Performance

It isn't fast enough! - You are impatient.

Remind that you are not on a workstation with dedicated hardware. This porgram does not ship with a PCI-card and you are using it probably on your powerbook with a harddisk which is slower than a RAID 0 diskarray.

We know that speed is a concern for you and try to find fast solution both for rendering and user interaction.

In the mean time, speed also depends on how you configure Belle Nuit Imagefilter. There are some settings, which can increase performance by a factor of more than ten (or decrease, if used wrong). The good news is that you can change these settings yourself:

• In the interface, work with half and quarter resolution as long as you can. In quarter resolution, the node has to process only 1/16 of the pixels. The node are aware of the resolution and adapt, except the filter nodes which always work at full resolution.
• With half and quarter resolution, however, the disk access time for the image sequence and the quicktime node is still the same. If you plan to stay longer in the interface, consider rendering proxies. This is very easy from within Imagefilter. Just connect an output node to the input node, set half or quarter resolution and render it. Work with a node with the new folder and just reconnect to the old node wehen you go definitive.
• Set your RAM cache big enough. All rendered images and intermediate results are cached, so that on a change only the changed nodes and their outputs must be reprocessed. Set it big enough for your needs, but do not cheat. You can set it larger than the real RAM, but then it will use virtual memory and swap and become very slow.
• The file cache is the second net to cache images. It makes only sense when it is larger than the file cache, because images for the RAM cache are also stored in the file cache. If the file cache speeds up your processing depends on the speed of your harddisk and the complexity of your nodes. For simple color correction and compositing reporcessing may be faster than reading the disk. Therefore you can also turn the file cache off.
• You also may generally better turn off the file cache on long renders. But not always. If you render to multiple nodes the result of a very slow node, it is better if it is cached.
• When you render to multiple output nodes, Imagefilter renders by offset and not by node. This way it can use intermediate results which cached. Set the RAM cache enough big to accomodate such a scenario.
• Normally, you want to work with Auto-Refresh so that you do not have to select Refresh on every parameter change. But sometimes, Auto-Refresh acts to fast, while you are still fiddling with the parameters. This is where the Render Wait Time comes in. This is the time in milliseconds the program will wait after a parameter change to render. Set it small enough to have a reactive system and big enough to not be overwhelmed. Note that the program always renders immeditalye when you change the offset to allow you to compare pictures fast.
• In the initial design, we had this fancy chasing arrows, which dance from node to node. While they show didactically the workflow from node to node, they slow down the computer lot. On my powerbook, every time i show them, they let the program 0.3 seconds! This is whay we have now the option Interface update time. This time in milliseconds defines the time a node runs silently, before the chasing arrows are shown. You will not get a feedback from every node, but everything runs faster. The "Transformations" example on my powerbook runs ten times faster with a interface update time 300 than 0. So if you want to see every node, use 0. For normal work, use 300-500. For rendering, set it to 1000 or 2000.
• Finally note that you can measure the time a Refresh needs: With the Debug Profile Option the Console shows timestamps of each important step of the Refresh. This way you can see if the changed settings improve your perfomance. Note: Use a Force Refresh to ensure that you measure the whole refresh and do not use cached intermediate results.