Bringing POWER to the People
Last month, Cirrascale joined the OpenPOWER foundation and soon after announced its entry into the market of POWER-based systems with its RM4950 OpenPOWER Development Platform at the GPU Technology Conference in San Jose. It was at this conference that IBM and the OpenPOWER Foundation were hosting their first OpenPOWER Summit. So, why is it that Cirrascale decided to get involved with POWER architecture?
Cirrascale has long been on the cutting-edge of HPC by providing reliable, flexible and high-performance hardware platforms that help customers achieve their goals as well as drive their business to meet their scale-up and scale-out needs. However, doing so has been extremely evolutionary and challenging over the past several years as HPC has expanded to include applications dealing with (buzzword alert) Big Data and Deep Learning. What hasn’t changed is the industry’s incessant need to keep exploring new avenues of technology to solve their compute intensive needs.
The POWER8 processors can execute many HPC workloads at over two times that of similarly rated x86 processors, due in part to the eight threads per core and outstanding memory bandwidth. Coupled with a base clock speed of over 3 GHz, and large cache sizes, this means a huge performance gain for CPU-based HPC workloads that are memory or thread constrained. This raw performance muscle is also delivered in a lower core count package. Many legacy HPC applications are challenged when it comes to scaling across many cores and therefore have not benefitted from the core count growth in typical next generation processors. These applications are inherently focused on fewer faster cores, which is another area where the Power Architecture excels, topping out at a whopping 5GHz! Aside from the overall computational goodness, the POWER8 Coherent Accelerator Processor Interface (“CAPI”) means that HPC workloads which take advantage of GPUs or other acceleration technologies now have a lower-latency communication path back to the CPU than was possible with other CPU architectures. Marrying these CPU capabilities with Cirrascale PCIe fabric yields a system with HPC performance capabilities heretofore unachievable. Cirrascale support of multiple NVIDIA Tesla GPU Accelerators attached to the POWER8 processor puts a lot of computing power in a single system – with an awesome price/performance ratio to boot.
The HPC landscape is definitely changing… and it’s becoming more exciting than ever to be involved.