Topic: Is Intel N3510 ok for pianoteq?

Currently I use pinoteq on desktop with 3.3GHz Core i3 processor, 3MB cache, 2GB RAM, Linux 32-bit (Arch) and have no problems. But I want to buy more slow laptop with processor Intel Pentium Dual-Core N3510 2.0 GHz  (4 cores), 2MB cache, 4 GB RAM. OS: Linux 64 bit (Arch). Is it OK for pianoteq?

I can't test the laptop before buy it, so I ask the community. The requirements from official FAQ is too ambiguous (just "modern" computer)? Thanks.

Last edited by Ross (28-06-2014 16:20)
Combine velocity curves: http://output.jsbin.com/cukeme/9

Re: Is Intel N3510 ok for pianoteq?

The N3510 has a PassMark score of 1682. Your i3 is probably an i3-2120 or i3-3220; it should have a PassMark score of 3874 or 4227, respectively. Therefore the N3510 is roughly 2.5-3 times slower. Since Pianoteq is (as far as I know) capable of using multiple CPU cores, it should work as long as on the i3 the load never exceeds about 30%. On my system (i5-3337U, Win8.1, PassMark 3247), Pianoteq 5 can reach about 50% load at 128 notes polyphony.

So my guess is: it's going to be a bit close, but it should work. Excluding 'extreme' circumstances, the effective polyphony should rarely exceed 64 notes anyway. (In other words, you will probably not 'lose' much if you limit polyphony to that value.)

Last edited by kalessin (28-06-2014 16:38)
Pianoteq 6 Standard (Steinway D&B, Grotrian, Petrof, Steingraeber, Bechstein, Blüthner, K2, YC5, U4, Kremsegg 1&2, Karsten, Electric, Hohner)

Re: Is Intel N3510 ok for pianoteq?

My Avoton (8-core 2.6ghz version of the 3510) is still dead at work so I haven't been able to test it against 5.0.  However, if we were look at Pianoteq 4.5 (see below link), 2ghz would be roughly a score of 280 which puts it in the solidly playable category.  You'd get note dropping if you piped through MIDI files with 100+ polyphony.

http://www.forum-pianoteq.com/viewtopic...33#p932733

It's a decent choice as long as you use mostly for Pianoteq.  Intel did a very nice job in boosting the floating point performance for Silvermont/Bay Trail but I can tell you they did nowhere as good boosting integer performance.  We bought it to use as a new firewall/network appliance -- and it runs like a crippled turtle doing these basic functions -- probably about the equiv speed of a 500mhz i3.

Re: Is Intel N3510 ok for pianoteq?

I would recommend Passmark score 2000+ to be able to play even most demanding music with lots of polyphony.  I was able to use my old Atom netbook (score 600 I think), but I had to lower maximum polyphony and sample rate.

Re: Is Intel N3510 ok for pianoteq?

I have found another laptop with Core i3 3217u processor. It has passmark 2291, but only 2 physical cores and 1.8GHz.

Should I prefer this laptop?

Combine velocity curves: http://output.jsbin.com/cukeme/9

Re: Is Intel N3510 ok for pianoteq?

My wife's laptop uses nearly the same CPU (3227u, 1.9ghz).

Depending on what you are doing, it is anywhere from somewhat faster to way fricking faster than the N3510.

Re: Is Intel N3510 ok for pianoteq?

Ross wrote:

Currently I use pinoteq on desktop with 3.3GHz Core i3 processor, 3MB cache, 2GB RAM, Linux 32-bit (Arch) and have no problems. But I want to buy more slow laptop with processor Intel Pentium Dual-Core N3510 2.0 GHz  (4 cores), 2MB cache, 4 GB RAM. OS: Linux 64 bit (Arch). Is it OK for pianoteq?

I can't test the laptop before buy it, so I ask the community. The requirements from official FAQ is too ambiguous (just "modern" computer)? Thanks.

I see you want to use Arch Linux.
For good performance you should also read/know about things like Jack and configure your system for realtime audio.

https://wiki.archlinux.org/index.php/Pro_Audio

I am quite sure you can make it work. I am using Pianoteq even on my EeePC 900 netbook, the generation just before the Atom processors were introduced. Not fast enough for serious usage, but not completely useless either. For serious pianoplaying I have a newer i5 laptop.

