Share your fastest high quality set up

I've been running a MOTU PCIe-424 with a card in an i7 6700k for years. It's hard to beat the low-latency performance of PCIe with a USB interface but other bus types may do better. And with PCs, good low-latency streaming performance is highly depndent on a machine with low and stable Deferred Procedure Call (DPC) latency, preferably under 200us as measured by something like Resplendence LatencyMon.

By itself, Pianoteq runs flawlessly on this older i7 machine with a 64-sample buffer and a measured RTL of 217 samples = 4.5ms at 48kHz - figure half that or less for output latency only with soft synths. I use a mioXL MIDI interface that delivers a low input latency of about 1ms. So, including a 1ms of keyboard scan time, I figure I'm getting around 4ms from the time I hit a key to the time audio is coming out of my monitors, plus another 3-4ms for it to get to my ears through the air. I feels totally instantaneous.

Because I generally host the Pianoteq VST in Sonar and layer it with a hardware digital piano that's subject to input latency, I actually have to delay Pianoteq's output to sync them, and lately I've found the whole setup feels more piano-like with a 128-sample buffer. I'm also using a second instance of Pianoteq as a resonance FX on the digital piano plus reverb and other FX in Sonar, plus I might be playing along with accompaniment from other synths and that larger buffer gives me some extra insurance against late buffers.

It's nice to be able to get really low RTL for monitoring through DAW software and working with hardware synths that are subject to MIDI transmission and reponse delays as well, but it's less important when the audio is coming from a single standalone synth inside the box that isn't subject to input latency (typically higher than output latency because it requires less processing to sample A to D than to render D to A).

I have to say I doubt many performers would be sensitive to the 0.5ms difference when the overall latency from key strike to ears (assuming open-air monitoring) is on the order of 7-8ms. At 48kHz, I can feel the 128 samples = 2.7ms round-trip difference of changing the buffer size from 64 to 128 when monitoring the digital piano through the DAW because it's a signifcant fraction (approx. 1/3) of the overall latency. But 0.5ms...? I don't think so.

Even if a performer could reliably tell the difference in blind testing, I doublt it would significantly affect their control once they get accustomed to a given response. The effective latency and "jitter" in the response of even the best grand pianos is certainly no better than this, and no one seems to have any problem with it.

Yep. The 4.263ms figure comes right out of RTL utility. And the even more relevant measurement is the delay between internal digital piano sound (supposedly has virtually zero latency, zero jitter) and pianoteq. I did this with a flat maximum velocity curve on both systems, and hammer delay set to off on the digital piano, because pianoteq also has velocity dependent hammer delay (up to 50ms for the weakest note), and cannot be turned off. The 0.5ms difference I'm talking about, I feel is almost JND. Note that 2.7ms noticable difference round-trip is actually 1.35ms difference in key press to sound, so 0.5ms may well be plausible? It seems like the velocity curve (and hammer delay) has synergistic effect with the latency, and the small difference in latency somehow registers as a blow distance change, or lost motion on my fingers. Which means best latency may not be lowest. Strangely small latencies sometimes gets me put extra mental effort into playing because my brain confuses the hammer motion felt by the finger and the implied hammer motion by the sound. For large latencies (20ms+) I can just tune out the sound and focus on the felt part, and the sound can turn out half-way decent but that's definitely not I want when actually playing.

Even direct-monitorng the DP with headphones under Local Control will have some delay due to keyboard scan time, audioe engine reponse time and D/A conversion - probably on the order of 2-3ms total.

Based on my experience with layering Pianoteq with a DP, I'm pretty sure Pianoteq does not do this. I suspect what you're seeing/hearing is just Pianoteq responding to delayed MIDI from your DP at low velocities when used as a controller.

I'm getting the full 2.7ms RTL difference because I'm input-monitoring the DP (under Local Control) through Sonar while Pianoteq is subject only to the output latency (and MIDI transmission delay), and has to be delayed with a plugin when layering the two).

That's interesting. I'm pretty sure I've never played any hardware DP or VSTi that does this. But, as mentioned, I do find that the extra delay of runnning a 128-sample buffer makes the reponse feel more natural and agree that lower is not necessarily better. But total latencies over about 15ms can get to be bothersome, if not unmanageable, to me.

Various dp models incorporate hammer delay into their sound and it's a adjustable parameter. Pianoteq has fixed dynamic-hammer delay relationship, this can be shown by use a single midi file, and two pianoteq vst instances in daw two tracks, so that one inverted phase nulls the other. Set one instance's velocity curve to constant max velocity and the other to constant lowest non-zero velocity, and generate audio output. You can see the weakest note has ~50ms delay compared to strongest, whilst mid-velocities have a few ms delay. iirc the devs had a post long ago addressed the hammer delay issue in the past which some felt was too large, reduced it a bit, but hadn't make it user adjustable. There also appears to be another delay related change in the change log. Most sample libraries don't do hammer delay at all. Layering pianoteq with them always have small misalignment at weakest velocities.

For measuring the time difference between pianoteq and internal sound, I connect internal sound into hsdp-9632 left channel line in and connect line out to right channel. In Reaper, I use one track for midi input to control pianoteq vst with send to master and another track to collect audio input without send to prevent feedback. In this way the recording latency is cancelled out. The scan time and dac latency in internal sound is probably irrelevant because dp manufacturer has various ways to compensate, like adjusting the sensor position and key dip so that the latency will perfectly match that of the acoustic, especially considering the key travel takes ~100ms so adjusting room is plenty. This brings some interesting point, a hybrid piano with real hammer action may have different sensor position than conventional digital, or that folded action keyboard may be different from non-folded, pianoteq may have applied the additional hammer delay on top of inherent hammer delay of the hybrid. This highlights the need of customizable hammer delay paramter, for different types of keyboards.

I was able to improve performance on the stand alone PT 8.4 to 96khz 256 samples on the USB-basee MOTU M2 by setting minimum CPU to 50%. I'm not sure this is not going to affect CPU life and there is a problem where the mouse will lag after an hour or two. Probably the CPU cooling itself  In reaper, multitracking with other VSTs and audio and effects etc I'm still at 48 128