Performance scaling with cores/clock speed

[Overmann]

Hi there,

I'm just wondering if anyone here has done any experimentation with Mixbus and performance scaling? How many cores is it able to utilize (and how well does plugins usually utilize multi-core cpu's) and how important is clock speed?

I've only ever been able to experiment with different clock speeds by overclocking my 5820k, and since that's a chip that overclocks really well I've been able to get a significant performance increase by overclocking. However, it only has 6 cores, and the 2011-3 socket supports upto 22 cores on the Xeon 2699. However, those Xeons run their cores at a significantly lower clock speed then my 5820k. I get very good realtime performance out of my system, and it seems to be well suited to DAW workloads. How would changing to a Xeon (or other 10-22 core CPU) effect performance?

If anyone has tried running Mixbus with such a chip, or have any information about how Mixbus (and plugins) scale with core count, please share!

[johne53]

How many cores is it able to utilize

You'd need to specify which OS you're running. I don't know about Linux or OS-X - but Windows takes no notice of the number of cores you specify within Mixbus (Windows itself allocates processes to cores - an app can request certain cores for itself but Windows doesn't have to accept that request).

[Overmann]

How many cores is it able to utilize

You'd need to specify which OS you're running. I don't know about Linux or OS-X - but Windows takes no notice of the number of cores you specify within Mixbus (Windows itself allocates processes to cores - an app can request certain cores for itself but Windows doesn't have to accept that request).

I'm using windows 10 64bit. Windows let you set affinity, and I assume that if you set affinity to all cores, and all cores within mixbus, then mixbus will be working on all cores?

Either way, I assume Mixbus (and plugins) to work more or less the same on all OS's, does it not? My question was ment as general: What is most benificial to mixbus, many slow cores, or fewer fast cores. And how well does mixbus (and popular plugins) scale with core count.

[johne53]

What is most benificial to mixbus, many slow cores, or fewer fast cores. And how well does mixbus (and popular plugins) scale with core count.

It's not easy to say but I'd guess that for both cases you'll see different results for different sessions. A small session with not many plugins might run better with faster cores (even if there's only 6 of them). But a more complicated session might run better on 20 cores - even when each core is running at a slower speed.

[whiskeynipple0088]

I was originally running mixbus on a 2nd gen i5 2400S - 4 cores at 2.5 ghz - not fun
I had to upgrade to a 2500k - 4 cores at 3.4 ghz - big improvement - usable
Then I upgraded to a i7 3770 4 cores with hyperthreading (8 logical cores) at 3.4 ghz - this was like a 50% increase to my DSP headroom.

I am on OS X and typically have projects between 16 to 42 tracks.

So what I have noticed is both speed and cores matter.
6 out of 8 cores seems to be the sweet spot for my setup. Not much DSP difference with 8 cores vs 6 but it allows other app's in the background to run more smoothly.
I too have been wondering how the xeons would fare but givin the cost and speed, I think the i9 8 core/16 logical cores seems to be the most practical IMO.

[madmaxmiller]

Either way, I assume Mixbus (and plugins) to work more or less the same on all OS's, does it not? My question was ment as general: What is most benificial to mixbus, many slow cores, or fewer fast cores. And how well does mixbus (and popular plugins) scale with core count.

Fewer fast cores. Robin (x42) has explained here in the forum how Mixbus multicore works, it's track-based. Just search for it.
Anyway, as you mentioned correctly, realtime is king, no matter which CPU and I dare say that more cores also create more overall overhead which can influence the actual realtime operation negatively.

MMM

[johne53]

For a long time I was running an old Mac Mini (dual core) although I was mostly using it with Windows 7. Then (a year or so ago) I managed to buy a cheap-ish Mac Pro with a 10-core Xeon (20 cores with hyperthreading). The results are stunning.

Even for a modest session the Mac Mini would typically show CPU usage at round 25% - and Mixbus's DSP could vary between 30% and 70%.

With the newer machine those same sessions show about 9% DSP (with the Portaudio backend) and just 5% if I use Jack!!

But the real stunners are CPU usage and boot-up time. The newer machine takes just 12-15 seconds to boot and I don't think I've ever seen the CPU usage higher than a few percent.