Compute Optimizer is currently in Private Beta, available for any customer. If you would like access, please get in contact with your Rescale Representative.
Rescale’s Compute Optimizer is a feature that helps users choose the optimal hardware for their workloads, based on performance, cost, and maturity. It uses machine learning (ML) to make recommendations based on Rescale’s Performance Index (RPI), Rescale Maturity Index (RMI), and Rescale Cost Index (RCI).
How it Works
When creating a new job, the Compute Optimizer will automatically select the most balanced CoreType for you, based on the software they have chosen and any hardware filters that are set. If you adjust the number of cores, the Compute Optimizer will determine a new recommendation if needed based on maturity.
You are not forced to use the hardware recommendation, at any point, you can choose from the list of hardware and opt-out of using the recommendation for a given job.
The Compute Optimizer also works for templates and cloned jobs, but it will not automatically select the hardware. Instead, it will show up in the “Recommended” category for you to choose from. If you would like to use the recommended hardware, simply click on the CoreType and it will be selected for that job.
In the future, the Compute Optimizer will be able to select the number of cores using its machine-learning model. This will help users get the most out of their hardware, ensuring that they have the right balance of performance, cost, and maturity for their specific workloads.
How long does it take to provide a recommendation?
It generally takes between 5-10 seconds. We plan to reduce this time significantly with future iterations of the product.
Do you provide recommendations for Cloned Jobs and Templates?
By default, we provide recommendations for all jobs. For new jobs, we automatically select the recommended hardware. For Cloned Jobs and Templates that have existing hardware, we do not change the default hardware, but show the recommended hardware under the “Recommended” section.
Are hardware filters taken into account?
Yes, we use hardware filters to filter the hardware.
How does the recommendation work?
We leverage Rescale Performance Index (RPI), Rescale Maturity Index (RMI), and Rescale Cost Index (RCI) to create a recommendation.
- RPI is created based on our ML-models and is software-specific
- RMI is created based on capacity limits and the type of hardware
- RCI is a normalized set of prices among available hardware