I also use pianoteq on the Linux commandline (with --headless option) to batch-convert MIDI files directly to WAV. In which case speed is not strictly required (but surely welcomed)

Re: Is Intel N3510 ok for pianoteq?

I recently bought a fanless Acer laptop E3-112, Passmark 1835, for 250€, with a Celeron N2940 processor. After installing 32-bit Kubuntu 14.04 and Pianoteq 5.1, I used a MIDI recording of a Rachmaninoff sonata (Op. 36, a wonderful recording by Marouan Benabdallah, http://www.piano-e-competition.com/ecom...006.asp#B) to test this rather puny processor's ability to run PT (with governor "performance", as recommended in README_LINUX). The sonata has some pretty hairy passages, and at 17:14 minutes the 'Current polyphony' jumps as high as 170. How this is possible for an 88-voice instrument I can't guess, but it is a big hurdle for Pianoteq.

On my office machine (core i5-3570K, Passmark 7161) with polyphony set to 256 this passage goes through without a hitch. On the laptop, sampling 48kHz, I have to reduce polyphony to 64 or to Auto (Optimistic). In a blind test I'm not sure I could tell the difference to unlimited polyphony, so it's not something that worries me, but still I would like to find the best settings. I notice that by setting Auto (Optimistic) the current polyphony can jump above 80, but not when I set 64. Anyone know which setting is better?

Here's something else: I added the privileges recommended in README_LINUX to '/etc/security/limits.conf', and starting PT from a terminal I get the message 'Multi-core: got real-time scheduling with priority 65'. I imagine PT is making a system call to get high-priority, real-time scheduling, but when I look at the process table with 'chrt -p <pid>' I see SCHED_OTHER with priority: 0, i.e., what any plain-vanilla program gets. I have to start PT with 'chrt -f 88 ./Pianoteq' to get SCHED_FIFO and priority 88. However, watching Audio load and Current polyphony, I don't see any difference in actual performance.

By the way, in the Rachmaninoff, "sensors" never reports a temperature higher than 58°C, so there's evidently no danger of overheating this laptop with long passages of high-load playing. For about €50 more there's a version of this laptop with a N3540 processor, which might allow a little bit more polyphony. This processor is called a Pentium but it's exactly the same as the Celeron N2940, just runs faster. For me the N2940 is entirely adequate.

Re: Is Intel N3510 ok for pianoteq?

MinorChord wrote:

...The sonata has some pretty hairy passages, and at 17:14 minutes the 'Current polyphony' jumps as high as 170. How this is possible for an 88-voice instrument I can't guess, but it is a big hurdle for Pianoteq.

Polyphony includes sustained notes. Hold down the sustain pedal and start pounding out notes, and your polyphony steadily increases. I ran a test and was able to get well over 100 with random sweeps. The notes do decay and drop out of the count; the highest polyphony I was able to get with one key was around 13.

170 sounds plausible with Rachmaninoff.

Re: Is Intel N3510 ok for pianoteq?

Hi,

MinorChord wrote:

I recently bought a fanless Acer laptop E3-112, Passmark 1835, for 250€, with a Celeron N2940 processor. After installing 32-bit Kubuntu 14.04 and Pianoteq 5.1, I used a MIDI recording of a Rachmaninoff sonata (Op. 36, a wonderful recording by Marouan Benabdallah, http://www.piano-e-competition.com/ecom … 006.asp#B) to test this rather puny processor's ability to run PT (with governor "performance", as recommended in README_LINUX). The sonata has some pretty hairy passages, and at 17:14 minutes the 'Current polyphony' jumps as high as 170.

I have a very similar laptop with Celeron N2930, a similar OS (Debian Jessie with KDE) and similar performance data. The last minute of Benabdallah09.MID is a very good "smoke-test", thank you!

MinorChord wrote:

On the laptop, sampling 48kHz, I have to reduce polyphony to 64 or to Auto (Optimistic).

Yes, with both modes I have no problems, but beginning with maximum polyphony 96 I see red CPU Overload detection at 17:14.

MinorChord wrote:

In a blind test I'm not sure I could tell the difference to unlimited polyphony, so it's not something that worries me, [...]

