KVR Audio

eidenk · Post by **eidenk** » Sun Aug 03, 2014 6:57 pm

AXP wrote:Eidenk: The convolution engine is the same for both the spring reverb and the speakersim, and it's the same as the one used in FM25. So we can narrow it down to the convolution. That's strange, as I've specifically made sure to even out the CPU load so it doesn't spike. I need more details: your CPU, OS, the host you are using.

Well, I am using an old OS, (Windows ME) and I think that's the problem so no worries if you can't/don't want to do anything about it. CPU is single core AMD Opteron 154 (equivalent to Athlon 64 FX-57) and host doesn't matter as it's the same with any.

I've had a similar problem with Mystran's convoluton reverb IRDust some time ago and he managed to fix it fairly quickly, the issue probably being "scheduling the worker before the audio thread". He said it could eventually affect Windows XP too. See discussion (about 10 posts) starting here: http://www.kvraudio.com/forum/viewtopic ... 0#p5644587 in case you're interested looking into it.

Thanks.

eidenk · Post by **eidenk** » Sun Aug 03, 2014 6:59 pm

Mmmmh, sorry for double post, I quoted instead of editing.

AXP · Post by **AXP** » Sun Aug 03, 2014 7:00 pm

Jonas78 wrote:Am currently testing and trying out all entries and come across a cpu usage issue in Renoise when enabling your plugin. When instantiating the cpu rises gradually 52% on top of what it was before! Is this normal? Am running Renoise 3.0 on a vista Core2 duo laptop, 4 gigs ram.

Well, tried the demo, but it's limited to directsound I/O only (no ASIO), so I'm not sure it's correct way to analyze the performance.

Mine shows about 15% load with everything enabled and limited to 1 core only.
For now I can only suggest lowering the oversampling setting.

Offtopic: I didn't know the trackers were still alive and evolved so much!

AXP · Post by **AXP** » Sun Aug 03, 2014 7:06 pm

eidenk wrote: Well, I am using an old OS, (Windows ME) and I think that's the problem so no worries if you can't/don't want to do anything about it. CPU is single core AMD Opteron 154 (equivalent to Athlon 64 FX-57) and host doesn't matter as it's the same with any.

That explains a lot

I've put a Windows XP requirement for a reason

eidenk wrote:I've had a similar problem with Mystran's convoluton reverb IRDust some time ago and he managed to fix it fairly quickly, the issue probably being "scheduling the worker before the audio thread". He said it could eventually affect Windows XP too. See discussion (about 10 posts) starting here: http://www.kvraudio.com/forum/viewtopic ... 0#p5644587 in case you're interested looking into it. Thanks.

It makes sense. Thread scheduling is indeed different between ME and XP. I'm surprised it works at all. Anyway, I'll see how fast I can fix this, but you can be sure I will not ignore it.
Thanks a lot for the link, I'm surely going to read it.

Dandruff · Post by **Dandruff** » Sun Aug 03, 2014 7:31 pm

Dandruff wrote:Too bad, rendering gives total garbage output here (REAPER 4.71 x64, Win8.1).

Flextron on Master track -> Master track rendering (48kHz, offline rendering): garbage output (chopped of sound)
Flextron on vsti track -> Master track rendering (48kHz, offline rendering): overload (> +600dB output)

Something definitely not right with this.

Mmh, looks like it's not happening all the time. Tried to make a small test-project but couldn't make it happen with it. Very strange ...

AXP · Post by **AXP** » Mon Aug 04, 2014 10:29 pm

eidenk wrote:Mmmmh, sorry for double post, I quoted instead of editing.

I've PM'd you with an experimental version. Please have a look when you have time.

AXP · Post by **AXP** » Mon Aug 04, 2014 10:52 pm

Meanwhile I also compared my convolution implementation against the Ignite Amps' NadIR.
For benchmarking I've used Reaper's offline rendering speed meter.
Here's the results (everything is for stereo signal):

performance_nadir.png

While there's obviously some room for improvement, there's nothing extraordinary as you can see.
Also for reference, Emissary preamp (without the poweramp section and at 4x OS) takes 6.97% CPU in the same scenario.

Note though that the Ambience algorithm takes a lot of CPU. I'll look at this later.

So I think we can narrow down the performance issues to specific cases when multi-threaded convolution algorithm behaves erraticaly. This is something I'm currently working at and hopefully will make an update soon.

eidenk · Post by **eidenk** » Tue Aug 05, 2014 12:01 am

AXP wrote:
eidenk wrote:Mmmmh, sorry for double post, I quoted instead of editing.
I've PM'd you with an experimental version. Please have a look when you have time.

All good now, you've earned your 4 points!

Btw, if you could implement the fix in FM25 too sometime, that'd be great.

