As we are only starting to make more information about InTime available publicly, there hasn't been release-oriented posts until now. Going forward we'll aim to accompany every release with a post highlighting new features as well as adding detailed release notes to the InTime documentation. If you're not quite sure what InTime is yet, check out the InTime Sneak Preview.
Here are the highlights of InTime 1.1.4.
Better handling of designs with very high utilization
Designs with very high utilization are handled much better than before. First of all, if an InTime strategy results in over-fitting in the target device, an estimate of the over-utilized device resource is given in the project history instead of just showing that the strategy failed.
Furthermore, the learning algorithms now take strategies which don't fit into account when the design goal is Area.
Discard bad strategies without spending time to evaluate them
In addition to the existing machine learning algorithms, this version of InTime predicts which newly-generated strategies are more likely to produce sub-optimal results without wasting time to run them. This directly improves the QoR and reduces the time required to find the best solution.
Design goal comparison
We found that the best solution for one design goal doesn't always produce the best overall solution. For example, when attempting to reduce a design's utilization, the solution with the best area might not achieve timing closure. Intuitively this may be so, but the numbers prove it.
Therefore, to help users determine what InTime strategies result in the best overall performance, firstly a new Goal Comparison plot was added to the Results Analysis capabilities. This plot can be used to compare Power, Speed and Utilization performance across multiple InTime strategies. An example is shown below for the example design shipped with InTime:
Chinese Translation Added
This is the first version of InTime which contains a complete Chinese (Simplified) translation of the user interface for our Chinese users:
This translation will automatically be used on systems with zh_CN locales set as their active locales.
Smaller improvements and resolved issues
This release contains many minor improvements such as smarter "workers" in the private cloud configuration, the ability to set runtime limits for local strategies and realtime status updates for running jobs in the project history. Numerous issues have also been resolved, including those encountered while distributing designs to different run targets.
Detailed release notes for this and previous releases are available here. If you are experiencing any issues with this release, or would like to give feedback of any sort, please get in touch at tellus@plunify.com