I tried to solve a large-scale MIP, but the poor performance of my PC makes the tests very time-consuming. If someone knows an alternative to Google Collab for paying with high-performance (e.g. 256G RAM) or any other suggestion, I would be appreciative.
If you need 256 GB of RAM NEOS server org is probably to small for you. In terms of cheap compute i would recommend a dedicated CCX51 instance with 128GB from Hetzner Cloud available for 0.5€/h. I have a referral link that should give you some credits to get started and will pay me the first 10€ if you spend money with them. Vultr has a 256 GB instance for around $1/hour an affiliate link that might give you some starting credits too. Other alternatives are to buy time at some interruptible spot instance at Amazon and do take care of check-pointing the solver using DMTCP. If your MILP solver runs on ARM you might also want to explore those instances on Oracle, Amazon, Alibaba Cloud as they tend to be a bit cheaper.
You could also see if you could get sign up at a local HPC center and submit a project request there. The last alternative is to excite someone else for your project so much that they pay for compute.
you could try WatsonStudio where you will be able to use CPLEX , SPSS and many open source
NB: I work for IBM
If the calculations are for academic purposes, many countries have national supercomputing facilities with lots and lots of computers each with 250GB of RAM or even more (some XSEDE computers had 12TB of RAM as early as 2018, and maybe more now).
Many of these national supercomputing facilities are described in great detail at this thread on the Stack Exchange Network: Supercomputers around the world
If you are not from a country with supercomputing facilities, then the alternative question Supercomputers around the world, which allow access for researchers outside the country may help.
Also, many commercial supercomputers are available such as the ones described in answers to What are some cloud services for computing?.
You could try Qarnot's cloud platform:
- You can create an account for free here (no banking information required) and get several hours of free computing time. Maybe it will cover your needs? Then prices range from 0.14€/h to 0.61€/h depending on the number of cores and RAM used.
- You can run many open-source softwares by providing appropriate docker images. Most of open-source softwares have official docker images on DockerHub but you can also create custom ones.
- Here is a tutorial example of how to run a simple or-tools computation on Qarnot's cloud platform.
Hope this helps.
NB: I work for Qarnot Computing.