LSSP demo performance issue

[astro-consulting]

Hi,

I'm very interested in the full LSSP bundle and am currently running the demo on a 2018 MacBook Pro ( 2.9GHz 6-core intel Core i9 with 32GB RAM, MacOS Monterey ). I've very recently upgraded the OS from Mojave to Monterey and all 'seems' to be well across applications so far ( Reason 12, Ableton Live Lite, Maschine 2, Reaper, VM, VCV Rack 2 Pro, Element, standalone VST's ). I wouldn't generally regard the laptop as slow or lacking in resources.

I am able to run tutorials 1, 2b,3 and 4 as well as the NinthTriangle demo without issue. When I load and run tutorial 5b, the London Crime Theme or Arwen demos, I am getting graininess and distortion. When I stop the performance, pull down the settings Audio/MIDI dialog and click on the Test button next to the output device I get the same level of distortion on the test tone with just the demo loaded but not running. I am running 44.1k with a 512 buffer. Increasing the buffer size doesn't seem to have made a difference currently.

It's not stopping my evaluation and, apart from resolving this issue, the full LSSP bundle is exactly what I am looking for.

I'm not seeing the same issue in other VM patches and have been able to load multiple VSTs (similar to tutorial 5) previously. I haven't tried loading VST's in since the Monterey upgrade but will be doing so in due course.

I have stopped most or almost all background processes in case that is the issue but it is making no difference.

Any clues, thoughts, pointers, ... solutions appreciated

Many thanks
Regards
Andy

[ColinP]

Hi Andy,

Some LSSP patches can be testing on CPU especially Tutorial 5 but everything works totally fine on my Ryzen 7 4800H which is not a a top of the range processor, so hopefully we should be able to get you up and running without glitches.

It's maybe worth looking at the performance tips on the Adroit website, if you haven't already done so...

https://www.adroitsynthesis.com/performance-tips/

The first thing I would recommend is running at 48 kHz rather than 44.1 kHz because 48 kHz is VM's native sample rate. Running at anything other than 48k will cause significant resampling overheads and reduced sound quality. I run at 48k with a 480 sample buffer (10 msec latency).

I think there were several options on the 2018 MacBook Pro so f you could let me know the full name of your machine's CPU then I can look up its specs and give you a more informed opinion. I've found that GFLOPS is a better indicator of what particular machines are actually capable of than generic processor family names and clock rates.

[basa333]

Hi,
I have same problems too (win 10, i7, 16 gb ram). I will try 48 khz and give info.

Thank for tip tom

[basa333]

Yeah!

thanks for this tip, it seems to be ok! I will try other modules, where I have these problems.
now I'm working with the whole LSSSP XL (but it's not easy ) and granular synth (I love both, great systems!!)

thanks and greets tom

[ColinP]

Hi Tom,

That's great news.

Hopefully now you're running at 48 kHz all your VM patches will be more efficient and sound a bit better too.

Yeah, there is quite a steep learning curve with LSSP but once people get their head around timing being voltage rather than pulse based it all starts to make complete sense.

I spent a lot of time on the website documentation but it takes some patience to plough through it. Hopefully the tutorials are just about strong enough to get people to the point where V/Bar, time-splitting and S-Poly no longer seem like alien concepts.

At the moment I'm working on putting together some videos on the Granular Synth but then plan to do some on LSSP as the former seems to be sparking a lot of interest in the latter.

[basa333]

Hi,

the london crime theme is working, but LSSP tutorial 3 is crackling again (CPU is running at about 40%). LSSP seems to be consuming a lot of system resources. I'll try to simplify the patch somehow, but I don't know...

greets tom

[ColinP]

Hi Tom,

Did you mean Tutorial 3 or Tutorial 5? Tutorial 3 runs at about 9% on my machine as it's essentially just three mono synths.

On the other hand Tutorial 5 uses four instances of Voltage modular running simultaneously and a total of 123 modules so really is a bit of a beast. As well as demonstrating hierachical sequencing it's meant to be a bit of a stress test so that people can fine tune the performance of their setups but it runs without glitching at about 27% total CPU on my Ryzen 7 4800H (around 22% for VM, the rest on other tasks I have running in the background). The numbers vary by a few percent depending on which way the wind is blowing.

