Myth #3: The Cloud Can’t Meet HPC Requirements
Editor’s note: This is the 3rd blog post of the ebook Dispelling the Myths of Cloud HPC. Read the full piece here.
In the early days of cloud HPC (~2010) the only types of processors available were variants of x86 architectures. Today there is now a wide range of classes of processors capable of driving HPC applications. In addition to x86 processors, there are now GPUs from NVIDIA and floating-point gate arrays from Xilinx that are readily accessible to anyone from anywhere. It’s clear that every type of processor that can be found in an on-premises IT platform is now available in the cloud.
The truth is the cloud enables organizations to accelerate the pace of HPC innovation in a way that is simply not possible in on-premises IT environments. HPC applications benefit from being able to tap multiple types of processors to optimize performance. Engineers and scientists can flexibly select cloud-based hardware for each specific workload they run using analytics tools that identify what workloads should be deployed based on real benchmark data.
In fact, a recent Benchmarking Public Cloud-HPC Hardware report finds on average organizations can reduce their overall simulation costs by 20% and increase their application performance by 30% by selecting the most optimal hardware available from public cloud providers.
Not surprisingly, the number of HPC applications being deployed in the cloud has been increasing at double-digit rates. A recent report published by Intersect360 finds more than 70% of organizations make use of the cloud, with more than half (53%) consistently opting to deploy HPC applications in the cloud. Similarly, a report published by Forrester finds 99% of respondents indicated their organizations are open to shifting more HPC workloads to the cloud. The respondents highlighted several factors that will enable this shift, including an expansion of HPC services in cloud environments, the ability to access workload automation management solutions across their organizations’ toolset, the availability of turnkey hybrid and multi-cloud platforms, and greater flexibility to choose hardware and tools that fit their organizations’ specific use.
With more demanding applications coming to market each year, we expect cloud architectures and infrastructure to continue their evolution. While Moore’s Law gains may be plateauing in x86 architectures, exponential cost-performance gains can still be realized by taking advantage of the latest architectural variants. Across GPU, x86, ARM, RISC-V and other specialized architectures, specialized hardware selection is top of mind, and cloud enables the flexibility to operationalize it. For example, new use cases with HPC+AI/ML are driving rapid adoption of new GPU cloud offerings. Cloud is especially valuable as a strategy in this case because physical hardware can be difficult to come by due to global chip shortages.
Making sense of the many options and possible HW/SW configurations can intimidate even the most seasoned IT/HPC professionals. Rescale has been pioneering cutting edge cloud and HPC use cases for over a decade and as a result we have deep data on the best configurations. Our team of experts understand the pain slow-running workloads and cloud cost overruns, and we commit to sharing our best practices with you to avoid these challenges. We’re confident we can help our customers run even the most demanding applications in the cloud. We work with customers who previously tried and failed in their cloud migrations for various reasons, and together we were able to build highly performant solutions in the cloud. Hopefully by now you overcome any hesitations that cloud can’t handle HPC and we look forward to hearing from you to tackle your next computing challenge!