I've tired to find an answer on the Reaper forum for a problem I'm having with plugin loads in Reaper but so far no luck. I wonder if anyone can explain to me what is happening. My specs are: Reaper 4 x64, Win 7 x64, AMD X6 1090T, 16GB RAM.

I noticed that when I ran one instance of SynthMaster 2.5 in Reaper 4 it caused the audio to break up when the synth audio quality was set to 'Best' using 12 note polyphony. When I set the polyphony to '8' it played fine.

I checked Reaper's performance meters and at 12 note poly FX CPU was 7% but RT CPU was 133%. Win 7 Resource Monitor reported CPU use at 7%.

I searched for an answer and came across this old SOS article:

http://www.soundonsound.com/sos/jan08/articles/pcmusician_01 08.htm

It states:

"The vast majority of stand-alone soft synths also seem to mostly use a single core, but as soon as you load the VSTi or DXi version into a host VSTi or DXi application, this host should distribute the various plug-ins and soft synths across the available cores to make best use of resources."

I then made a test and loaded 6 instances of SynthMaster 2.5 on separate tracks in Reaper. As long as they were set to 8 note polyphony they played fine and used 94% of all six CPU cores. As soon as I made one instance 12 note poly the audio crackled and everything slowed down.

My question is why can Reaper handle multiple tracks of SynthMaster at 8 note poly (94% CPU) but not a single track of SynthMaster at 12 note poly (even though CPU is only 7%)?

Why you don't ask it at the KV331 SynthMaster forum?

http://www.kvraudio.com/forum/viewforum.php?f=98

I have posted there but this isn't a SynthMaster problem because the same thing happens with any plugin that overloads a track in Reaper. The drop-outs are caused by the way Reaper handles plugins. I'm just trying to find out if anyone knows why.

For reference, this is the Reaper forum thread:

http://forum.cockos.com/showthread.php?t=101660

AFAIK no host can distribute one plugin's load across cores. As someone says in that thread the plugin is like a black box to the host. Its load is atomic (can't be sliced up).

On the other hand if you use multiple instances of a plugin spread across tracks then the host WILL be able to distribute the load.

There was a thread that I think might have helped on the reaper forums, but I personally can't find it. Get a load of this, the search engine over there yields ZERO results for "cpu"

Anyways, maybe a search for reaper/cpu on google can help, I tried to find it (it was in the last few days over there)

EDIT: I found the preference that may help:

Try audio/buffering/allow live FX multiprocessing on (select cores)

Not sure if it will help, but thought I'd mention it.

Sounds to me like that option is only useful when recording audio input, probably from multiple sources to multiple tracks which would then be split across cores. The hover help warns that it isn't ideal for low latency performance. I doubt it would boost performance of a single synth plugin on a single track.

What I'd like to know is how Synthmaster performs at 12 note poly and "best" setting in a different host on the same system. Does it max out a single core in another host?

All I know, is on my machine, things are distributed really well. I hate to load that demo just to check it, there has to be others that use it and use reaper. Plus the dev is around somewhere right????

Ok, I dl'ed the demo and found a pad (with movement and have no problems with cores here. It even ran ok in studio one!

However, it's quite hungry. Rarely does anything reach the 7% mark here in reaper so with much polyphony it can suck the juice.

btw, I have an i7 860 (first gen i7)

I have a dinosaur with reaper in linux, and it uses two cores very close to
50-50, regardless what plugins are loaded, undulating at times between
45-55 and 55-45, for no apparent sonic reason.

There is a zynaddsubfx preset that will send the cpu beyond the moon in seconds,
sometimes returning after a coffee break, sometimes not
Very pleased with reapers cpu management, even if it's a fluke

I suppose when i9 comes out, I can mow a few lawns to get an ancient i7

these days the super cheap $400 pc you can buy from virtually any shop isn't really that bad. usually includes a windows license and everything so all you need is your midi and audio and that's it.

a far less powerful machine would have been $2000 five years ago. things have changed fast since they reached the peak of speed they could manage by multiplying clock frequency.

i'm definitely not playing that game anymore. i'm still on a core2duo from six or so years ago that was 2k. i cna get the same power now for like $300. sad.

since i got this machine i've never once ran out of power for any audio project. i can max things out by using 16x unisons and 200 voices and stuff, but that's really super unrealistic. for the 40 or so voices most projects usually use up and the ten or twenty effect plugins i rarely go beyond 30%.

Chances are unless this computer goes toenails up, I'm done (people might remember prior to 2010 when I got this I was ALWAYS hungry for more ) i7 with reap is pretty powerful stuff. Other hosts? We won't go there

Thanks for trying it out. I'm using a lead from one of the Rob Lee banks 'LED Plucked Trance RL' set to 'Best' and using 12 note poly. It maxes out at this setting but if I reduce it to 8 note poly it plays fine and I can load six of them before my AMD X6 1090T buckles.

If it is the case that a plugin has to be coded for multi-cores to distribute the load in a host then why isn't this coded as standard in this day and age? And how many other synths and FX aren't multi-core? This single core barrier poses a problem with synths that use a lot of juice. It seems such a stupid limitation when we are now using multi-core CPU's.

it isn't a "standard" because the os/language doesn't provide for it, it's extremely volatile and difficult to implement.

new additions to languages are starting to include this stuff and you'll be able to see it more often by about 2015-2020.

it's just without the language/compiler being able to help you diagnose problems and develop the feature in the first place it's a very dangerous thing to add. especially when you have to consider the latency/overhead of thread/core switching and all the shared/redundant data you have to move around.

if you were a programmer you'd understand. it's difficult to explain. short version: it's not as easy as you think.

most hosts don't even have good implementations!

What you usually want in sound processing is short latency and as little jitter as possible. General-purpose CPUs built into PCs are already not-so-good at this. When you go multi-core, you basically have to synchronize the tasks you have split up and get your information back together to provide a result. On a multi-core CPU this is likely to add to the jitter and to the latency as not all cores are available at all times. Synchronisation btw is mostly a fancy term for "waiting for another", and this is precisely what you do not want when you process sound, as the human ear is really good at spotting delays.

There are special-purpose architectures that make latency and performance guarantees. When you fork over the royal amount of dime they ask for the hardware that they sell for pro-level studios you basically get hardware that does it all in parallel, with software to match, and with known performance characteristics, etc. This is not nearly possible on PC hardware, so what we settle for is a sort of best effort approach - keeping it to one core per plugin makes it a lot easier to program, a lot less error-prone and so on.

Don't forget music-making is a niche market in many respects. The vendors out there are likely to sell in much smaller numbers than let's say game vendors or even digital image processing vendors. So the development time for the amateur-to-mid-level market where people are rather cost-conscious must be rather limited.

Thanks for your replies. I can now understand why it's so difficult to go multi-core with plugins. Having said that, I think DIVA is or is about to become multi-core so perhaps it's already possible with the right expertise.