FedCentric Technology has a long and storied history with scale-up hardware. Working closely with our partners in SGI and HPE, we have paved the way of scale-up supercomputing at the United States Postal Service, and many others. Additionally, we have worked in scale-out supercomputing and have developed solutions in the cloud. Actionable real-time results require millions of answers per second. Scattered data creates delay among network servers and slows down results. Combining RAM and computing power into a single server significantly speeds the time to insight and simplifies the programming model. There is one huge block of memory to hold the data and hundreds or thousands of compute core to crunch results.
HPC – High Performance Computing
HPE/SGI has the largest share of HPC platforms providing 140 of the TOP 500 supercomputers. FedCentric brought HPC into the enterprise where our customers are using it to provide actionable information and make split second decisions. Scale-out Cluster solutions include: SGI ICE XE and HPE Apollo 8000. Scale-up large memory Symmetric Multi-Processing (SMP) solutions include: HPE Superdome X and the SGI UV300.
HPDA – High Performance Data Analytics
FedCentric Technologies’ strength stems from High Performance Data Analytics. We define this as “workloads that are daunting enough to require HPC technology.” The primary factors riving the HPA trend are the complexity and time criticality of the most challenging big-data workloads. We are at the forefront of HPDA with our hardware, software, and interdisciplinary teams. We currently leverage them in cyber security, tax fraud, science, and genomics.
GP-GPU – General Purpose Graphics Processing Unit
FedCentric uses the dense power of GPUs for both HPC and HPDA workloads. With 150B transistors delivering 5.3 DP TeraFLOPS in a single card, the NVIDIA P100 can be the core of a hybrid HPC solution. HPDA and machine learning applications are further sped up by the 21.2 TeraFLOPS half-precision mode. Training time for machine learning applications is nearly 4x faster with no impact on accuracy. The HPE Apollo 4500 supports eight P100 GPUs in a 4U rack configuration. These servers can run stand alone or be clustered for enterprise-scale HPC or HPDA applications.
The advent of hard disk drives (HDD) and solid state drives (SSD) greater than 10TB each provides an opportunity to deliver over a PetaByte (PB) of storage in 6U of a rack or about 8PB per 19” 42U rack. SSDs are now exceeding the capacity HDDs, and while SSDs cost about 10x HDDs, they provide 1000x the IOPS. Better yet the SSD cost per TB is dropping at twice the rate of HDDs and will crossover in the early 2020s. FedCentric tests all the emerging storage platforms at FedCentric Labs before making customer recommendations. Our storage engineers replicate the workload, level of parity (RAID), file system, and interconnect fabric to ensure the customer needs are met.