Performance is an important concern when overlaying a Kate stream on top of a video. This section highlights some of the most time consuming operations that Tiger performs.

Before all, Tiger is a renderer for Kate streams, so the performance of rendering is directly tied to the input data. While Tiger offers a variable rendering quality level, changing it will usually have less effect than changing the Kate stream to make lighter use of slow features.

In addition to the quality level, the Tiger API also offers a way to know if rendering the current state would yield a different image from the previous one. If so, rendering may be unnecessary. It all depends on whether the last render was overwritten (eg, if a new video frame was displayed, which the Tiger renderer has no way to know).

Note also that rendering is affected by whether or not a particular operation is done in hardware or software. Hardware is usually so much faster than software than a very slow operation in software might well become faster than most others if done in hardware. Therefore, the following is only indicative, as measured roughly on a particular setup.

Tiger uses caching to speed up rendering. Static text can be drawn once to a cache, and then reused for the lifetime of the event it appear in. Depending on the complexity of the text, the savings can range from insignificant to considerable. Memory usage should also be considered, as caching is currently naive and uses a lot of memory. Caching may be disabled if memory is scarce.

Most motions will prevent caching. While some some motions such as draw motions will not cause text not to be cached, others (eg, text position) will prevent caching.

The slowest operation of all is alpha blending. While a transparent background for all text may look nice, it's very slow to draw. Also, while static text with alpha can be cached, rendering the cache will also need alpha blending, so will be slow too.

Font outlines can be slow to render as well, since they are drawn by displacing the text slighly and drawing its shape several times.

Text/image size has a direct impact on performance. If possible, select smaller font sizes.

The number of different events visible at the same time also effects performance. Consider delaying events to lessen the maximum number of events onscreen at once.

Inline markup should be kept to a minimum. Font changes are expensive.

Text path is expensive. Not only can text on a path not be cached, but the computations needed to keep text following the path are fairly complex.

Setting a region clipping can improve performance, if text or images would otherwise overlap the outside of the region.

The pixel format of the destination buffer is also important. Tiger can draw to ARGB or ABGR buffers, in host endianness. It is typically much faster to ask Tiger to draw on the correct format than to perform a RGB swapping step afterwards.