Maybe you can improve the question by being a little more specific: 1) What do you mean by reliable (by standard metrics, all major cloud services are as reliable as any university computer if not more)? 2) What services/platforms are you looking for? Do you want bare iron where you have to do system administration, installing programs, etc? 3) What kind of hardware are you looking for? The fields you listed (without actual software) are as wide as it can get. — Greg Apr 1 at 13:29

Hi @Greg, I edited the question and added the information you ask. — Camps ♦ Apr 1 at 13:42

I get the feeling that you aren't using the power of the cloud with your requirements. Let me ask a few questions regarding them. 1) Why do you need performance to be similar to a local computer? 2) Your say that your demands are very heterogeneous, yet you want one server to handle everything. Why not use a server optimized for the individual task? 3) Why do you want a bare iron system and not virtual systems? 4) Do you need GPUs for all simulations? If not, why do you always want GPU access? 5) When you need GPU access, do you need fast CPUs? If not, why pay for it? — Polygorial 2 days ago

@Polygorial I can't answer all those questions, but I think the first one makes sense: supercomputers and cloud service machines typically by default provide less computing power than a local personal laptop (on Compute Canada supercomputers, you're limited to about 250MB of RAM for jobs on a login node, and likewise cloud services like Vultr provide extremely minimal resources for people paying the smallest amount, then people have to pay a lot more/month to get the type of power equivalent or greater to a personal laptop (typically). He wants the service to be "at least" as good as PC — Nike Dattani yesterday

@NikeDattani you just made my point. If the current job just needs 250MB RAM and not much processing power, why pay for more? If another job needs 128GB and loads of CPU that job can be run on a more expensive machine optimized for computing. And when no job is running no server is needed. With cloud computing it's possible to configure the server based on your requirements for the individual job instead of having an all-purpose server. As an example [AWS EC2]{aws.amazon.com/ec2/instance-types} have loads of different server configs, each with their uses. — Polygorial 1 hour ago