Dual Xeon Scalable LGA 3647 workstation. Anyone?

[Intini]

Hi, everyone.

I am thinking of either building a dual xeon scalable LGA 3647 workstation for Pro Tools (based on ASUS WS C621E SAGE) or getting one ready (HP Z8). I want the best possible workstation for music composition for VIs synthesisers (Roland Cloud, Native Instruments, PianoteQ). It is important to note that the sessions would use mostly these VIs - not one or two...but a lot of them.
I would like to believe that the many cores (up to 58 cores, 116 threads) and great expandability can give me some reassurance of being able to compose without glitches and stops as I really don't want to freeze tracks or increase buffer sizes too much while composing.
I plan to start with cheaper processors (maybe 2 x Xeon 4116) and then later on give some longevity to the system by upgrading to 2 x Xeon 8180, when they hopefully will be cheaper.
I used Pro Tools on Windows for 5 years (some 10 years ago), then now on Mac for the last 7 years and thinking about moving to PC to get more power.
I would love to stay with the Mac OS but the new Mac Pros aren't ready from Apple - only sometime next year and I want to work now.

Does it make any sense? Anyone would care to give me some advice? Has anyone build a Pro Tools workstation with the Xeon LGA 3647 platform?

Useful links for the Xeon Scalable LGA 3647 products cited:

ASUS Motherboard I am talking about:
https://www.asus.com/Commercial-Serv...WS-C621E-SAGE/

HP Z8 Workstation:
https://www8.hp.com/us/en/campaigns/...-z8/index.html


Any help is welcome!

[Bob Olhsson]

The very latest generation 5 GHz. Intel 6 core would be my first choice. More cores will speed up servers and video rendering but slow down real-time audio processing which is what we do. They are not worth sacrificing processor speed.

[Intini]

Originally Posted by Bob Olhsson

The very latest generation 5 GHz. Intel 6 core would be my first choice. More cores will speed up servers and video rendering but slow down real-time audio processing which is what we do. They are not worth sacrificing processor speed.

But aren’t different VIs allocated to different cores according to the system load? This would only mean distributing the processing needs into smaller parts to be able to reach the processing schedule for real time operation of a given audio session, very similar to HDX with respect to resource allocation. You get lower clock speeds BUT each core does much less as the load is distributed.

[Bob Olhsson]

This is at the expense of an increase in latency which audio developers try to avoid.

[The Weed]

Another option is to host your VIs in VEP, maybe even on another computer, perhaps the Mac you're replacing. Others far more knowledgeable than I about integrating PT and VEP will hopefully post.

[Intini]

Originally Posted by Bob Olhsson

This is at the expense of an increase in latency which audio developers try to avoid.

Minimum latency for a given load is related to the capacity of a given CPU to process the DSP in real time and keep the audio buffer fed.
When comparing CPUs with the same clock speed (frequency) and a higher number of cores, you will notice that for a given buffer size (and latency) you actually get MORE processing done (more instances of plugins) within the same slice of time, of course with the same latency (for any given fixed sample rate).

You can see that study here on Sound on Sound, despite not being so new it shows the concept:

https://www.soundonsound.com/sound-a...sors-musicians


Of course, this study assumes you can keep the same CPU clock speeds while varying the number of cores.
In our case, for Pro Tools, assuming that the allocated load to a single thread (of the, say 118) would be able to keep the buffers fed - and thus it needs to be at a minimum frequency for given load -, you would essentially have a LOT of power for VIs.

This is absolutely the same idea as VEPro on many machines but this all happens in a single machine. VEPro is a parallel synchronised computing system with distributed load (even if it is manually distributed). The thing is that maybe VEPro would be cheaper - yet much more complicated and less elegant to work with - than having a single really powerful workstation. Of course VEPro has other benefits for orchestral works (such as keep instruments loaded while changing sessions, the instruments themselves, etc), but this is beyond the scope of this processing analysis.

So, in short, if the cores of a given CPU - of whatever number of cores - is fast enough to keep your most demanding non-parallelizable DSP and other code processing feeding the audio buffer (at the value you set) - and keeping it fed - then you should be fine.

Of course, the question then is: what is the minimum frequency that a core must be run to be able to achieve that with, say, a buffer size of 64, with a load of XYZ Plugins at a certain sample rate. And that is exactly why I place this post. And I am guessing that the answer relies on actually buying the thing and testing.

[Bob Olhsson]

That study is pretty old. My information came from several developers over the past few years based on tests using 6 core and higher processors. They felt higher speed beats more than 6 cores.