No worries for me too. In the past I always used max. polyphony 48, just because it was the default setting. I will now elevate to 64 after this smoke-test, just because it is possible :-)

MinorChord wrote:

but still I would like to find the best settings. I notice that by setting Auto (Optimistic) the current polyphony can jump above 80, but not when I set 64. Anyone know which setting is better?

Good question. I have more confidence in static settings. But this is not rational and the algorithm 'Auto (Optimistic)' shows higher polyphony on my system too.  Does it cost latency? I don't know.

MinorChord wrote:

Here's something else: I added the privileges recommended in README_LINUX to '/etc/security/limits.conf', and starting PT from a terminal I get the message 'Multi-core: got real-time scheduling with priority 65'. I imagine PT is making a system call to get high-priority, real-time scheduling, but when I look at the process table with 'chrt -p <pid>' I see SCHED_OTHER with priority: 0, i.e., what any plain-vanilla program gets. I have to start PT with 'chrt -f 88 ./Pianoteq' to get SCHED_FIFO and priority 88. However, watching Audio load and Current polyphony, I don't see any difference in actual performance.

Yes, no effects from fiddling with 'chrt' on my Debian kernel (3.16x amd64 with default scaling_driver 'intel_pstate' and set to governor performance).

MinorChord wrote:

By the way, in the Rachmaninoff, "sensors" never reports a temperature higher than 58°C, so there's evidently no danger of overheating this laptop with long passages of high-load playing. For about €50 more there's a version of this laptop with a N3540 processor, which might allow a little bit more polyphony. This processor is called a Pentium but it's exactly the same as the Celeron N2940, just runs faster. For me the N2940 is entirely adequate.

May I ask, what performance index you have with your N2940? With a N2930 I see index ~20 with 48kHz and 64 samples buffer.

Thanks!

Re: Is Intel N3510 ok for pianoteq?

In answer to groovy's question: on my Acer E3-112, set up as described, I get Performance=21. On my office machine (intel i5-3570K), *without* realtime privileges, I get 28 and have no problems with the last minutes of the Rachmaninof sonata, with Polyphony at 256, so I've never bothered to modify /etc/security/limits.conf. On a 10-year-old Dell laptop, with a Pentium M, 1.8GHz (PassMark = 391!) using a setup I described on a different posting, I get Performance = 14, which is marginal but quite usable with reduced polyphony and sampling rates.

Re: Is Intel N3510 ok for pianoteq?

MinorChord wrote:

In answer to groovy's question: on my Acer E3-112, set up as described, I get Performance=21.

Thank you. Then there seems to be no significant difference between 32 and 64 bit within that CPU familiy.

Re: Is Intel N3510 ok for pianoteq?

Update on Acer E3-112 performance

Here's an update to my posting of 03.03.15 in this thread on how I'm running Pianoteq on my lightweigt Acer E3-112 laptop under Xubuntu Linux (should work with all Ubuntu varieties). I've gotten Performance index up to nearly 23, and I have a nice-looking program launcher that sets everything up.

When I'm using the laptop for other tasks, especially on battery, I don't want maximum performance from the processor, but when when I start Pianoteq, I want to suspend unnecessary background tasks and change to the power management regime "performance" to get the most polyphony and audio bandwidth I can. To do this from a shell script you normally need to get administrator privileges (e.g. with gksudo), but you can circumvent this with by adding a line in the file /etc/sudoers (use sudo visudo -s) that looks like this, assuming your user name (whoami) is "pianoplayer" :

  pianoplayer  ALL  = (root:root) NOPASSWD: /usr/bin/cpufreq-set

This allows you to use sudo for cpufreq-set without entering a password, even in a script.

I then wrote a short script that turns off networking and revs up the power management regime only when I start Pianoteq. You might also want to suspend other background services, like file indexing, that you don't need while practising. Use Task Manager or ps to see what other processes are using noticeable cpu time.


