More effective sharing of GPUs

IBM Spectrum Symphony GPU Harvesting is available as an optional add-on for both the IBM Spectrum Symphony Standard Edition and the IBM Spectrum Symphony Advanced Edition. The software monitors GPU resources and optimally places GPU-sensitive workloads on the most appropriate systems through configurable exchange policies and priorities. This reduces latency and improves application performance. At the same time, the applications are provided as needed and more simplified GPU
performance.

Better performance for better competitive advantage

This supports GPU-sensitive applications developed based on the NVIDIA CUDA environment to maximize all cluster resources for optimal performance. By sharing the GPU hosts through the various applications, these can share the GPUs and save infrastructure costs. The combination of GPU applications and the distributed computing capabilities of IBM Spectrum Symphony supports scaled applications with many GPU services that provide maximum performance in parallel.

Reorder GPUs via known API

From a customer perspective, the programming interfaces in IBM Spectrum Symphony are unchanged when invoking GPU-sensitive services. IBM Spectrum Symphony provides the convenient way to invoke GPU services, query GPU device identifiers, or fix GPU-specific runtime errors. As customers’ application programming interfaces remain unchanged, GPU computing services can be seamlessly added and accessed from a wide range of clients in a variety of application environments.

More benefits through more features

GPU-sensitive resource dashboards:

System administrators benefit from simplified management of GPU-enabled hosts. These hosts can be automatically combined in a GPU resource group; the application profiles for GPU-sensitive applications can then simply refer to this resource group.

Flexible GPU scheduling options: IBM Spectrum Symphony GPU Harvesting enables services based on different CUDA library versions and GPU hardware models to co-exist concurrently on the same IBM Spectrum Symphony cluster. They configure the number of GPU devices, monitor different GPU attributes, and select runtime GPUs based on flexible resource requirement formulations.

Automatically handle exceptions for reliable execution: GPU applications work reliably when managed by IBM Spectrum Symphony GPU Harvesting, which detects common issues such as ECC errors and overheating and transparently handles them. Formulating flexible resource requirements aligns computing services with the appropriate hosts and minimizes the risk of execution failure.

Why IBM?

IBM Spectrum Computing provides a comprehensive portfolio of software-defined infrastructure solutions that help your organization deploy IT services in the most efficient way, optimize resource utilization for faster results, and reduce costs. These offerings help maximize the full potential of your infrastructure and accelerate analytics, HPC, Apache Hadoop, Spark, and cloud-based applications of all sizes. In addition, they extract insights from your data and higher-quality products are brought to market faster.

Whether deployed in a datacenter or in the cloud, IBM Spectrum

Computing supports product development, critical business decisions, and breakthroughs in financial services, manufacturing, digital media, and more. Oil and Gas Industries, Life Sciences, Public Sector, Research and Development. From Formula One racing car design to loan risk analysis, organizations in a variety of industries use IBM Spectrum Computing as the foundation for software-defined large data, analytics, HPC, and cloud infrastructure solutions to deliver tangible business benefits.