Hardware lessons?

I just secured a good chunk of funding for dedicated deeplabcut workstation. It’s enough for a Dell 7920 workstation with a 2TB SSD, 64 GB RAM, and a TITAN RTX GPU (following general suggestions in the DLC wiki). I will probably have to run Windows on it since we sync data files on Box, which doesn’t have a Linux client.

This machine will be used on many projects, but not so many there will be time conflicts, so I am thinking one powerful GPU rather than several smaller.

With studies on lots of different types of animals, there will be a lot of different training sessions, rather than a lot of analysis (at least at first).

Has anyone gone this route and regret it? Wish they had more but weaker GPUs? Less SSD more RAM? I won’t see this kind of funding again for a while, so I want to make sure I’m somewhat future-proof, hence the monster workstation with room for new GPUs down the road.

You might consider building a machine vs. buying a Dell (saves money and more flexibility) …
also unless you use the machine for something else, the CPU/RAM won’t me used much, so even 16 GB of RAM is fine (and saves $$). I’ll let others give feeback on more GPUs vs. 1 better one. It’s too bad you can’t go for linux though! Docker is amazing!

Thanks! University restrictions may stick me with Dell, but I think we can dual-boot to be able to install deeplabcut with the docker.

In our University we plan to build a new workstation for analysis:

RAM: 200 GB
3.7 GHz CPUs (Intel) 8 cores or more
Graphics: NVIDIA Quadro P4000 8 GB or more
2x 1To PCIe SSDs

I think the main questions are
what is the type of your dataset? (Confocal, lightsheet microscope etc…)
Because we have in our core facilities Airyscan microscope and lightsheet.
Airyscan +/- 10 Gb per file and up to 100 Gb for timelapse
lightsheet +/- 100 Gb per file up to 400Gb for big organs

And which software you use for your analysis. the software requirement are very different.