more cores or more clock speed? for DP

[Galileo]

Hi there
wich gives the better performance in DP9 more cores for multi threading or more clock speed?
it means, should i buy a new MacPro or a iMac 5K ?
i work in general with 16 channel stereo or mono, min 6 plugins per channel
3 subgroups (aux channels) with 4 plugins per channel, 3 aux (EFX) an the master channel
48khz 32 bit

now im equiped with a 12 Core mid 2012 MacPro / Motu HD 192

thanks for helping

[mikehalloran]

There are many old threads on this. If you are using many VIs, multiple cores is the way to go. Each instance you open will run in its own core. If, however, you run multiple instruments in only one instance of a VI, it will run one core only.

I don't run that many VIs and find an iMac does the job for me. That said, my next Mac will probably be the nicest Mac Pro I can afford.

[Rick Cornish]

Mike O is right.

I would also hasten to add to buy as much RAM as you can afford. I sprung for 32bB on my Mac Pro and have never been sorry. (Getting a couple SSDs doesn't hurt performance either.)

[HCMarkus]

now im equiped with a 12 Core mid 2012 MacPro / Motu HD 192

Based on the description of your usage, I can't imagine you are pushing your current 12 core 5,1 to anywhere close to its capacity. If you somehow are, I would encourage you to consider simply replacing your CPUs (likely running at 2.4 gHz) with 3.33 or 3.46 gHz X5680s or 90s, which you can easily do for under $400. By doing so, you will create for yourself a Mac Pro that is compatible with your PCIe hardware and very expandable yet almost as powerful as the most expensive current Mac Pro. You will also have a more responsive machine, due to the nearly 50% clock speed bump.

Of course, you have lots of RAM and are running PCIe or SATA SSDs, right? And if you need more GPU horsepower, multiple options, up to the fastest GPUs currently made, can be installed in your Mac Pro. All can be done for pennies on the dollar compared to what a new trashcan Mac Pro or a high end iMac will cost you.

OTOH, if you just have an itch to buy a new computer and money is no object, I understand, and let me know if you want to sell your 2012…

[Galileo]

Based on the description of your usage, I can't imagine you are pushing your current 12 core 5,1 to anywhere close to its capacity. If you somehow are, I would encourage you to consider simply replacing your CPUs (likely running at 2.4 gHz) with 3.33 or 3.46 gHz X5680s or 90s, which you can easily do for under $400. By doing so, you will create for yourself a Mac Pro that is compatible with your PCIe hardware and very expandable yet almost as powerful as the most expensive current Mac Pro. You will also have a more responsive machine, due to the nearly 50% clock speed bump.

Hmm i tought the same...
but i have to change for a studio all audio interface to MOTU AVB. and now they need thunderbolt Macs.
it is the reason why they can not use any more these pre thunderbolt MacPro mid 2012.
And they do time to time also really heavy movie scoring sessions with up to 90 channels and native instruments.

[HCMarkus]

Hmm i tought the same...
but i have to change for a studio all audio interface to MOTU AVB. and now they need thunderbolt Macs.
it is the reason why they can not use any more these pre thunderbolt MacPro mid 2012.
And they do time to time also really heavy movie scoring sessions with up to 90 channels and native instruments.

New AVB interfaces can use USB… not quite as fast on i/o thru computer (a little higher latency, only pertinent when performing with VIs and monitoring thru effects) and lower physical i/o channel count. Per MOTU's Mr. Miller:

Yes, up to 64 ins + 64 outs simultaneously over USB at 1x rates. Thunderbolt and AVB are 128 ins + 128 outs.

Maybe you should get the AVB interface(s) first, and try via USB…?

[mikehalloran]

If you are taxing a 2012 12 core, any iMac makes little sense.

[kdm]

While there might be some variation for DP on OSX, plugins and VIs actually benefit more from processor and memory speed than core count. Higher core systems typically benefit disk streaming more.

The other major factor is audio card driver efficiency. RME cards provide a significant boost in plugin counts, and lower real time performance requirements on MAS (core audio / ASIO, etc). The other factor is the interface itself - a USB interface is going to perform quite a bit below a PCIe or Thunderbolt interface, and even firewire, though I've heard some USB2 interfaces perform well. USB3, and especially USB 3.1 should provide an even bigger jump for audio interfaces. But for now, TB and PCIe are the top performers. Couple that with a high clock speed, decent core-count system, and SSDs and it should easily run a few hundred tracks. I'm on PC, but for comparison, I've run over a hundred tracks with plugins on a modest quad core system, so cores aren't the most important factor.

Toss a cpu-hungry VI and reverb or two into the mix and all of a sudden those cores sit barely used, while the performance meter is peaking as the host app tries to keep up with the real time rendering demands, especially if the audio card's buffering isn't as efficient - they you end up running at higher latencies.

[mikehalloran]

While there might be some variation for DP on OSX, plugins and VIs actually benefit more from processor and memory speed than core count. Higher core systems typically benefit disk streaming more...

This is one of those many threads I was referring to. Magc Dave's instructions for taking advantage of multicore Macs are quoted in this thread:
http://www.motunation.com/forum/viewtop ... es#p451456

[kdm]

Hi Mike - I've seen that thread, and how much to split VIs among hosting plugins (Kontakt, Play, etc) has been debated, but in tests by DAW users and DAW techs/builders it's been concluded that this suggestion only applies in limited situations (such as users only running a single instance of Kontakt and fully loading it on a lower powered/low cpu speed system).

