1.27 beta slower than 1.26 ?

Some users said, that the 1.27 beta 2 release is much slower on their system so I started to optimize the performance and did some tests.

On my system (Win2k, 2GHz Athlon, 512 MB RAM) the official 1.27 beta build is about 5 percent faster than the official 1.26 release build.
Maybe the problem is CPU dependent so that the new version is only slower on some CPU`s. I therefore need some feedback from you to see if this is true and fix the problem as soon as we know more.

To test the performance, do the following: Choose a song that uses many samples and VSTis and needs much performance. Play only a pattern from this song and watch Renoises CPU meter (not the taskmanagers) as we only want to test the audio performance. Write the two values down. Preferible the average, and the upper and lower peaks.
If you write your results into this topic, make sure that you you also write what CPU, how much RAM and what OS you are used for the test and if you`ve used ASIO or DirectSound.

When running the test, also make sure that you use the same “processing buffer size” (configs/audio) as this affects the performance !!!

my result is :

System : Win2k, 2GHz AMD Athlon, 512 MB RAM
Processing buffer size : 512 samples, using DirectSound
1.27 beta 2 : 39 percent (average)
1.262 : 43 percent (average)

You already know my vallues but I write them down anyway

System : WinXP, AMD Athlon 1800 XP+, 1024 MB RAM
Processing buffer size : 256 samples, using ASIO

Only internal samples:
1.27 beta 2 : 70 percent (average)
1.262 : 55 percent (average)

VST’s and samples:
1.27 beta 2 : 50 percent (average)
1.262 : 50 percent (average)

PC: P3 733 with 592MB RAM,
Audio: Terratec EWX 24/96 using ASIO
Video: ATI Rage PRO 2000 16MB RAM
Renoise: 1024 buffer size using NI Kontakt

ReNoise 1.261 (btw, what’s that 1.262 you’re talking about!? :unsure: )
pattern 17: 40%
pattern 1a: 50%
pattern 20: 53%

ReNoise 1.27b2
pattern 17: 50%
pattern 1a: 73%
pattern 20: 76%

1a and 20 produce crackles on 1.27b2, while everything goes smooth on 1.261
crackles are a bit reduced when not using the pattern follow option.

System:
P4 2,52Ghz - FSB 157Mhz - i845E - 512MB PC400 - SBlive /w KXAsio 5, 10, 00, build3528

Renoise Audio Settings:
ASIO, 21ms Outputlatency, Samplerate @48Khz, ProcessingBufferSize @128 samples


Song with VST/VSTi/Samples

1.27 beta2 : peaking @51%
1.262 : peaking @51%

Song with Samples / Internal FX
(using my bb2 entry “Martian Spaceboots” looping Songposition 12 / Pattern 0B)

1.27 beta2 : peaking @31% (!!) :o
1.262 : peaking @17%


as a (very important) side note, you can also very well observe the difference in the way the integrated filter fx is handled in 1.27 beta1/2 when compared to 1.262, which i already pointed out earlier before.
in order to reproduce this malfunction, play “Martian Spaceboots” and compare the sound of 1.27 to 1.262. focus your attention to lowpass filter activities and you should be able to hear the very well audible difference at for example [songpos12/pattern0B] on track00 between v1.26 and v1.27 sound.
so in version 1.27 beta2, the imported filter of songs written with 1.262 is treated in an improper way - there’s no doubt about it imo.
please somebody, especially taktik, let me know, if he’s able to reproduce this.

System : WinXP, AMD Athlon 1200MHz, 384 MB RAM
Processing buffer size : 256 samples, using ASIO

Song 1:
1.27 beta 2: 49%
1.261: 49%

Song 2:
1.27 beta 2: 64%
1.261: 63%

Song 3:
1.27 beta 2: 72%
1.261: 73%

if i can still trust my eyes, twilek has got an athlon xp system and also got the higher CPU usage with tunes that use samples/internal fx only.
same as here. i don’t guess it’s CPU brand dependant.
i’m going to test this on my 2nd PC now, which is an Athlon TB 1,4Ghz / 640MB. i’d be very suprised if things would turn out to be different on that rig.

did the test on my secondary system which looks in detail as followed:
Athlon 1,4Ghz - 640MB PC133 - FSB 133 - KT133A

Renoise Audio Settings:
Directsound(no ASIO hardware available on that sys), 50ms Outputlatency, Samplerate @48Khz, ProcessingBufferSize @512 samples

Song with Samples / Internal FX
(using my bb2 entry “Martian Spaceboots” looping Songposition 12 / Pattern 0B)

1.27 beta2 : peaking @38%
1.262 : peaking @23%

I experienced Keith’s problem in my very own song(s): Renoise 1.27 beta handles moogLP instrument envelopes (and others ?) different compared to Renoise 1.261 - I played a 1.26 song in 1.27, and the cutoff instrument envelope of the moogLP didn’t let pass as many high frequencies as in the older version.

exactly what i meant. the LP filter cuts off “more” than it used to in 1.262.
btw: hi kenny, long time no see :D

in general, 1.27 sounds more bright, with more higher frequencies (when not filtered).

Could be that taktik has found that sound quality bug someone has reported before?

I finally also found a song now wich is slower in the 1.27 beta release so I hopefully can find and fix the bottleneck now.

About the different sounding Filter : I also found the problem here and will send you keith303 a build for testing the fix, ok ?

in general, 1.27 sounds more bright, with more higher frequencies (when
not filtered).

Could be that taktik has found that sound quality bug someone has
reported before?

I found no bug there, but we are using again the old compiler that we also used for 1.25 again. It might be possible that it produced faster code because of more unprecise floatingpoint calculations wich resulted in a this bad quality. Just a guess, this needs some more testing :)