KVR Audio

Sampleconstruct · Post by **Sampleconstruct** » Sat Feb 07, 2015 1:43 am

Processing a granulated glass patch (HALion 5) with Kaleidoscope, using various images I made with ArtMatic and a waveform tuning file created in Metasynth.

Highly experimental stuff:

synzh · Post by **synzh** » Sat Feb 07, 2015 2:18 am

same problem on my Win7 64-bit using FL Studio 11.1.1 and latest update 1.0.1 of KS, lots of stuttering clicks/pops playing exactly same preset as on 1.0.0 that had very little clicks/pops

Sampleconstruct · Post by **Sampleconstruct** » Sat Feb 07, 2015 2:46 am

Mixolydian Sweep:

dayjob · Post by **dayjob** » Sat Feb 07, 2015 3:04 am

jc_vt wrote:That *appears* to have installed the update, but the download still refers to version 1.0.0 and the date created has not changed....

Galbanum wrote:try this:

http://www.galbanum.com/shop/product_in ... cts_id=116

follow the directions there.

Good?

yeah .. the zip file is labeled 1.0.1 but the installer is labeled 1.0 so i'm guessing they haven't renamed the installer to reflect the new update???

Andrew Souter · Post by **Andrew Souter** » Sat Feb 07, 2015 3:46 am

jeremydb wrote:The string resonators are much better, but I still get clicks in many presets (for instance, the Auto Strings presets at the top of the Factory Bank) when using all 512 resonators (regardless of the resonator model), despite the fact that I don't appear to be maxing out my CPU. Backing down to 256 solves the problem, but I can't really explain the behavior. I'm on a Retina MBP running 10.10.2, 2.3 GHz Core i7, 16GB RAM, should really be adequate...

What is your host? What is your host/hardware buffer size? What is your audio card?

Upload a screen shot of the INFO page of the GUI please.

I can run 10 or more max setting, instances for Strings and Springs on my Mac Pro even at a buffer of size of 256. At 2048 I can run 14 instances in Cubase 8 and Logic 10. I would guess you should be able run at least 3 or 4 max setting instances live on your laptop.

jeremydb wrote: Actually, I just observed this, while testing in Live (I can test in Logic if necessary). Note that I have low-resolution mode enabled to avoid the Retina problems.

1. Instantiate a Kaleidoscope, run the transport, enable noise input, etc.
2. Open preset Auto Strings 3
3. "Rapid clicking", as if oversampling were enabled.
4. Go to "Info", back the resonators back down to 256. Then return to 512.

Now it works ok, although if I go to a different preset and back to "Auto Strings 3", I have to do it again. The programmer in me guesses "uninitialized variable", but you're the boss! I was seeing this in 1.0.0, as well.

Well, if you switch the info Page pref to 256 and then immediately back to 512, the number of active resonators will NOT automatically return to 512. You would have to go to the MAIN page and move the "Max Enabled Lines" slider up to max. (the red slider if you are using the Cyan color scheme.) This preference is basically a limit as to how far the slider can go up. If you change the pref that requires the slider to come down, it will. But it does not go back up automatically. You have to change it, or change presets or whatever. So in this case, you might actually simply be comparing 256 vs 512 active lines. 256 will use less CPU obviously. Roughly half.

But as I said you should really be able to run several instances of 512 lines, dual mod, dual res modes. Please provide the info requested above. Thanks.

Andrew Souter · Post by **Andrew Souter** » Sat Feb 07, 2015 3:50 am

jc_vt wrote:That *appears* to have installed the update, but the download still refers to version 1.0.0 and the date created has not changed....

The installer file itself on OSX was labeled 1.0.0, yes -- even though it was 1.0.1. My mistake. Corrected -- well I have actually just removed the version number from the installer itself to avoid forgetting to change this in the future -- so now only the download zip, and the unzipped folder show the version number.

The installer itself should have been correct. 1.0.1 is internal build 379. You should see b379 on the INFO page of the GUI. Do you?

Andrew Souter · Post by **Andrew Souter** » Sat Feb 07, 2015 4:02 am

xx JPRacer xx wrote:Sorry if it was asked before but what program did you use to generate those images?

Almost everything in the Architecture Volume One content library (which is also the source of most of the factory content) was created between 2000 and 2005 -- maybe even some in 1999). Almost all of the images were done with a combination of Photoshop and Metasynth, and almost all of them were done "manually" in some fashion. Artmatic may have been used a little for a few of the "fractal-looking" images. But it was actually NOT used that much, and may not have even existed at that point -- I don't quite remember.

The "New Additions" folder is stuff that I have done more recently during KS development. I have created proprietary methods to make these kind of images. I do them in a similar way to the way I made Galbanum Architecture Waveforms. I do not use any 3rd party tools. I create them all myself.

I am currently developing Architecture Volume Two which will offer even more resources of this nature targeted specifically for Kaleidoscope.

Andrew Souter · Post by **Andrew Souter** » Sat Feb 07, 2015 4:27 am

Anyone who says that 1.0.1 seems LESS efficient than 1.0.0:

1) Check the info page of the GUI. Does it show b379?