Cheers

AXP · Post by **AXP** » Tue Aug 05, 2014 1:29 am

eidenk wrote:All good now, you've earned your 4 points!

Btw, if you could implement the fix in FM25 too sometime, that'd be great.

Cheers

That's awesome. I'll give it a fresh look tomorrow and then release it publicly.
FM25 will be updated too, of course. Just maybe a bit later when I get more time.

It's funny that I have implemented dynamic thread prioritization from the beginning, but then dropped it after not being able to achieve any performance/stability gain on the systems within my reach.

It's even more funny that me and Mystran seem to have stepped on the same rake (it's a Russian idiom for making a stupid mistake more than once)

AXP · Post by **AXP** » Tue Aug 05, 2014 10:06 pm

The aforementioned update has been released.
It's available on my site, on KVR (simply re-download the plugin), and via the "update" button in the pugin's GUI.

jPod · Post by **jPod** » Wed Aug 06, 2014 7:06 pm

2 Little bugs encountered:

1 - Loud pops when first loading or switching between oversampling rates even with audio input disabled by the host.

2 - When closing and then re-opening the GUI Bias/EQ/Tube section reverts to Off.

OS - Windows 7 SP1 32 bit.
Hosts - Reaper4.71 and Cantabile Lite 2.0.

Also, I don't know if it's really a bug, but is the Clean channel supposed to distort so much? Not what I thought of as solid state amp behaviour, but I haven't heard the hardware it's modeled on.

All around great work though.

AXP · Post by **AXP** » Wed Aug 06, 2014 10:08 pm

Hi,
first of all thanks for reporting this!

jPod wrote:1 - Loud pops when first loading or switching between oversampling rates even with audio input disabled by the host.

I'm aware of the clicks and pops. I just didn't have time to work around them. This is something I will do eventually. The most prominent happen when switching to clean channel.

jPod wrote: 2 - When closing and then re-opening the GUI Bias/EQ/Tube section reverts to Off.

Since it's a visualization only and doesn't affect processing at all, I decided not to make it part of the exported parameters. Otherwise it would be stored in presets and jump around when switching them.
On a second thought I could just save it temporarily within a session. I guess that's more than is allowed before DC ends, so I will do it afterwards.

jPod wrote: Also, I don't know if it's really a bug, but is the Clean channel supposed to distort so much? Not what I thought of as solid state amp behaviour, but I haven't heard the hardware it's modeled on.

Well, first of all, check that input level knob is not at the maximum setting. 0dB peak corresponds to 1 Volt. I suggest to keep it at the middle setting (0dB boost). Also check the Tube stage setting - try switching it off.
Also keep in mind that the real amp doesn't have a master volume, so the clean channel volume is usually set very low. The topology of the channel is such that the stage that clips is after both the tonestack and the level control, so their setting would affect the clipping behavior.

jPod wrote:All around great work though.

Thanks

Dominus · Post by **Dominus** » Thu Aug 07, 2014 5:05 pm

Odd question - I get an error trying to open the x64 version in Reaper. x86 version works fine. When scanning it never shows up. If I uncheck "Get VST names" on scanning it will come through but then Reaper disables it as soon as I load it. It looks like FM25 does the same thing, and I downloaded that before I upgraded my computer.

I've reinstalled and reset Reaper's configuration, same thing. Other people are having similar issues lately, one other person said that it occurred after he upgraded his computer. (I did the same recently.) SoftDrive GV loads and one of the other DC2014 plugins works just fine, so it doesn't seem to be a folder permissions/Windows issue. I'm wondering if there's a newer compiler for VSTs that you might have used on Flextron and FM25 that has an issue with Reaper or the new Intel Z97 chipset / i7-4790k CPU.

Anyone else experiencing anything similar?

Ivan_C · Post by **Ivan_C** » Thu Aug 07, 2014 5:07 pm

Have you tried to remove the x86 version when you try to use the x64 one ?

Dominus · Post by **Dominus** » Thu Aug 07, 2014 5:29 pm

Wolfen666 wrote:Have you tried to remove the x86 version when you try to use the x64 one ?

I only used the x86 version after the x64 version failed to load.

I even did a test with only 4 plugins in a new VST64 folder. Flextron and Emissary (both which failed to load), NadIR which loaded fine, and Acon Digital's Multiply, which also loaded fine. Emissary gave an error message, Flextron didn't do anything. When I rescanned with the option for VST Names unchecked both plugins would load, Emissary would crash Reaper and Flextron would get disabled immediately.

I'm pretty sure FM25 worked before I upgraded my computer, so it really seems like a compiler/CPU/Chipset issue. I haven't gotten a chance to try another host yet, but I'll try to do that tonight.

AXP Flextron DC14 support [1.0.03 / 11.08.2014]