Cloud Update: In its cloud, IBM has created a machine with affordable AI.

Furthermore, it is built for AI tasks.

Although IBM's response to the affordable supercomputer has been operational for some time, it has only lately made any concrete disclosures about its so-called Vela project.

Looking to its blog(opens in new tab) for more information, IBM disclosed that the study, which was written by five company employees, addresses issues with earlier supercomputers and their unpreparedness for AI tasks.

The business clarifies the choices it made regarding the use of low-cost but powerful hardware in order to adapt the supercomputer model for this future type of workload.

We need your assistance to create a better website for our users. By responding to our poll and sharing your thoughts and ideas about the tech sector in 2023, you can contribute. Your responses will all be kept private and secret, and it won't take long. Once more, thank you for your assistance in improving TechRadar Pro.

Although it's obvious that these supercomputers can handle demanding AI workloads, including the one built for OpenAI, the company behind the well-known ChatGPT live chat software, a lack of optimization has left traditional supercomputers at risk of lacking valuable power and having an excess in other areas, which results in an unnecessary expenditure.

Although it is well known that bare metal nodes are the best for AI, IBM wished to investigate making these available inside of a virtual machine (VM). Big Blue claims that the outcome is significant efficiency improvements.

We developed a method to expose all of the node's capabilities (GPUs, CPUs, networking, and storage) into the VM so that the virtualization overhead is less than 5%, which is the lowest overhead in the industry that we are aware of, after conducting a substantial amount of research and discovery.

Vela has four 3.2TB NVMe storage drives, 1.5TB of DRAM, and 80GB of GPU memory in terms of node architecture.

According to The Next Platform's estimation, if IBM wished to include its supercomputer in the Top 500 rankings, it would produce approximately 27.9 petaflops of performance, putting it in 15th place as of November 2022.

Although existing supercomputers are capable of handling workloads related to artificial intelligence, the rapid advancements in this field and the pressing need for cost-effectiveness highlight the need for such a machine.