How much memory does sfz take?
sfz features four loading modes, and the amount of memory a SoundFont file can take depends on the loading mode selected, as follows:

SF32
In this mode, sfz will attempt to load the whole SoundFont file to memory at once, and will convert all samples to 32-bit during the loading operation.
This scenario results in optimum performance and flexibility: Bank/Program change can be sent to sfz while voices are playing, without interruptions, clicks or any kind of artifacts.

However, this requires a memory amount of around 2.5 times the size of the SoundFont file. In other words, in order to load a 100Mb SoundFont file, for example, you will need about 250Mb of free RAM. For computers with a lot of RAM, or for small SoundFont files, this is the suggested setting.

If the required amount of memory for the 32-bit storage isn't available, sfz will switch to SF16 automatically, and will load all samples as 16-bit, performing the conversion to 32-bit 'on-the-fly' during playback.
Using this method, memory usage is cut by half, but performance will suffer as the conversion process takes some extra CPU cycles.

Only processors featuring SSE accelleration can use SF32 mode.

SF16
sfz will load the whole SoundFont file, but all samples will be loaded and played as 16-bit. This mode has the advantage of taking half of memory space than SF32, but the processor utilization will be higher due the realtime conversion process.

PR32
When in this mode, only the samples required by current preset will be loaded from disk, and converted to 32-bit during the loading operation. As a result, only the memory required for those samples will be used, which can drastically the RAM required to play certain SoundFont files.

In this mode, when a Bank/Program change message is received (or when changing programs with the interface selector), all voices will be inmediately stopped before loading the next set of samples. Consequently, changing programs during playback could result in a undesirable click.

Only processors featuring SSE accelleration can use PR32 mode.

PR16
This mode is similar to PR32, but the samples are loaded as 16-bit. This is the setting that takes the less RAM amount, suitable for old computers.

DTuna