A GUS can not do software mixing on the back end.
But if you have multiple source channels playing at the same time you have several options:
1. You use software to mix the channels to stereo and use the card as stereo out.
1.1. The output is done with a GUS that a small buffer is transferred to the cards RAM (DMA or PIO) and the GF1 plays from there, upconverting to 44.1 kHz
1.2. The output is done with a SB that a small buffer from memory is played via DMA.
2. You use the hardware to mix the channels and let the DSP mix to stereo.
2.1. The output with a GUS is straightforward, you transmit a small buffer for each source channel and activate all for play, the GF1 will mix it and put out stereo.
2.2. This kind of output is not possible with an SB, as the card is limited to 2 channels only.
Now with the usual UltraMID functionality that a lot of games use (AIL) a small part of the GUS onboard memory is kept free and the remaining is filled with the instrument patches data. A fixed number of channels gets reserved for Music (24 to 30) and 2 channels for digital playback.
In such setup the GUS can play only 2 channels, thus stereo, as digital playback. Several digital playback channels require software mixing.
Of course you may reserve only 24 channels for music and have 4 stereo playback channels left that could use hardware mixing. But I strongly doubt that any game use this, because this would mean:
- You have to reserve a buffer in the GUS onboard memory for each digital channel, reducing the available RAM for instrument patches.
- You have to have support by the API which would be more than a classic Soundblaster could do, so it would be special.