View a scalability graph showing the predicted approximate performance characteristics of the selected site. The Scalability graph is automatically updated if you change modeling assumptions in the Suitability Report.
Two types of graphs appear depending upon the selected Target System. The X axis of the graph shows either CPU Count to model CPU usage or Coprocessor Threads to model Intel® Xeon Phi™ coprocessor usage.
Scalability Graph for CPU Count Usage
If you select a Target System of CPU or select Offload to Intel Xeon Phi and uncheck the column Offload to Intel Xeon Phi for the selected site, a green-yellow-red graph appears and the X axis shows CPU Count.
The number of CPUs appears on the X axis and the program's predicted run-time performance gain appears on the Y axis.
Near the top of each vertical line for a CPU number, you will see a box and a circle that indicate the minimum and maximum predicted gain values. The color shading indicates the likely benefit. For example, if the minimum-maximum range appears in the red-shaded area, this site is hurting your program's performance and you should significantly modify or remove the site annotations.
If the minimum-maximum range appears in the:
Red-shaded area, this site is hurting your program's performance. Remove or significantly modify the site annotations.
Yellow-shaded area, this site is helping your program's performance, but maybe the gain may not be enough to justify the effort you needed to refactor and maintain your program. You should investigate how the results can be improved. Also, it might be more energy (power) efficient to increase the serial throughput using program changes or faster processors (parallel machines are designed to run at a lower clock frequency than certain serial machines).
Green-shaded area, this indicates good results. If you can achieve this level of performance, you will likely have a significant improvement to your program. The green area also indicates energy efficient (power) parallel speedups.
If the minimum-maximum range forms a flat line, this indicates that the maximum number of CPUs specified exceeds the number of task instances.
Scalability Graph for Intel® Xeon Phi™ Coprocessor Threads Usage
If you select a Target System of Intel Xeon Phi or select Offload to Intel Xeon Phi and check the column Offload to Intel Xeon Phi for the selected site, a gray-green graph appears.
The X axis shows Coprocessor Threads and the program's predicted run-time performance gain appears on the Y axis.
The lines between the graph's gray and green areas is a reference baseline, where the reference CPU chosen to calculate the Intel Xeon processor peak baseline is a dual-socket 8-core Intel Xeon processor E5-26xx product family (2.7 GHz, 16 cores total). When the Maximum Site Gain exceeds this baseline, you might consider using an Intel® Xeon Phi™ coprocessor rather than an Intel Xeon® or similar processor.
Near the top of each vertical line for Coprocessor Threads, you will see a box and a circle that indicate the minimum and maximum predicted gain values. The color shading indicates the likely benefit depending on whether the number of predicted threads would saturate the manycore coprocessor.
This graph shows the predicted parallel performance of the manycore parallel coprocessor without accounting for data exchange amongst Intel Xeon Phi coprocessor cores and the host CPU. For many applications, the number of task instances does not scale enough to fully utilize the many cores of the parallel coprocessor, as indicated by a hover tip. Applications that are not appropriate for a Intel Xeon Phi processing system have values that appear in the gray part of the graph; in this case, try modeling other types of the Target System.
For Each Site, Decide Whether to Modify or Keep Annotations
Use the Suitability Report window to view the predicted parallel performance of each parallel site and its impact on the Maximum Program Gain for All Sites. For example, if a site either has a Site Gain of less than 1.0 or does not contribute to Maximum Program Gain for All Sites, modify or remove its annotations. In contrast, any site that contributes to Maximum Program Gain for All Sites should be kept. For most sites, carefully examine the annotations, overhead assumptions, and related code.
Within the upper-right area of the Suitability Report window, if multiple parallel sites were detected during execution, select a different Site row to display its details.
Implementing Modeling Assumptions Later When Adding Parallel Code
In the lower-right part of the Suitability Report under Runtime Modeling, changing a checkmark does not resolve an issue - it configures the modeling of your proposed parallel program execution. To implement the modeled improvements indicated by the check boxes, use specific parallel framework constructs in the Add Parallel Framework step of the workflow and consider tuning your parallel code.