Topic by SteveLy

  • Avatar
  • SteveLy
  • Member
  • Offline
  • From: Melbourne, Australia
  • Registered: 16-11-2015
  • Posts: 377

Topic: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

This might be helpful for some...
_Summary_: Turn Intel Speed-up off in BIOS = Pianoteq CPU performance index increase from 30 to 80.

_Details_:

I recently acquired a Dell Latitude XT3 laptop/tablet hybrid. CPU specs are modest by today's standards: i5-2520M 2.5GHz.

With default BIOS settings, running Manjaro Linux (fresh recent stock install), Pianoteq 6 could handle 44.1 kHz, 48-note polyphony, 128 sample buffer with only the occasional CPU overload when "torture" tested. Pianoteq gave a CPU performance index of 30.

This is the important bit:
*Turn Intel Speed-up off in BIOS*
(I also turned Turbo Boost off - just because I don't like it, fan spinning up etc.)

... after which the performance index went up from 30 to 80. Now running Pianoteq 6 at 48 kHz (41.1 also works fine), 128-note pp, 128 sample buffer and no matter what I throw at it, I cannot trigger a CPU overload. Buffer of 64 samples (for a 1.3 ms latency, which is overkill) is also very playable but will trigger CPU overload when pushed hard.

Note: "Intel speed-up" is a misnomer. It should be called "Intel slowdown", because what it really does is lets the CPU clock drop when the system thinks there is less to do and speeds it back up as needed. But apparently not quickly enough to keep up with Pianoteq.

Other details:
Multicore rendering is on (both in BIOS and Pianoteq).
Hyperthreading (in BIOS) is on (though I doubt it makes any difference).
Audio device is ALSA with output PulseAudio Sound Server.

3/2 = 5

SteveLy's Website

Reply by stamkorg

  • Avatar
  • stamkorg
  • Member
  • Offline
  • Registered: 05-04-2014
  • Posts: 452

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

Yes,
These options (intel speed step, speed-up, power save or boost) should always be turned off to avoid variations of the frequency at which the CPU works. The variations cause throttling and are responsible for audio dropouts.
By turning all these off, the CPU stays stable and at a fixed frequency.

Reply by steve200

  • Avatar
  • steve200
  • Member
  • Offline
  • Registered: 20-12-2017
  • Posts: 59

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

Very interesting.  Thanks.

Will this be relevant to my three year old Lenovo T440s?
i5-4300U CPU @ 1.9GHz, M5/S1/F6, x4 (12G ram)

If i put Win 10 into High Performance mode, I get an effective 1.8GHz.   Would I still benefit from playing with the Bios?

If so, how do i do it?

steve

Reply by grazie

  • Avatar
  • grazie
  • Member
  • Offline
  • From: UK
  • Registered: 19-12-2016
  • Posts: 15

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

Always good when folk post really useful info like this. Thanks!

Reply by music_guy

  • Avatar
  • music_guy
  • Member
  • Offline
  • Registered: 18-07-2017
  • Posts: 45

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

I think this is dependent on computer system so requires that users test different configurations.

With laptops, disabling SpeedStep or SpeedShift schemes often locks out turbo speeds (to control heat) so CPU performance drops a lot. Not always the case so you need to check.

On my laptop:

- For recording and other situations that require no dropouts, keeping processor frequency stable can help reduce clicks but requires a bit higher latency

- For home practice where lowest latency might be most important, I seek highest CPU speeds with few dropouts. Another trick I use in PianoTeq to reach very low latency is-  Options>Perf>uncheck "CPU Overload Detection"

Reply by SteveLy

  • Avatar
  • SteveLy
  • Member
  • Offline
  • From: Melbourne, Australia
  • Registered: 16-11-2015
  • Posts: 377

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

music_guy wrote:

With laptops, disabling SpeedStep or SpeedShift schemes often locks out turbo speeds (to control heat) so CPU performance drops a lot. Not always the case so you need to check.

Unless you're overclocking turbo boost is typically no more than 10-20% increase in clock speed. Fortunately on my systems (laptop and desktop) I can independently set speed-step/up or throttling and turbo boost. Though as I noted in OP I turn the latter off anyway. These things can also be controlled via software but setting it in BIOS makes it a lot simpler.

music_guy wrote:

- For home practice where lowest latency might be most important, I seek highest CPU speeds with few dropouts. Another trick I use in PianoTeq to reach very low latency is-  Options>Perf>uncheck "CPU Overload Detection"

Thanks for that tip. I never tried that before. However, 2.7 ms latency is already equivalent to a sound source that's less than a metre (or about three feet) away from the listener, so a six-foot grand would have considerably more natural latency than that.

