UA's own guidance is that DSP plug-ins belong in the DAW only while mixing, because audio inserted on a track round-trips through the hardware. Native versions have no such trip. Everything happens inside your host buffer, which sounds like pure gain until you look at what the buffer is doing.
With a card or an Apollo, the limit was a number you could count. So many slots, so many instances, and a meter telling you when you were out. Native plug-ins remove that meter and hand you a softer limit instead.
You can run as many instances as your CPU handles. That phrasing does a lot of work. Reverb is one of the more expensive things you can ask a processor to do. Two EMTs across twenty tracks is not a trivial ask.
Universal Audio's advice for pops, clicks, and distortion is to start high and work down. Set the buffer to 512 or 1024, then reduce it in steps until problems appear, and back off one notch. Watching your DAW's CPU meter climb is the clearest sign the buffer has gone too low.
The list of things that move that ceiling is short and dull. Processor headroom, whatever else is running in the background, and the rest of your plug-in chain. Universal Audio names the same three factors, which is a polite way of saying the plug-in is rarely the problem on its own.
Sample rate pulls in two directions at once. Higher rates cut latency, which is useful, and they also increase the load from native reverbs leaning on your processor. Running a session at 96 kHz to shave milliseconds can cost you the instance count you were trying to protect.
The numbers behind it are specific. Input Delay Compensation offers settings of off, short, medium, medium-long, and long, worth 0, 100, 200, 300, and 1000 samples. At 44.1 kHz, 100 samples works out at roughly 2.3 milliseconds. UA notes that short and medium settings are generally not noticeable to a performer.
There is a detail worth knowing if you record. On the Console monitoring path, only the minimum delay needed gets applied, so unused headroom costs you nothing. On the DAW record path, the full compensation amount is always added to every input, whether any plug-in needs it or not. Console aux channels add roughly 72 samples on top.
The catch runs deeper than most people expect. A DAW cannot compensate for latency on input channels in real-time. A heavy plug-in sitting anywhere in the session can push your recorded takes late, not only the ones on the track you are recording.
Native plug-ins have no equivalent of that path. If you want a plate in a singer's headphones while they track, an interface with onboard processing is still the way to get it. The native version is a mixing tool, and it is very good at being one.
Printing wet returns to audio is the other lever. Once a plate is doing what you want on a snare, bouncing that return frees the instance permanently. Keep the unprinted version muted underneath if you might revisit it. This is old practice from the hardware days, when a plate was booked rather than instanced.
Send routing does more for your CPU than any setting. One EMT 140 on an aux, fed from eight tracks, costs the same as one EMT 140 on a single track. Eight separate instances cost eight times as much and rarely sound better, because a shared reverb is what glued mixes together in the first place.
That was never a choice on the hardware. There was one plate in the building, three floors down, and everybody sent to it.
With a card or an Apollo, the limit was a number you could count. So many slots, so many instances, and a meter telling you when you were out. Native plug-ins remove that meter and hand you a softer limit instead.
You can run as many instances as your CPU handles. That phrasing does a lot of work. Reverb is one of the more expensive things you can ask a processor to do. Two EMTs across twenty tracks is not a trivial ask.
Buffer size replaces the DSP slot as your hard limit
Slot counting was simple arithmetic, and the EMT pair leaving the DSP card behind replaces it with a number that changes per machine. Your processor, your background tasks, your other plug-ins. All of it moves the ceiling around.Universal Audio's advice for pops, clicks, and distortion is to start high and work down. Set the buffer to 512 or 1024, then reduce it in steps until problems appear, and back off one notch. Watching your DAW's CPU meter climb is the clearest sign the buffer has gone too low.
The list of things that move that ceiling is short and dull. Processor headroom, whatever else is running in the background, and the rest of your plug-in chain. Universal Audio names the same three factors, which is a polite way of saying the plug-in is rarely the problem on its own.
Sample rate pulls in two directions at once. Higher rates cut latency, which is useful, and they also increase the load from native reverbs leaning on your processor. Running a session at 96 kHz to shave milliseconds can cost you the instance count you were trying to protect.
Apollo still wins the one thing native cannot do
Monitoring through Console bypasses the DAW completely, and UA describes it as having no noticeable latency at any buffer size. That is not marketing language about a fast plug-in. It is a different signal path that never enters your host application.The numbers behind it are specific. Input Delay Compensation offers settings of off, short, medium, medium-long, and long, worth 0, 100, 200, 300, and 1000 samples. At 44.1 kHz, 100 samples works out at roughly 2.3 milliseconds. UA notes that short and medium settings are generally not noticeable to a performer.
There is a detail worth knowing if you record. On the Console monitoring path, only the minimum delay needed gets applied, so unused headroom costs you nothing. On the DAW record path, the full compensation amount is always added to every input, whether any plug-in needs it or not. Console aux channels add roughly 72 samples on top.
The catch runs deeper than most people expect. A DAW cannot compensate for latency on input channels in real-time. A heavy plug-in sitting anywhere in the session can push your recorded takes late, not only the ones on the track you are recording.
Native plug-ins have no equivalent of that path. If you want a plate in a singer's headphones while they track, an interface with onboard processing is still the way to get it. The native version is a mixing tool, and it is very good at being one.
Higher buffers and committed tracks buy back headroom
Timing stops mattering once you are past tracking. That is the moment to push the buffer up rather than leave it where it sat for overdubs. Most people forget. They mix an entire record at 128 samples and blame the plug-ins.Printing wet returns to audio is the other lever. Once a plate is doing what you want on a snare, bouncing that return frees the instance permanently. Keep the unprinted version muted underneath if you might revisit it. This is old practice from the hardware days, when a plate was booked rather than instanced.
Send routing does more for your CPU than any setting. One EMT 140 on an aux, fed from eight tracks, costs the same as one EMT 140 on a single track. Eight separate instances cost eight times as much and rarely sound better, because a shared reverb is what glued mixes together in the first place.
That was never a choice on the hardware. There was one plate in the building, three floors down, and everybody sent to it.