2) What host are you using?

3) What OS?

4) What computer hardware?

5) What host/hardware buffer size are you using? (set to 1024, and relaunch your host. any better?)

6) What KS buffer size are you using? (set to 1024, and relaunch your host. any better?)

7) What KS Mulithreading pref are you using (set to 100%. and relaunch your host. any better?)

1.0.1 should absolutely be significantly more efficient than 1.0.0, and DRASTICALLY so for Strings. If you find otherwise, please answer the above so that we may explore more. We do NOT see this on our machines. 1.0.1 is much better than 1.0.0 in our internal use and tests.

synzh · Post by **synzh** » Sat Feb 07, 2015 4:33 am

for me the GUI shows "Version 1.0.1 [A379] AVX2"
host=FL Studio 11.1.1 currently 32-bit
OS=Win 7 SP1 64-bit
Intel i5 3.4GHz quad-core 8Gb RAM, soundcard=Soundblaster Audigy FX
host buffer size=640 samples
KS buffer size=1024
KS multithreading=100%

jackmazzotti · Post by **jackmazzotti** » Sat Feb 07, 2015 4:33 am

Galbanum wrote:
jc_vt wrote:That *appears* to have installed the update, but the download still refers to version 1.0.0 and the date created has not changed....
The installer file itself on OSX was labeled 1.0.0, yes -- even though it was 1.0.1. My mistake. Corrected -- well I have actually just removed the version number from the installer itself to avoid forgetting to change this in the future -- so now only the download zip, and the unzipped folder show the version number.

The installer itself should have been correct. 1.0.1 is internal build 379. You should see b379 on the INFO page of the GUI. Do you?

mine says a379

synzh · Post by **synzh** » Sat Feb 07, 2015 4:35 am

aha - setting host buffer size (ASIO4ALL) from 640 samples to 1024 samples DOES help quite a bit.
I presume if I use the 64-bit FL Studio (with 64-bit KS) I may get better results as well -

[edit] actually 64-bit makes no difference which in retrospect makes sense; but looks like keeping the buffer size larger than normal for the host (at 1024) while running KS is my resolution

Andrew Souter · Post by **Andrew Souter** » Sat Feb 07, 2015 5:12 am

synzh wrote:aha - setting host buffer size (ASIO4ALL) from 640 samples to 1024 samples DOES help quite a bit.
I presume if I use the 64-bit FL Studio (with 64-bit KS) I may get better results as well -

All of this "variable buffer size" stuff is a new topic for us. So we are gathering intel based on people's experiences now.

As mentioned earlier in the thread, KS generally likes large buffers. You probably want to use at least 256 samples for KS. Using the max of 2048 vs 256 will likely gain about another 40-50% in efficiently. i.e. if you can run two max-setting instances live at 256 samples, you might be able to run three at 2048. But 256 should be completely workable, and is prob the best compromise if you are using live instruments etc. since you will probably want low latency for performance reasons. This more latency you can tolerate, the bigger the buffer can be, and the more efficient KS can be.

Also as mentioned, some audio cards (fire-wire in particular, but maybe there are others) use weird buffer sizes. 640 samples for example, is strange. Usually the buffer size is a power of 2 (32, 64, 128, 256, 512, 1024, 2048...). If you have the option, try to use one of these settings.

We find that in some cases performance is best when the host buffer and the KS buffer are the same size. The first thing to try is to set BOTH to 1024 samples. Performance should be very good in this case. Is it?

Once that is verified, then you can try other settings to see how low you can realistically go.

We continue to explore this topic and look for ways to optimize this and handle all host/hardware cases in the most efficient way possible. We appreciate your feedback on these topics.

Andrew Souter · Post by **Andrew Souter** » Sat Feb 07, 2015 5:14 am

synzh wrote:
[edit] actually 64-bit makes no difference which in retrospect makes sense; but looks like keeping the buffer size larger than normal for the host (at 1024) while running KS is my resolution

Using a 64-bit host, does make some minor difference. Performance is slightly better in 64-bit hosts compared to 32-bit hosts.

synzh · Post by **synzh** » Sat Feb 07, 2015 6:02 am

That's good to know; I couldn't discern much difference between the 32-bit and 64-bit versions performance wise but as time goes it's getting easier to stay with 64-bit stuff anyway.
Also couldn't tell much difference between setting KS to 1024 vs. 2048, I had left it at 2048, but now that you mentioned keeping it same as host buffer size can help I'll leave it at 1024 for now.
My original setting of 640 samples was actually set manually in the ASIO4ALL generic driver using the slider there, not from the soundcard itself.

Andrew Souter · Post by **Andrew Souter** » Sat Feb 07, 2015 6:23 am

There are diminishing returns in performance as buffer size gets bigger. Difference between 64 and 256 (which is 4x as big) is MUCH more pronounced, than between say 512 and 2048 (which is also 4x as big).

Basically if you use 256 or higher, and match both the KS and the host buffer things should be pretty good.... if you can go higher you will get some additional benefit. but 1024 vs 2048 for example won't be a huge difference.

2CAudio Kaleidoscope | It's A Trip | Latest Update 1.1