3/2 = 5

SteveLy's Website

Reply by music_guy

  • Avatar
  • music_guy
  • Member
  • Offline
  • Registered: 18-07-2017
  • Posts: 45

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

SteveLy wrote:

Thanks for that tip. I never tried that before. However, 2.7 ms latency is already equivalent to a sound source that's less than a metre (or about three feet) away from the listener, so a six-foot grand would have considerably more natural latency than that.

Hi SteveLy,

I'm sure you know this but a few quick comments to add to the discussion.

The "PianoTeq latency" of 2.7ms represents a buffer of 64 samples and sample rate of 48KHz (2.7ms = 64/48,000). However, that only represents a portion of the total latency that is relevent for a piano player. A round-trip latency is probably more relevant, say the time it takes from hitting the key until sound hits the eardrum.

Quick-and-dirty "real-world" latency is probably closer to 10ms:

~1ms for digital piano processing of keystroke and transmitting info in slow MIDI protocol
~2.7ms for "Pianoteq latency"
>4ms for computer in/out buffers & DAC processing
~3ms sound travels from speakers to ears (speed of sound is about 3ms per metre)

This performance requires a fast computer with OS optimizations and good ASIO drivers (e.g. RME). (A simple way to confirm/estimate real-world latency is to put a microphone at the piano seat and hit a few keys and estimate the latency between keystroke and sound output in Audacity charts.)

PreSonus had a neat diagram which illustrated many of the system latency hurdles; the diagram is a bit dated and of course a marketing pitch but is helpful.

http://proaudioblog.co.uk/2015/09/usb-3...nterfaces/

An uber expensive computer can only reduce the "Pianoteq latency" so much so we need to look to different areas to reduce real-world latency. Headphones are cheap and can shave ~3ms latency. Other options can help but start to get expensive for top-performance (better interfaces with better ASIO drivers, move from USB to PCIe, etc.).

Also there is a point where reducing latency provides no perceptable benefit to the musician. That varies by person and probably by instrument (with drums and drummers probably being extra sensitive). For typical musicians, I have seen 5ms to 10ms as informal limits posted in reasearch, marketing materials and forums but have not seen any conclusive research there.

Regardless, I suspect jitter is more important than latency but that is really tough to measure. My sense is that is where PianoTeq really shines compared to sampled VIs and that is why PianoTeq "plays so well."

Last edited by music_guy (25-01-2018 06:36)

Reply by SteveLy

  • Avatar
  • SteveLy
  • Member
  • Offline
  • From: Melbourne, Australia
  • Registered: 16-11-2015
  • Posts: 377

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

Thanks for the very informative reply re latency. You clearly know more about this than me (I mean it - not being sarcastic - just saying because one can never tell on forums like this ).

Btw I get a quoted latency of 2.7 ms quoted for 128 sample buffer, which does check out at 48 kHz sample rate (128/48kHz = 2.67ms). I understand that is the bare minimum possible for the buffer size, not the actual latency. (I used to be able to gauge the latency directly when I had piano keyboards that had on-board presumably instant-on samples. But now I only have controllers.) With Pianoteq (Linux version) I find the 2.7 ms setting very satisfactory.

I tried Pianoteq back in the early days (version 2 or 3?). I really liked it but did not buy it for years, until well into version 5, because I could not get the latency down with the hardware and lack of know-how I had back then.

Btw I spent my early teens living in a church and playing pipe organ. The base latency there was over 0.1 s. For the larger pipes (used for pedal esp) it was even more as the air column inside the pipes takes time to start resonating. Somehow I got used to that so much that I would not even be conscious of it after a while. But with piano emulation it's a different story.

Last edited by SteveLy (25-01-2018 16:39)
3/2 = 5

SteveLy's Website

Reply by music_guy

  • Avatar
  • music_guy
  • Member
  • Offline
  • Registered: 18-07-2017
  • Posts: 45

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

SteveLy wrote:

With Pianoteq (Linux version) I find the 2.7 ms setting very satisfactory.

This is all that matters. Job done!