As far as I can tell (given the lack of metering in VM) the vast bulk of the CPU hit in patches like Tutorial 5 is from the audio modules. Most of LSSP is extremely optimized code doing the absolute minimum of work, it still has to run at 48k but apart from a few special cases it's not using much CPU.

It's a problem with large patches that even when modules like oscillators and filters aren't contributing anything to the sound they still use up CPU. It's something that GS addresses as its 500 voices "know" when they are doing anything useful and shut down the rest of the time to save CPU. But in things like Tutorial 5 and Arwen all the VCOs, VCFs etc are working flat out all the time. So for instance when you are hearing the verses play in Tutorial 5 the chorus modules are still using CPU even though you can't hear anything from them.

The processor hit in large scale patches could be solved by being able to switch off processing in a plug-in host with a special gate signal and then we'd all be able to build enormous patches that run on modest machines. But CA have shown no interest in this suggestion. I am pondering whether to implement my own plug-in host dedicated to just running embedded instances of VM as this facility to shut-down sub-patches is a pretty essential component for taking VM to the next level.

Returning to your particular problem could you look up the full name of your processor and let me know then I can give you more informed information. Many people don't seem to realise that things like "i7" are pretty meaningless marketing terms. There are about 40 or 50 different i7 CPUs covering a wide range of performance.

Best wishes,
Colin

[basa333]

Hi Colin,

thank you for your time and your extensive response.
I'm afraid my approx 9 year old laptop (acer aspire 3-772 G/i7-4702MQ 2,2-3,2 GHz /16 GB RAM/Steinberg UR22) just doesn't stand a chance.
Tutorial 5 runs at about 77% performance, the result is almost unlistenable.
I'm thinking about a new pc, but it will be a while yet.
It's a pity that I can't somehow reduce the sampling rate even at the cost of some reduction in sound quality, for example to 22050 kHz as with VCV.


greets and thank you tom

[ColinP]

Hi Tom,

Thanks.

Choice of sample rate depends on the audio driver. I'm wondering if you were using 44.1 kHz if you have DirectSound selected as the Audio Device Type in VM's settings. If so try changing it to Windows Audio. DirectSound is unusable on my machine. I use ASIO with my Focusrite interface.

But yeah the i7 4702MQ is unfortunately only capable of 9 GFLOPS.

If it's any consolation, in 1984 the fastest computer in the world was the Cray X-MP supercomputer. It did 0.8 GFLOPS and cost $15,000,000 (disk drives not included) so your computer is still pretty powerful.

To run Tutorial 5 I think you need something like 20 GFLOPS and even then everything needs to be perfectly tuned. But Tutorial 5 really is an unusual patch and something of a brutal stress test. You should still be able to do plenty of interesting things with your machine.

I'll say a few things for a general audience about CPUs as this subject comes up frequently in discussions.

I use GFLOPS as a benchmark because the main thing a CPU is doing while running VM is floating point multiplication, addition and subtraction. So although one shouldn't read too much into the absolute GFLOPS rating it's a useful base indicator of what your machine will be able to do.

Here's a range of CPUs and their approximate GFLOPS ratings...

i7 4702MQ = 9
i5 3470 = 11
i7 4980HQ = 12
i7 4790K = 15
i9 9980HK = 33
R7 4800H = 38
i9 9900K = 41
i9 12900K = 106

As you can see there's a big difference in computational power and unqualified terms like i7 or i9 are not very useful.

On finance I use the dollar a day three or four year rule.

So I think it's reasonable to spend $1 each day on computer hardware and I think machine performance improves at a rate that it makes sense to change computer every three or four years.

By this logic I think it's reasonable to spend about $1000 to $1400 on a computer. Some of this expense can be offset by selling the old machine. Or you can donate it to someone who will be very happy with its performance level.

Just my thoughts after 40 odd years of upgrading.

By the way, I appreciate that stumping up $1000 if you are living on a low income and have no savings may not be an option. It's just the way I think about balancing the expense over a three to four year period.

[basa333]

Hi Colin,

sorry for the late reply (holiday).

I will have to make some compromise, but VM is very important to me and I want to spend more time on it, so there is no other way.
Thanks for your time for the performance comparison, it's useful info for me

greets tom