How to choose a powerful computer or supercomputer for machine learning?

Question

Iron

Problem
It is possible to spend some (large) amount of money to buy a computer to solve various machine learning problems. The field of machine learning includes many methods and models, as well as methods for pre- and post-processing of data. And they are no longer parallelized or, if they can, they are broken into large pieces that are executed sequentially and very slowly. Using the GPU here does not give any speed gain, but it makes life much more difficult for the developer.
Offers on the market usually rest on GPU complexes. For 250t-1000t you can find many offers of ready-made assemblies on NVIDIA cards. Everything is simple and clear here. There is one BUT.
GPUs are good for fine-grained parallelism. This works great in neural networks or linear algebra, where it is possible to represent everything in matrices and matrix operations.
A simple example. Evolutionary algorithms are inherently parallel (a population in which decision individuals are independent). The most costly is the calculation of the fitness scores of individuals. In an applied task, you will have to make several tens or hundreds of such estimates, and each estimate is a work with data or another model that is calculated sequentially. A great option is multi-core CPUs - one core makes one evaluation. The simplest implementation and almost proportional acceleration.
Question
I'm not strong in iron. Are there any solutions (semi-professional, MSU-level supercomputer not needed) that can use more than one multi-core CPU + GPU at the same time? Or some platforms for building such systems?

Answer 1

AWS https://aws.amazon.com/en/machine-learning/
Azure https://azure.microsoft.com/en-us/services/machine...
GCP https://cloud.google.com/products /ai/
and stop shaggy grandma and give big money for something that will then stand in the corner

Answer 2

Google colab free gpu

Answer 3

In general, I would also advise you to rent something if necessary, but usually any task can be solved on a regular household computer, without any frills, just reduce the amount of data, or do you want to become a world champion in data analysis? if this is the task, then it is better for you to show some result at the master level at the beginning, and then cooperate with someone who will provide the cluster.

Answer 4

I'm not strong in iron

Then hire someone who is strong in it and who will pick it up exactly for your tasks. And perhaps he already has access to integrators who will immediately supply the hardware.
Or do you seriously think that someone here will advise you free of charge how to spend the mulien?

Answer 5

>> Problem
>> It is possible to spend some (large) amount of money to buy a computer to solve various machine learning problems.
Everyone should have such problems :)
It seems that you still do not know what tasks you will solve, but you are already trying to optimize for it.
Specify the problem statement first.