IBM Platform Computing

IBM Platform GPU Harvesting Software

IBM Platform GPU Harvesting software for IBM Platform Symphony helps unleash the power of general-purpose GPUs by harvesting idle cycles. Automatically adapting resources to meet demands can reduce run times, improve service levels and help eliminate additional capital spending.

The Challenge

Not enough money for more GPUs?

For many large enterprises, grid computing is the best solution to handle large-scale jobs like analyzing big data. But with increasingly restrictive economic pressures, organizations like yours are looking for better ways to improve IT performance without incurring the large capital expenses required for infrastructure expansion. But for GPU-intensive applications, there never seems to be enough power available. And there’s usually little or no budget available for buying more.

The Solution

Get more GPUs without buying them

IBM® Platform™ GPU Harvesting software for IBM Platform Symphony extends the harvesting capabilities in Platform Symphony to unleash the power of general-purpose graphic processing units. This enables applications to scale beyond the confines a single GPU. This enables applications to share expensive GPU resources more effectively and to scale beyond the confines a single GPU. By sharing GPUs more efficiently among multiple applications, and by detecting and addressing GPU specific issues at run time, service levels are improved, and unnecessary capital spending is avoided.

Platform GPU Harvesting is an optional add-on available for both the Standard and Advanced editions of Platform Symphony. The software dynamically monitors demand for GPU cycles by applications. Then it can transparently tap your Platform Symphony grid’s available GPU cycles across multiple environments. That helps reduce wait times and improve application performance, while delivering more on-demand GPU power to applications that require it with simplified management.

The Benefits

Creating more GPU power for less

Platform GPU Harvesting software can improve your productivity picture.

• Shares GPUs between departments and applications
• Scales performance beyond the confines of a single GPU
• Simplifies GPU cluster management
• Improves application efficiency and reliability
• Reduces manual administration

The Specifics

Maximum GPU utilization

In most instances, Platform Symphony programming interfaces are unchanged when calling GPU-aware services. These include methods to invoke a GPU service, retrieve the GPU device ID or handle GPU-specific run-time error conditions. Because client APIs are unchanged, GPU compute services can be added seamlessly and made callable from heterogeneous clients running diverse application environments.

• Flexible GPU scheduling options: With Platform GPU Harvesting, services built using CUDA different library versions and different GPU hardware models can coexist on the same Platform Symphony cluster.
• Automated exception handling for reliable execution: GPU applications run reliably with exception handling at multiple levels when managed by Platform GPU Harvesting. By providing flexible resource requirement expressions, compute services are matched to appropriate GPU hosts minimizing the chance of a failure at run time due to software incompatibilities.

Read the datasheet

Back to Platform Symphony family