SteveLy wrote:

Thanks for the very informative reply re latency. You clearly know more about this than me (I mean it - not being sarcastic - just saying because one can never tell on forums like this ).

Latency for virtual instruments is incredibly complex so can be an endless sinkhole. We all have plenty to learn!

SteveLy wrote:

Btw I spent my early teens living in a church and playing pipe organ. The base latency there was over 0.1 s. For the larger pipes (used for pedal esp) it was even more as the air column inside the pipes takes time to start resonating. Somehow I got used to that so much that I would not even be conscious of it after a while. But with piano emulation it's a different story.

Real instruments and real halls have some factors that still are difficult to emulate. Both for passive audiophiles and active VI musicians. There is plenty of progress to be made.

Reply by peterws

  • Avatar
  • peterws
  • Member
  • Offline
  • From: North of England
  • Registered: 26-04-2016
  • Posts: 199

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

stamkorg wrote:

Yes,
These options (intel speed step, speed-up, power save or boost) should always be turned off to avoid variations of the frequency at which the CPU works. The variations cause throttling and are responsible for audio dropouts.
By turning all these off, the CPU stays stable and at a fixed frequency.

One further question; my laptop is used exclusively for Pianoteq, and along with the piano, never gets switched off, although the screen is allowed to go blank after 10 minutes inactivity.
Might this be a problem; would the disc be spinning madly around all this time?

I'm playing all the right notes but not necessarily in the right order

Reply by groovy

  • Avatar
  • groovy
  • Member
  • Offline
  • Registered: 12-06-2009
  • Posts: 717

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

music_guy wrote:

Another trick I use in PianoTeq to reach very low latency is-  Options>Perf>uncheck "CPU Overload Detection"

Hello music_guy,

seems to have no significant effect on latency.

I checked this quick and dirty by microphone-recording the sound of my fingernail hitting a key and the sound of pianoteq arriving from my speakers. The timespan between both soundevents is something like an "overall-latency" of the entire chain.

In this experiment the overall-latency was around 45 ms on my system, independent of CPU Overload Detection being On or Off.

1st track: Fingernail hitting the key, Pianoteq disabled.
2nd track: Pianoteq now enabled with overload detection OFF.
3rd track: Pianoteq enabled with overload detection ON.

https://s31.postimg.org/qkd9ismh7/pianoteq-cpu-overload-detection-off-vs-on.png

Testtone: Steinway D Prelude (unmodified preset), key C2, Pianoteq Standard v6.1.1.

Reply by music_guy

  • Avatar
  • music_guy
  • Member
  • Offline
  • Registered: 18-07-2017
  • Posts: 45

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

groovy wrote:
music_guy wrote:

Another trick I use in PianoTeq to reach very low latency is-  Options>Perf>uncheck "CPU Overload Detection"

seems to have no significant effect on latency.

Works on my system as it allows me to select more aggressive settings on the RME ASIO drivers (e.g. lower buffer or higher sample rates) without clicks and pops. Everyone's system will differ.



groovy wrote:

I checked this quick and dirty by microphone-recording the sound of my fingernail hitting a key and the sound of pianoteq arriving from my speakers. The timespan between both soundevents is something like an "overall-latency" of the entire chain.

In this experiment the overall-latency was around 45 ms on my system, independent of CPU Overload Detection being On or Off.

1st track: Fingernail hitting the key, Pianoteq disabled.
2nd track: Pianoteq now enabled with overload detection OFF.
3rd track: Pianoteq enabled with overload detection ON.

https://s31.postimg.org/qkd9ismh7/pianoteq-cpu-overload-detection-off-vs-on.png

Selecting Options>Perf>uncheck "CPU Overload Detection" alone will not change latency. But it allowed me to use more aggressive settings on the RME  ASIO drivers. Apologies for not making that more clear.

Reply by SteveLy

  • Avatar
  • SteveLy
  • Member
  • Offline
  • From: Melbourne, Australia
  • Registered: 16-11-2015
  • Posts: 377

Re: easy huge performance boost 2nd gen i5 Intel laptop w/ Pianoteq

Interesting, but the problem is usually not with just one note you hit and let sustain, but when you are playing complex passages, using sustain pedal (and maybe pitch bend if you go "organish") and suddenly there is a lag or even a glitch, and then it all returns to normal. It's very annoying at home and something intolerable when playing live at a gig.

3/2 = 5

SteveLy's Website