SOSCIP's Advanced Computing Platforms

A high performance combination you won't find anywhere else

The SOSCIP advanced computing platforms feature three distinct and unique systems that provide academic and industry researchers with a competitive advantage not available anywhere else in Canada. Our technology is bolstered by high-level technical support from our experts at the University of Toronto and Western University.

GPU-Accelerated Platform

The SOSCIP GPU-Accelerated Platform is a high-performance compute cluster built on the latest generation IBM Power System S822LC for HPC servers powered by NVIDIA Tesla P100 GPUs and POWER8 CPUs. The systems have NVLink technology to accelerate computing by allowing for greater speed, programmability and accessibility of data. It delivers the fastest performance on a wide variety of GPU computing applications. Tesla P100 GPUs enable massive parallelism and high memory bandwidth  with NVLink for faster communication between the CPUs and GPUs. In addition to its high performance and scalability, it has advanced energy management features to optimize and conserve power usage, and to maximize performance per watt.

The SOSCIP GPU-Accelerated Platform is one system rack consisting of 14 IBM Power S822LC “Minksy” Servers each with 2x10core 3.25GHz POWER8 CPUs and 512GB RAM. POWER8 CPUs are capable of Simultaneous MultiThreading (SMT) with 8 threads per core allowing the 20 physical cores to support up to 160 threads. Each node has four NVIDIA Tesla P100 GPUs, each with 16GB of RAM and CUDA Capability 6.0 (Pascal), with connections to each other and to the CPUs via NVlink. The primary OS is Linux Ubuntu 16.04  that manages the compute nodes and the system.

The SOSCIP GPU-Accelerated Platform leverages the IBM PowerAI software that makes machine learning/deep learning with popular frameworks such as Caffe and TensorFlow more accessible and higher performing on IBM Power Systems. The SOSCIP GPU-Accelerated platform is ideal for batch computing big data applications such as: deep learning with training /inference over large models, astronomy, computational fluid dynamics, genomic/next generation sequencing, molecular dynamics, video processing, IoT, geophysics, and computational chemistry.

GPU-Accelerated Platform

The new SOSCIP GPU-Accelerated Platform is a high-performance GPU cluster based on IBM Power System AC922 powered by NVIDIA Tesla V100 GPUs and IBM Power9 CPUs. The systems have NVLink 2.0 technology to accelerate computing by allowing for greater speed, programmability and accessibility of data. This platform provides 148 GPU-equivalent allocation on SOSCIP-SciNet cluster Mist which consists of total 54 IBM AC922 servers each with 2×16 core Power9 GPU and 256GB RAM. Each compute node has 4 NVIDIA Tesla V100 GPUs with 32GB of RAM and CUDA capability 7.0 (Volta), with connection to each other and to the CPUs via NVLink. The OS is Red Hat Enterprise Linux 7.6 for power9 (Little Endian).

The new SOSCIP GPU-Accelerated Platform supports popular AI, Machine Learning and Deep Learning frameworks such as TensorFlow and PyTorch provided as Python Wheels and Conda packages from Watson Machine Learning Community Edition (PowerAI). The platform also has capability of accelerating Molecular Dynamics codes (NAMD, Gromacs, etc) and other computationally intensive applications in computational chemistry, geophysics, next generation sequencing and astronomy.

Parallel CPU Platform

The SOSCIP Parallel-CPU platform is a homogeneous high-performance system attached to the SciNet supercomputer Niagara which is designed to enable large parallel jobs in order to optimize throughput of a range of scientific codes running at scale, energy efficiency, and network and storage performance and capacity. The platform provides 2880 core-equivalent allocation on Niagara. Each compute node (based on Lenovo SD530 server) has 40 Intel Skylake/Cascade-Lake cores with 202GB (188 GiB) of RAM. A fast 100Gbit Infiniband network is connected in a ‘Dragonfly+’ topology with Adaptive Routing.

