When It Comes To Viewperf,

When It Comes To Viewperf,
Don't Compare New to Old

by Rob Putney

Regular readers of the OPC Review section of The GPC News have noticed that there is now a new version of Viewperf and new numbers assigned to the CDRS, DX and DRV viewsets. One astute reader, Greg Weiss of D.H. Brown & Associates, noticed that the issue of performance comparisons between different versions of viewsets had not been clearly addressed in The GPC News. We hope this article clears up any confusion.

In general, only results with the same viewset name and number are comparable. The change in a viewset's number indicates that code revisions were made. Depending on the nature of the revisions, the effect on a system's performance can be slight or major.

Here are examples of some of the changes made in the latest version of Viewperf (5.0):

In addition to changes in Viewperf that affect viewsets, the viewsets themselves are sometimes revised to more accurately reflect how an application is used in the real world. Here are specific changes that have been made to the CDRS, DX and DRV viewsets:

CDRS

The rendering mode in CDRS-03 was changed from independent vectors to line strips to more accurately reflect the mode used by the application. In many systems, this caused an increase in performance, depending on display list optimization and how the system implements GL_LINES and GL_LINESTRIP. The change affected test 1, which has a weight of 50 percent, and test 7, which has a weight of 0 percent. Tests 2 through 6 are identical and the results should be comparable. But, this is a different version of the code, and even though the inner rendering loops shouldn't have changed, there could be performance variations of a few percent or more. And, because of the changes in test 1 and test 7, composite numbers and price/performance data are not comparable between CDRS-03 and previous versions.

DX

Results for DX-03 should be roughly comparable to those from previous versions. Neither the tests nor the dataset have changed. It would not be recommended to make direct comparisons of results between the current version and earlier ones, but the data could be used as a rough gauge of performance trends.

DRV

DRV-04 has undergone major changes that make it incomparable to previous versions. The biggest change is the addition of walkthrough simulations, which are now supported by Viewperf 5.0. The walkthrough simulations provide a better representation of how the application is used in the real world.

The OPC project group realizes the value of continuity for determining performance trends and making historical comparisons, so the decision to add functionality to Viewperf and to update the viewsets did not come lightly. On the whole, the OPC group feels that the upgrades provide a benchmark and test files that are more representative of real-world performance. Now that these major changes have been made, the group is not anticipating a break in continuity anytime soon.

Rob Putney is the former vice-chair of the OPC project group and an OpenGL development manager for IBM.

[GPC Home] [OPC Project] [PLB Project] [XPC Project]