster.

Also, what resource is more used on the analysis? Is it CPU, RAM or the hard disk?

If someone could help me it would be awesome. I really want to buy the best PC possible for this task.



I found that cellprofiler was mostly CPU bound, so for our workstation which we use for running small jobs we prioritised number of CPU cores for running as many workers as possible (you can get Intel Xeons with 40+ virtual cores), though it seems like 2GB of RAM per worker seems to be enough memory.

Though we still resort to a compute cluster for anything more than half a dozen 384 well plates.


@ezmarques Currently CellProfiler does not use any GPU optimized algorithms, so, as @Swarchal says, the CPU is a bottleneck. Check out these resources to see why:

  1. https://stackoverflow.com/questions/41567895/will-scikit-learn-utilize-gpu
  2. https://github.com/scikit-image/scikit-image/issues/1727

Also to echo @Swarchal, running CellProfiler from the gui will utilize all available cores/threads on your machine, so processing large sets of images can be done in parallel. Therefore, the Intel Xeon chips are a good option. Check out the Core i9 series of chips from Intel, too. Core chips tend to have higher clock speeds than Xeon chips and the top of the line i9 has 18cores/36threads.

What computer specs is best to run Cell Profiler?

Thanks @Swarchal and @karhohs. I’ll check out which Xeon processor we can get here.

@Swarchal, you said 2GB per worker. Then if I have a processor with 16 cores, its best if I have 32GB of RAM. Is that it?



@ezmarques In our case we have a 10 core machine that runs 20 workers with hyperthreading and gets by fine with 32GB of memory - although this will be dependent on the size of your images and what sort of analysis you’re carrying out.


Perfect @Swarchal

Thanks again!


One last thing @Swarchal. Did you tried to run using SSD as well?

I guess it will also improve performance, but I’m not sure how much it will improve.



@ezmarques Sorry, It’s not something I’ve compared. I think most setups pull in images from a networked drive rather than storing them locally.


It makes sense.

Thanks again!