The SOSCIP Parallel-CPU platform is an ideal system for running parallel code that cannot obtain cost-efficient speedup through GPUs. Computational Fluid Dynamics codes (OpenFoam, Nek5000, etc) and Quantum Chemistry codes (VASP, CP2K, etc) are all good candidates on this platform.

AGILE COMPUTING PLATFORM

Agile computing systems use reconfigurable hardware that can be adapted to the computing task at hand, delivering significant improvements in performance while using less power than traditional CPU-only systems.  One such reconfigurable hardware technology, called Field-Programmable Gate Arrays (FPGAs), is emerging as a powerful accelerator device for a wide range of applications such as real-time data stream processing, machine learning, big data analytics, and physical systems simulation.

The SOSCIP Agile Computing Platform is the first cloud-based FPGA accelerator development platform in Ontario.  It is a set of heterogeneous systems consisting of IBM POWER8(R)- and x86-based servers and Nallatech PCIe-385N(TM) accelerator boards with Altera(R) Stratix(R) V FPGAs. POWER8 servers include IBM’s Coherent Accelerator Processor Interface (CAPI) technology, which extends shared virtual memory capabilities to the accelerator and significantly reduces device driver overhead.  Development tools available on the platform for FPGA accelerators include Altera Quartus(R) II, OpenCL(R), and the IBM CAPI Development Kit. Additional tools and support for agile computing projects are available separately from CMC Microsystems.

BLUE GENE/Q PLATFORM

The SOSCIP IBM Blue Gene/Q (BGQ) is Canada’s fastest supercomputer and is designed to handle large-scale, distributed applications that require massively parallel processing power, such as molecular modelling, drug discovery, climate change forecasting, and computational fluid dynamics.

The BGQ is well-suited for applications that can use 1,024 cores or more at a time and require low-latency, high-bandwidth communication between processors. A highly dense and energy efficient supercomputer, the BGQ is built around a system-on-a-chip compute node with a 16-core 1.6 GHz PowerPC®-based CPU with 16 GB of RAM. The SOSCIP BGQ is a four-rack 65,536-core system capable of 840 Tflops peak theoretical performance. Each rack has 16 I/O nodes that run a full Red Hat Linux OS that manages the compute nodes and mounts the file system. The compute nodes run a lightweight Linux-based operating system called CNK and are connected using a custom 5D torus high-speed interconnect.

CLOUD ANALYTICS PLATFORM

The SOSCIP cloud analytics platform is the first research-dedicated cloud environment in Canada.  The SOSCIP cloud is unique in that it provides access to a broad array of IBM software tools for application development and data analytics which can be combined with user-specific and other open source software to create customized virtual machines to meet project demands.  The cloud platform is ideal for complex data analysis, streaming and managing large data volumes, and data mining applications.

The SOSCIP cloud platform is based on OpenStack, and provides users with a self-serve big data analytics computing environment. Running on a cluster of Intel Xeon hypervisors, it has 1440 CPU cores, 7.5TB of RAM, and 850TB of disk storage in total. Using Red Hat Enterprise Linux, the cloud provides an environment for IBM analytics software including InfoSphere Streams, BigInsights, and DB2, amongst others.

HIGH SECURITY LARGE MEMORY SYSTEM

The SOSCIP High Security Large Memory System (LMS) is a virtual symmetric multiprocessing environment, aggregating three IBM x86 servers using ScaleMP software into a single virtual system.

The primary hardware node consists of four Intel Xeon Processors providing 32-cores at 2.7GHz with 1.5TB of RAM. Two secondary hardware nodes consist of four Intel Xeon Processors providing 16-cores at 2.0GHz with 1.5TB of RAM each. The result is a single virtual 64-core, 4.5TB environment. The network topology of the LMS is a mesh, with each hardware node having a direct-attached QDR Infiniband link to other nodes.

The SOSCIP LMS is outfitted with the latest IBM analytics software and is ideal for data intensive projects that involve in-memory databases, key-value storage, and large computations.

Use of SOSCIP’s computing platforms are governed by our Terms of Use.