With its latest upgrade, cloud users will supposedly be able to deploy fewer servers to crunch the same amount of data, reducing the overhead of high-performance computing. Furthermore, the technology giant also insists that, even with fewer machines deployed, data sets will be reasoned through at faster-than-ever rates.
How is that possible?
In a high-performance computing environment, using GPUs in conjunction with CPUs for processing can offload computationally intense aspects of an application onto the GPU. Given that GPUs usually consist of thousands of cores whose sole job is to process information at blazing-fast speeds, it makes sense to put the GPUs in the lead of crunching bigger data sets. Any CPUs onboard a server can be used to manage applications and run the infrastructure supporting the GPUs' efforts.
“With NVIDIA GPU technology on IBM Cloud, we are one step closer to offering supercomputing performance on a pay-as-you-go basis, which makes this new approach to tackling big data problems accessible to customers of all sizes,” says Jerry Gutierrez, High Performance Computing leader of IBM’s SoftLayer. “We’re at an inflection point in our industry, where GPU technology is opening the door for the next wave of breakthroughs across multiple industries.”
For engineering firms, having the ability to spin up supercomputing performance on demand could signal a big change in how far the state of the art can be pushed even by the smallest firms. Computational fluid dynamics, generative modeling, seismic simulations for structural engineering and other computationally intensive investigations could be undertaken without fear of running up a project's budget.
While it’ll likely be a few years before Gutierrez’s dream of full-blown, on-demand supercomputing is a reality, that time is near, and might have a huge effect on how designers create work, and how they address critical engineering problems.