Supercomputers used to be fairly limited in number, with the major users being governments, public research facilities, and scientific organizations.
The accessibility of sophisticated cloud workload management (CWM) tools has substantially reduced the barrier of entry owing to the development of cloud computing and its wide adoption.
Last month, a CWM company called YellowDog in Bristol, United Kingdom, built a virtual super computer with its own platform and for about 10 minutes, it had amassed a force of more than 3.2 million vCPUs.
For a few minutes, while it was no match for Fugaku, it was nonetheless in the top 10 of the world’s fastest supercomputers.
The provisioning of a popular drug discovery software as a single cluster was assisted by the pharmaceutical firm. The raw cost of the project, according to an amateur estimate, is around 65,000.
This is a cost-effective and scalable solution that provides the same performance, security, and power as bare metal servers at just 1.6013 per hour. That’s accounting for 33,333 instances of AWS 96-core c5.24xlarge instance, one of the instances utilized during the run (essentially comparable to baremetal servers or dedicated servers).
So that’s 53,376 per hour or 57,824 to take into account the entire duration of the session (65 minutes in all).
“With access to this on-demand supercomputer, the researchers were able to analyze and screen 337 million compounds in 7 hours. To replicate that using their on-premises systems would have taken two months,” said Colin Bridger from AWS.
What’s remarkable is that this level of firepower is available to everyone who has the money for it. It’s based on the same hardware as our cloud computing world: web hosting, website builders, cloud storage, email services, and so on.
CWM platforms have grown over time to include algorithms and machine learning capabilities that can identify the best source of computing regardless of its origin or type.
Take the case of selecting a cloud service provider who has the lowest-cost spot compute. However, if this was not accessible in the customer’s territory or there were not really a sufficient amount of servers of the right instance type available within that cloud provider, another source of compute would be utilized.