32-bit float shows up in two places for hardware producers. In your DAW, it is the session format: Logic, Reaper, and Pro Tools all run their internal processing at 32 or 64-bit float. For hardware, it is the format the Make Noise Morphagene requires and nearly everything else rejects. Knowing what the format actually does is the useful part; the practical decisions follow from there.
Integer PCM vs. floating point
Standard PCM audio stores each sample as a signed integer. In 16-bit PCM, each sample is a number between -32,768 and 32,767, giving you 65,536 possible values across the full amplitude range. 24-bit expands that to roughly ±8.4 million values. More bits mean finer quantization steps between silence and maximum amplitude, which translates directly to more dynamic range and lower noise at the bottom of the level range.
The practical ceiling for 16-bit is about 96 dB of dynamic range. 24-bit pushes that to around 144 dB. For reference, the best analog tape tops out around 60-70 dB and a very quiet studio might be 30 dB below the loudest sound you can record cleanly on a good mic. In other words, 24-bit exceeds what most audio chains can actually deliver, and 16-bit is sufficient for the vast majority of playback applications.
32-bit float is a fundamentally different representation. Instead of an integer, each sample is an IEEE 754 single-precision floating point number. The key property: floating point can represent values above the nominal 0 dBFS ceiling without distortion. A peak of 1.4 (which would clip and truncate in integer PCM at full scale) stores as exactly 1.4 in 32-bit float. The value sits above 1.0 without truncation.
This is why DAWs use floating point internally. A plugin adding 12 dB of gain won’t clip the signal at 0 dBFS; the values go above 1.0 and stay there until they hit the D/A converter or get written to an integer file. The signal only clips when it reaches hardware output or when you export to integer PCM. That “overflow headroom” is the entire point of 32-bit float in a production context.
Why the Morphagene requires 32-bit float
The Morphagene’s DSP engine does real-time granular processing at audio rates: vari-speed pitch shifting, grain manipulation, and parameter modulation that can fire at CV speeds up to audio rate. To do that kind of processing on integer PCM would require a format conversion step either on load or inside the signal path, adding latency or placing extra demands on the module’s processor.
By requiring 32-bit float input, the firmware loads each reel directly into its processing chain without any conversion step. The DSP code already operates on floats; the file just needs to be in the same format.
The consequence is that the Morphagene will not load 16-bit or 24-bit files. Not “loads them at lower quality.” Not “shows a warning.” It skips the file entirely. Sweep the Reel knob over a card with a 24-bit WAV on it and that reel simply will not appear.
Before anything goes on a Morphagene card, it needs converting to 32-bit float, 48 kHz, stereo WAV. An existing library of 24-bit files from a DAW session is not directly usable. Neither are the 16-bit files most free sample sites deliver. The conversion does not improve the audio quality (more on that below), but it produces the exact container the module requires.
What converting to 32-bit float does and doesn’t do
Converting a 16-bit file to 32-bit float does not improve the audio. Dynamic range, noise floor, and quantization characteristics are fixed by the original recording. The conversion just re-wraps the same integer values inside a floating point container. You get a larger file with identical audio content.
The headroom advantage of 32-bit float only applies if the source was recorded or processed at floating point in the first place. If your DAW session runs at 32-bit float internally and you export for the Morphagene, the export can carry any peaks above 0 dBFS without clipping. A 16-bit export of the same session would hard-clip anything above the ceiling.
For a sample from a 16-bit or 24-bit source that is already normalized and under 0 dBFS: there is no quality difference between the original and the 32-bit float version. They will sound identical through any playback path that supports the format.
This is worth repeating because the number “32” is larger than “16” and larger numbers in audio often do mean something. Here they mean something specific, not something general. The format is useful when you need to carry headroom through a floating point signal chain. It is not an upgrade to existing samples that are already sitting at or below 0 dBFS.
Other hardware that works with 32-bit float
The ER-301 by Orthogonal Devices goes the opposite direction from the Morphagene: it accepts 16-bit, 24-bit, or 32-bit float WAV files and converts everything to 32-bit float in RAM after loading. A 32-bit float source file skips the conversion step, which means any over-0-dBFS peaks in the file stay intact on load. A 16-bit source loads fine too; the ER-301 just promotes it internally. In practice, this means you can bring your existing 16-bit library to the ER-301 without any extra work, while also being able to pass it 32-bit float files from a floating point session if that is what you have.
