View Only

Introducing GPU optimization with IBM Turbonomic

By Patrycja Hubl- Lis posted 28 days ago


As the demand for advanced graphics processing units (GPU) grows to support machine learning, AI, video streaming and 3D visualization, safeguarding performance while maximizing efficiency is critical. 

IBM Turbonomic is dedicated to optimizing GPU workloads to promote maximum efficiency without sacrificing performance at the lowest cost. The benefit for customers includes performance optimization to promote faster response and smoother experience, and better efficiency by reducing resource waste and potentially keep costs down. 

Turbonomic is excited to announce the release of IBM Turbonomic 8.12.0 with comprehensive GPU optimization!Turbo can now monitor and optimize GPU on-prem and on cloud. The more permissions Turbo is granted for observation, the more optimizations it can drive. Here are a few ways Turbonomic is supporting GPU optimization: 

On Cloud 

Developers find it difficult to decide which GPU cloud instances would serve them best and, in most cases, they end up over-provisioning. With GPU instances costing close $100 an hour, this can result in a steep increase in their cloud bill.  

Turbonomic enables users to scale GPU instances continuously and automatically to the best instance type. 

Turbonomic uses NVIDIA’s GPU metrics such as the number of GPU cards and the amount of GPU memory in use for AWS EC2 instance types to drive these VM scale actions.Currently, Turbonomic supports P2, P3, P3dn, G3, G4dn, G5, and G5g instance types. You can view these metrics in the Capacity and Usage and Multiple Resources charts.  


On Premises 

The usage of GPUs is increasingly prevalent, especially in virtual machine (VM) environments. It's becoming common to configure VMs with virtual GPUs (vGPUs) to leverage their powerful processing capabilities.  

IBM Turbonomic now supports the following capabilities in support of GPU enabled VMs: 

  • VM Placement Optimization: In recommending VM placement actions, Turbo evaluates the GPUs installed on both the source and destination hosts, as well as the vGPU types assigned to VMs. It ensures that placement actions are only suggested if the destination host has compatible GPU cards and vGPU types, along with an adequate GPU memory buffer to support the vGPU type of the VM. 

  • Recognize Passthrough GPUs: VMs with Passthrough GPUs cannot be migrated from one host to another. Turbonomic now recognizes VMs that have Passthrough GPU attached to them, and blocks move actions on such VMs.    

Turbonomic is committed to developing GPU optimization services to provide performance insights and generate actions to achieve application performance and efficiency targets. We expand capabilities in GPU optimization to bring even more value to your Turbonomic investment.  

Generative AI / LLM applications run on GPUs, and we will soon share updates around intelligently scaling these workloads.  

Stay tuned for more exciting announcements! 

1 comment



27 days ago

More GPU Optimization updates in Turbonomic coming soon!