#! /bin/sh
# deactivate Gnome NetworkManager (nm-applet)
nmcli nm sleep true
#
# put all cores in "performance" regime
#   requires the line
#     pianoplayer  ALL  = (root:root) NOPASSWD: /usr/bin/cpufreq-set
#   in /etc/sudoers (use sudo visudo -s)
sudo cpufreq-set -c 0 -g performance
sudo cpufreq-set -c 1 -g performance
sudo cpufreq-set -c 2 -g performance
sudo cpufreq-set -c 3 -g performance
#
# start Pianoteq with first-in, first-out, real-time priority 88
#   assumes user is member of group audio and that the line
#     @audio - rtprio 90
#   appears in /etc/security/limits.conf
chrt -f 88 '/path/to/my/Pianoteq 5'
#
# reactivate NetworkManager
nmcli nm sleep false
# return to power saving regime
sudo cpufreq-set -c 0 -g powersave
sudo cpufreq-set -c 1 -g powersave
sudo cpufreq-set -c 2 -g powersave
sudo cpufreq-set -c 3 -g powersave


To make a nice-looking launcher icon, I extracted the Pianoteq logo using a trick I described in a different posting (09.03.5) and made a desktop laucher icon containing the paths to the icon, to the Pianoteq executable, and to the folder where I keep midi files (in Xubuntu, a right-click on the desktop background).

In my earlier posting I reported an average Performance index of nearly 21, but with the adjustments made here I get about 22.5. In the Rachmaninov prelude I see the polyphony jump as high as 83 with both sampling rates at 44.1kHz. Processor temperatures never exceed 58°C after half an hour. As before, it appears that Pianoteq requests real-time scheduling with priority 65 (as reported when started from a terminal), but the shell command ps shows that it is really running with priority 88, as requested in by chrt, with first-in, first-out scheduling policy -- which, I've read, can give a small improvement over the default round-robin, policy. In fact, it appears that -f (Fi-Fo) gains about half a Performance point over -r (round-robin).

You can look at the scheduling (and other) parameters using the shell command ps (see man ps):

$ ps -C Pianoteq -l       shows priority/nice? as -29

$ ps -C Pianoteq -o policy,rtprio,vsz
POL   RTPRIO    VSZ
FF        88  211628

where:
POL =  scheduling policy or class of the process;
   class FF = a First-In, First-Out real-time process
RTPRIO = realtime priority
VSZ = virtual memory size of the process in KiB (1024-byte units).


By the way, for various reasons, I wanted to replace the Acer laptop's mechanical hard drive, but I had serious problems when I changed to a Samsung EVO 850 SSD, and a recent report in the German magazine c't suggests this may happen with laptops from other manufacturers as well. I tried Kubuntu, KXstudio and finally Xubuntu on the SSD, and I found that with all of them the shutdown sequence hangs; when you forcibly turn the power off, the file system gets corrupted in the way in does when power fails. My solution: in /etc/default/grub change the appropriate line to GRUB_CMDLINE_LINUX="". This must be done from grub's "recovery mode", *before* you start the system the first time. Make sure to use only ext4 file systems. If at startup you get file system error messages, do thorough file system recovery before anything else. Apart from that, no problems that can't be solved by normal adjustments in the Settings applets.

Re: Is Intel N3510 ok for pianoteq?

Hi,

MinorChord wrote:

/usr/bin/cpufreq-set

are you using an older or special kernel? As far as I know the acpi-based cpufrequtils are deprecated in standard-kernels -->

# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver
intel_pstate

# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
performance

I'm getting a similar perf index of 22 on my older Acer E3-111 by just setting the scaling_governor to "performance" (and of course the limits.conf).

MinorChord wrote:

but I had serious problems when I changed to a Samsung EVO 850 SSD

Interesting, I have a smaller "Samsung SSD 840 EVO 120GB" inside the Acer - -  and no trouble. I just tested a hard power-off by pressing the power-button long enough. No filesystem troubles here (ext4).

cheers

Re: Is Intel N3510 ok for pianoteq?

Hi groovy,

Thanks for bringing me up to date about cpufrequtils and reporting your performance index without rtprio. I now see that rtprio is superfluous.

I'm running kernel 3.16.0-46-generic, and, yes, sys/.../scaling_drivers shows intel_pstate. Looking around, I see that cpupowerutils would be the thing to have, but it hasn't yet made its way into the Xubuntu repository.

I see, however, that in principle I could get along quite well without cpufreq-set. All I would need to do is

#echo performance > /sys/devices/system/cpu/cpu<n>/cpufreq/scaling_governor

for each of the four cores <n>.