Some developers have suggested running one instrument per instance of Kontakt, Play, etc to maximize multithreading, and that I'm sure that is technically correct from a programming perspective. But in reality, each instance also uses more memory, and some amount of cpu power. So what you gain in marginal core splitting, is lost in the additional memory and cpu loading. For running a lot of VIs as with a scoring template, we'll have dozens of instances of Kontakt, Play, etc in VEPro, which does it's own multithreading.

For a mix of audio and plugins, the cores vs. cpu tradeoff really depends on whether you run smaller projects (less than 50 tracks) with resource intensive plugins (multiband compressors, reverbs like 2C's B2, etc), or run a lot of tracks with basic EQ and compressors on each track. The former would benefit from faster clock speed, the later from additional cores (though both with benefit from the faster clock speed).

As another poster said, I would be surprised if a 12-core 2.6Ghz system is being pushed by 16 tracks of audio. However, it could easily be maxed out with some resource intensive plugins.

Here is a great video created by a VI-Control member on general DAW performance and where cpu power and other system specs intersect, and diverge: http://vi-control.net/community/threads ... daw.46807/

[bayswater]

Here is a great video created by a VI-Control member on general DAW performance and where cpu power and other system specs intersect, and diverge: http://vi-control.net/community/threads ... daw.46807/

Thanks for posting that. Explains a lot. I wonder if anyone publishes a real time performance benchmark?

[kdm]

These are a few years old, but will give you an idea of realtime benchmarks:
http://www.dawbench.com/win7-v-osx-1.htm

Page 6 of these tests covers VI performance. I would be interested to see recent numbers with Win10 and OSX. As OSs, cpu families and builds change, performance can as well. To me, the tests of most interest are audio interface comparisons. Performance can vary wildly from very good low latency performance with RME, to poor performance with some low cost USB interfaces.

[mikehalloran]

Great video, explains a lot. Doesn't change my answers at all.

What did we learn? That SSDs, more memory and buffer size have more to do with performance than the CPU? OK... That's not news, really,

Since MagicD was writing specifically about VI performance in DP, I have to take his word for it on the multicore issues with VIs.

A 5K iMac might do the job -- it can certainly handle anything I do now. I know this because I'm using similar machine but with a lower clock speed and an SSD at SATA II instead of a PCIe blade. So what? I'm not you.

Any of the new Mac Pros can outperform the iMac any way you want to measure. Forget what you are doing now. What do you want to do in the future? You are buying for then, right?

Heavy use of VIs in DP benefits from more cores if set up right; audio does not. It depends on what you are doing now and what you plan to do. Really. I'm only restating my original post.

[kdm]

I understand, but I do doubt that DP operates differently than any other DAW since this is really determined by basic computer architecture. I'm just passing along what DAW experts and computer techs have all said, and what I've found in my own observations and tests. If you have tested and found deviations from any of this in your own work with DP, that would be great know - not everyone's requirements will fit a specific either/or recommendation.

I've heard the opposite core vs cpu recommendation from reps and developers recommending the many instance, few instruments per instance setup (which was disproven to be far less of a gain than the memory and cpu loading costs). Nothing against Magic Dave as I'm sure he is only stating the information he has (and it may be based on a general recommendation, not the specific scenarios we are discussing here).

This is what many of us have found in real world testing and comparing of different systems.

For VIs and plugins specifically: I can tell you that DP does not perform 3x better for VIs or plugins on my i7 that runs 12 virtual (6 physical) cores, than it did on my quad core that has 4 cores - I ran DP on both, and had Omnipshere, Kontakt, Serum and Play running on both.

Clock speed on my i7 is roughly 18% higher on my i7 than my quad (3.3G vs 2.8G). VI performance might be 30-50% better in DP (and Nuendo). Some of that is likely from a much higher memory speed as well - memory speed on the i7 is quite a bit higher than the quad as I intentionally went for as much increase there as I could afford (64G, DDR4 vs. DDR2 in the quad).

Quite frankly, I was expecting a very large jump in performance when going to this hex (12 virtual core) i7, but it didn't happen, and the reasons stated above are why. At first I thought I had something setup wrong, but I didn't. It's just the reality that multicores only benefit certain processes, and virtual cores only provide a marginal performance bump, not an exponential bump. You will find the same results from the film and graphics design worlds - only certain fx/render processes benefit from multithreading, but not processor intensive fx (similar to VIs like Omnisphere, Serum, Reaktor, heavily scripted VIs in Kontakt, etc). VIs and VI players are typically designed to load a single thread. That's why Dave (as have other developers) suggested running fewer instruments per instance of, for example, Kontakt or Mach5, and more instances of those sample-players. It appears that more cores would benefit that scenario, but it does not for the reason that most of the resource requirements are cpu speed, and disk streaming, not process threading.

I found my new system scaled more directly with clock speed. Cores handle threading, which is where audio streaming is managed - hence you get better performance with large audio streaming systems with higher cores - but a much bigger jump with SSDs. DAW builders have typically recommended dual xeon systems for audio post, mixing, and live tracking, but high clock speed i7s, etc for VIs for these reasons.

Of course the more of anything, the better, and that includes cores. This is only comparing relative improvements of clock speed vs. cores for specific uses.

For the OP, if recording and mixing is the main focus would gain better mixing plugin performance (reverbs, eqs, comps, chorus, etc) with faster cpu speeds. If he wants high track counts (over 100 tracks) and high plugin performance - the MacPro would be the way to go.

I didn't see a mention of VIs in the original post, but which VIs you use most also factor in - streaming vs. synth/cpu heavy have almost opposite requirements.

[nk_e]

Hi all.

I found this video to be very helpful in understanding the basics on some of these issues. Perhaps the OP will also.

It's very well done and there is some advice (PC centered) on improving RTP towards the end around the 20 minute mark.

https://m.youtube.com/watch?v=GUsLLEkswzE

Cheers.