Good pc for QuPath v.0.2.0 M9 and future versions

Hi! We are using QuPath v. 0.2.0 M9. We are very interested in the pixel classifier and we have decided to update our pc to work at a good speed. The performances of our pc are the following:

Processor: Intel Core i7 7700L CPU 4.20 GHz


We cannot expand the hardware of this PC for “bureaucratic” reasons, so we opted to buy a new one.

Could you recommend a good pc to work on? (possibly you could provide me with the purchase link).

We want to work at maximum speed, even for future versions

Thank you


Really depends on what you want to be doing with it, even in terms of QuPath projects, but some thoughts:

NVMe primary hard drive. The speed increase is worth it. I generally use something like the 970 EVO.

Large solid state hard drives for storage. I just RAIDed two 1TB HDDs for a project with 500 whole slide images. You will get much better performance than plugging in external hard drives or working off of a network/cloud. If you will also be handling large 3D data sets, say lightsheet, you may want to go much larger (8TB or so).

There is generally a tradeoff between single core speeds and having a lot of cores… unless you start spending a lot of money. Some processes will only ever run on a single core, and so a 32 core multi-cpu monster might end up being slower in some cases than a basic desktop. More cores allows anything tiled too be run more quickly though, N tiles can be processed at once. Great for large numbers of cells. If you want both high individual process speeds and high multi-core speeds, the CPUs can start costing tens of thousands of dollars.

On the other hand, the pixel classifier is huuuuungry, and I have run into a few cases of it simply closing down QuPath with no warning even with 25GB of RAM available (of 32). It seems like total CPU throughput and having plenty of memory is key to using that efficiently. I would recommend 64 GB of RAM, especially if you want to run more than one instance of QuPath at a time.

With 200+GB of RAM I was able to run 4 projects pixel classifying 10 images each all at once without problems, but dipping below 25GB of RAM per instance started causing me problems. Not sure if that will be an issue for you.

If you want to do other kinds of image analysis, say 3D analysis using other software, you will likely want to consider a nice video card as well. In the case QuPath allows you to train deep learning models directly in the future, an Nvidia card with plenty of RAM will definitely come in handy. The newer RTX cards have some nice improvements for deep learning applications. I haven’t used it, but there is also a CLIJ plugin for QuPath created by @haesleinhuepf that might allow you to improve processing speeds using the GPU, but that is not a default part of QuPath, and I’m not sure how many versions of QuPath it is compatible with.

Obviously not all of this is needed to simply run QuPath, but depending on how much you plan to push the limits… Anyway, I never like using pre-built systems so I don’t have any recommendations there.

1 Like

Hey @Research_Associate,

Thanks for mentioning clupath :slight_smile:

It is a prototype for a QuPath-CLIJ bridge which is only compatible with QuPath 0.2.0-m3. I didn’t continue working on it because the API of QuPath changed back in the days and higher priorities kept me from updating it. But if someone wants to use it in the current QuPath, I’m happy to make it happen. All I would need would be a clear use-case, example data and likely some support by @petebankhead - who was already very supportive when I wrote this prototype.



Hi @Manuel_Baldinu, hi all,

I’ve written a little bit about system requirements here. Apart from that, everything you might improve in your computer could help QuPath a bit (memory, hard disk, processor). I would rank them in that order of importance.

Expanding on that a bit:

  • A fast hard drive is especially important for whole slide images, because any software will need to frequently return to the hard disk to access tiles. But once the image tiles have been read, then RAM is what matters - because that determines how many of them can be kept in memory.
  • An NVIDIA GPU is a good investment, given the huge & growing popularity of deep learning… you might not need it today, but could miss it in the future
  • The pixel classifier in particular needs a lot of RAM
  • More processor cores is good up to a point… but there are limits because
    • Some things are inherently restricted to one processor core (including pretty much everything that directly relates to the user interface in a Java application like QuPath)
    • Accessing image tiles from the disk can become a bottleneck
    • If you have 16x the number of processors, you’ll need roughly 16x the amount of RAM to be able to use them all…

I myself have never worked with a computer anywhere close to as powerful as those @Research_Associate has used…


Excellent point there. I had to turn down the number of cores (from 44->~20) because it was killing a 19GB whole slide fluorescent scan’s cell detection, just a couple of days ago. The user I was testing for also couldn’t get the cell detection to even start on their computer, which I suspect had to do with storing the 19GB image in a separate network location. Even letting it run overnight never resulted in the tiles showing up to indicate cell detection had started.

Thank you. The computer we use now has a very large disk, since it was designed to store the scans of the slides (several TBs). Indeed also the PC that we will buy must have this characteristics, since working on an external disk slows down the calculation processes. We had no major problems in classifier training, but rather in converting areas into annotations (several minutes) and counting the cells within them (here too several minutes, if all goes well, sometimes even Qupath crashes, losing all the work). The movements along the image are very slow, you necessarily need a new computer to work. Are there already computers with these characteristics or is it better to buy the various pieces and assemble it? thanks

I think Pete has posted in a few different places that large numbers of annotations result in QuPath slowing down in M9, and that should be improved in M10. If you are crashing during cell detection, I would recommend lowering the number of cores QuPath is allowed access to in the Preferences.

As Pete mentioned, the main issue you are probably running into is RAM based. 8 GB is not that much when several GB of that are being used up by the operating system itself. If you could upgrade your current computer to 32GB of RAM, you would probably see significant improvements in many parts of the program. Large numbers of annotations will be an issue, however, no matter how much memory you have.

Have you tried creating Detections rather than annotations?

@petebankhead if people are going to be using the pixel classifier to make increasingly complex and vertex heavy annotations in order to run cell detection inside of them, have you considered being able to pass ROIs or Detections to functions like Cell detection (with some serious warnings, or maybe only through scripting)?


I honestly didn’t think it was possible to perform cell detection within the Detections. do you think it’s a less heavy method?

It currently is not, as far as I know. Detections are pretty much always easier to work with in terms of stress on the system, but they are less flexible. That’s why the program can handle 2 million cells, but not 2 million annotations. You can’t use the wand or brush tool to edit your cells, though, as an example of the loss of flexibility.

1 Like

Hi! Can it be a good pc in your opinion?

I would personally prefer more RAM as 32GB limits what I can do on my home computer. I am far more comfortable (and have less issues) with larger amounts of RAM on work computers. Of course, that is fairly easy to upgrade after the fact.

Similarly, the 4TB hard drive won’t be great for access speed, but could be great for storing inactive projects. The 512GB SSD probably won’t be enough space for many active projects, and I don’t see where it states whether the drive is a normal SSD or an NVMe drive. I would recommend the NVMe, and more like 1TB if you are going to have large projects. Maybe an extra 2TB SSD if you expect to have large projects… but that depends on the projects you will be running. We have one project where the raw image data is over 1TB.

Video card looks very good.

Processor doesn’t have great multithreading capacity, but should run fairly quickly for single threaded processes while also giving you some multithreading capacity. Probably a good balance given the small amount of RAM.

theoretically they are 32 GB of RAM plus two other 16 GB RAM. the technicians confirmed it to us. So it’s 64gb of ram. Our project are not so big (less than 50 gb)

1 Like

Hello everyone, I found two more PCs within my reach. This week I have to buy one. Do you think the first PC I posted is better, or this

or this?


Not really sure, but the one thing that jumps out at me is that the first computer listed does not seem to have a solid state drive. The second at least has an SSD, though, again, an NVMe drive would be a better option.

Would some of these “AI ready” workstation laptops be useful options?!#1885,1886