However, I don't seem to be able to do this as sudo, only as root. Even if I could to do this from my launcher script, as a non-root user, I'd have to give 'echo' non-password status instead of giving this status to 'cpufreq-set', which is possibly a little dangerous.

When I start PT without rtprio (but with all processors under performance regime), I see that it gets "real-time scheduling with priority 65", Perf index is consistently 23, and I see instantaneous polyphony of 86 at 17:13 in the Rachmaninoff. Trying it with rtprio -f 88 today, I get exactly the same result. I don't know why I've gotten slightly lower values in the past; perhaps it had to do with older kernels.

The Samsung EVO 850 was a trial by fire. On forums I could find very little relevant help, only that eliminating "quiet splash" from GRUB_CMDLINE_LINUX_DEFAULT helped with a different problem. I had no idea where the file corruption was coming from, until one day I accidentally pulled the power connector from my laptop and saw similar errors from fsck when I restarted. The EVO 850 works fine in my office machine, and I even did a dd if=dev/zero to make sure everything was clean before installing yet another Linux on it. I was at least comforted by a remark in a recent review of new laptops in c't that reported a similar problem upgrading to SSD. I've read, too, that some SSDs take power outages much worse than mechanical drives because of the way they buffer writes. I normally have no battery in my laptop, so when the power goes, it goes instantly.

Re: Is Intel N3510 ok for pianoteq?

Hi MinorChord,

thanks for your detailed feedback!

MinorChord wrote:

However, I don't seem to be able to do this as sudo, only as root.

If you need to set the scaling_governor as a privileged user, you could try this:

$ echo performance | sudo tee /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor

(for each cpu-core 0-3)

MinorChord wrote:

I've read, too, that some SSDs take power outages much worse than mechanical drives because of the way they buffer writes. I normally have no battery in my laptop, so when the power goes, it goes instantly.

Then why not shutdown the laptop softly? ;-)
Or you can influence the "button events handling" in some desktop-environments, for example in KDE the option "When power button pressed" can be set to "Shutdown" or "Prompt log out dialog" (that's what I use).

Re: Is Intel N3510 ok for pianoteq?

MinorChord wrote:

I normally have no battery in my laptop, so when the power goes, it goes instantly.

For the longest time, I ran all my laptops w/o battery thinking that would save them from degradation.  Then after several years, I'd pull the batteries out of the cabinets and plugged them back in ... 30% max capacity.  Basically, laptop batteries degrade whether you have them plugged in or not.  I now run with both battery + UPS on all my computers.

You can increase the power-off resilience of hard drives/SSDs using the following options:

* mount ext3/ext4 with data=journal
* turn off write caching at the disk level using hdparm -W0 /dev/sdX
* turn off write caching at the OS level using /proc/sys/vm/dirty_ratio=0 and /proc/sys/vm/dirty_background_ratio=0 to force flush to happen with every write

These 3 options will make your disk slower but with a SSD, it may not make much of a difference.

Last edited by Mossy (04-09-2015 10:32)

Re: Is Intel N3510 ok for pianoteq?

Hi Mossy,

Thanks for the caching tips. I wonder if turning off caching in the SSD won't increase wear, since I would think it will produce more frequent writes to the eproms. Reinstalling everything after a crash is, of course, far worse.

My motivation for taking out the battery is mainly that I want it to have nearly 100% capacity on the relatively few plane trips I take. For frequent local travel, I take along a small external 12V rechargeable -- actually, the Acer E3 seems to work fine on as little as 6.1V, although below 9V the power LED starts flashing.

According to data I've seen on manufacturers' data sheets, the internal lithium batteries degrade fastest when they're warm and fully charged. At 40% charge and 4°C (ie. in the fridge) they could last many years, if what I read here is right:

http://www.batteryuniversity.com/learn/..._batteries

Re: Is Intel N3510 ok for pianoteq?

Turning off caching will increase wear by a bit but it's probably not by a lot.  It's not often you continually rewrite the same location for typical user usage.

Have you considered a cheap UPS?  The batteries do die every few years but the brick-style ones use standard car/motorcycle batteries and I've had them replaced several times already without hassle.