Some other Eurorack modules process at 32-bit float internally even when their file format requirements are more modest. The Expert Sleepers Disting NT, for example, accepts 8, 16, or 24-bit WAV but processes everything in 32-bit float internally. For those devices, 32-bit float source files do not hurt if the hardware can parse the format, but the audio result is equivalent to what you would get from a well-prepared 24-bit file.
Outside of Eurorack, consumer hardware rarely touches 32-bit float. CD audio is 16-bit. Most standalone samplers are 16 or 24-bit on their playback engines. The format is a DSP convenience at the software layer.
Where 32-bit float doesn’t belong
The Elektron Octatrack accepts 16 or 24-bit WAV at 44.1 kHz only. A 32-bit float file will not load. The Digitakt takes 16-bit at 48 kHz. The Squid Salmple uses 16-bit, 44.1 kHz mono. None of these devices gain anything from 32-bit float source files, and all of them reject the format outright.
File size is also a real consideration when building out a library. A 32-bit float stereo file at 48 kHz uses 4 bytes per sample, compared to 3 bytes for 24-bit and 2 bytes for 16-bit. That is twice the storage of 16-bit at identical length. On a Morphagene card with 32 reel slots, the overhead is manageable. For a large Octatrack or Digitakt library with hundreds or thousands of files, using 32-bit float would double card requirements for no audible benefit. The format rejection on those devices is not arbitrary; there is genuinely nothing to gain there.
Conversion gotchas
Converting from 32-bit float down to integer PCM is the step most likely to produce problems. If the source has any material above 0 dBFS, those peaks will clip when written to 16 or 24-bit integer PCM. Before converting down, check your peaks and apply a normalize or brick-wall limiter pass. A file that sounds fine at 32-bit float can still export to clipped 16-bit if its peaks were above 1.0 in the float representation. Checking the waveform in the float source and checking the waveform in the converted file is the only way to be sure.
Converting upward from 16 or 24-bit to 32-bit float is lossless. No audio data is destroyed. The values change representation, and any device that requires 32-bit float will load the result without issue. The audio is identical to the source.
One specific workflow gotcha: the Morphagene includes a recording input and records reels at 24-bit internally. If you record material directly into the module and then want to re-use those recordings as source for new reels, the resulting 24-bit files will need converting to 32-bit float before the module will load them back. The conversion is lossless, but the step is easy to miss if you do not expect it.
DAW export settings are also worth checking. Most DAWs let you set the export format independently from the project’s internal bit depth. If your session is configured for 32-bit float internally, that does not mean you should export everything at 32-bit float. Export at the format each target device expects: 32-bit float for the Morphagene, 16-bit for the Octatrack and Digitakt, 24-bit for devices that accept it. Matching the container to the destination is the right move; matching the container to the session format is not. The WAV vs AIFF vs FLAC guide covers the container-choice side of sample format decisions in more detail.
The actual decision
For the Morphagene: 32-bit float, 48 kHz, stereo WAV, every time. The module is designed around that format and refuses anything else.
For everything else: match the bit depth to what the hardware expects. If you’re prepping a mixed library for multiple targets, keep master files at 24-bit (enough quality, reasonable size) and convert to each device’s required format when building a card. A conversion step from 24-bit to 32-bit float is not a quality loss; it is a change of container required by the destination hardware. A conversion from 24-bit to 16-bit is also not a problem for most material in a hardware context, where 16-bit’s dynamic range is more than sufficient for percussion, loops, and the majority of one-shot content.
Most of the confusion around this format comes from DAW users who see 32-bit float in their project settings, assume higher always means better, and then wonder why the Octatrack or Digitakt reject the files. At the hardware playback layer, those devices want integer PCM at the sample rate and bit depth their engines are built around. 32-bit float is not a quality tier above them; it is a different representation suited to a different use case.
If you are building cards for several hardware targets in the same workflow, SampleStack reads WAV, AIFF, and FLAC at any bit depth, converts each file to the exact format its target hardware requires (32-bit float for the Morphagene, 16-bit for the Octatrack and Digitakt), and writes the result to each device’s expected folder layout in one pass.