In direct graphics output to the display, we noticed that the results deviated from the average here and there, despite identical hardware and identical test conditions, but then later leveled back to the same results again. We tried to reconstruct this paradox and things pointed us to the Task Manager.
After our first two articles, we received countless emails reporting connections between the irregular test results and having the Task Manager open. Therefore, we chose the term ‘Task Manager paradox’ as a working title for this issue. We used our test system to compare the overall results in our Tom2D benchmark, using cards from both graphics companies. We did five runs on each, with and without the Task Manager open. The results confirm the initial assumptions quite clearly:
This simple test shows that having the Task Manager open leads to reproducible higher benchmark results for both graphic cards. We found that the consolidated 2D performance increases of the Nvidia card were greater than those of the ATI card. We realize that this statement is way too general, so we had a closer look at each of the benchmark subcategories. Test candidate number one is our ATI Radeon HD 5870:
Only the two values for the polygons and the stretching are considerably different on the Radeon HD 5870. The other results are at approximately the same level as without the Task Manager. As seen in the previous graph, the overall combined difference is so small that it is irrelevant.
Next, we tested the Nvidia card and were curious to find out what benchmark sub categories resulted in the 17% overall performance increase. The results are somewhat surprising:
Compared to the Radeon HD 5870, the GeForce GTX 285 sees an even higher performance increase in the polygon and stretching tests, as well as a staggering 72% extra blitting performance.
But what is the explanation behind the very significant performance improvements in some of these tests? Looking at the features that Windows 7 can actually accelerate in hardware, only blitting and stretching are affected. The polygon performance, which showed a nice bump as well, does not fit in here. And the Task Manager in itself is a bad explanation. Clearly, there has to be another reason.