Composer's Workstation

Configure and optimize you computer for Audio.
RELATED
PRODUCTS
Apollo x6 Cubase Pro 13 Vienna Ensemble PRO

Post

Hello KVR,

Occasional reader, first time poster. This is somewhat of a cross-post to an identical thread I created on the Linus Tech Tips forum within the past hour. I am posting this here as well because this is an audio-centric forum that will no doubt have even more input on this subject. Also, I plan on posting here a lot as I learn more and grow as a producer and artist.
  1. Location and Unknown Budget
    This build will be conducted in the U.S. and will have an unknown budget. I would prefer staying close to best performance per dollar, unless spending a bit more allows more future proofing.
  • Aim
    I want to build a Intel-based composer's workstation using multiple Mini ITX systems as networked sound modules that are controlled by one (up to EATX) master system using Vienna Ensemble Pro 6 and Cubase Pro.

    My maximum number of systems is currently hovering around 3 sound module builds and 1 master for a total of 4 systems, but this could (and probably will) change.

    The basic idea here is to load up instances of sample libraries and processor intensive VSTi and spread that across several barebones rigs so that the CPU strain can be offloaded and so that all of those resources are ready and available while producing music.

    Junkie XL has a similar setup with 6 sound module PCs and one master system.
  • Displays
    Main displays: Samsung C34F791 34" Curved Displays (3440 x 1440)
    Why: I want an immersive experience with the option to arrange and mix simultaneously. I am open to changing this to one display if it is a 4k model. I would prefer it to be curved, but I don't want to spend thousands on it.

    Secondary displays: AOC I1601FWUX 15.6" IPS (1920 x 1080)
    Why: I would like to run ShareMouse on these machines so I can make adjustments as needed without using Remote Desktop to keep latency low. This means I will need a small format display for my Mini ITX systems purely for making those occasional adjustments. These, or something better, could be placed on my desk for seamless access to the desktops of the sound module PCs.
  • Peripherals
    This is a "from the ground up" build, meaning everything is being planned and purchased from a starting point of having nothing.

    I still own my prized Logitech G9x and have enjoyed using it for navigating my studio sessions in the past, but I am interested to know if there is a successor to this mouse I should consider; one that is just as dependable but even more customizable, in terms of available buttons.
  • Why I Am Upgrading
    • I need to arrange and compose music in an uncompromising digital environment.
    • I need to have a dependable, powerful, future-forward network of machines that will serve me for the next decade or so and can be upgraded as necessary.
    • I need a modern build plan to realize the goal of moving toward music production as a full time passion.
    • I need guidance to determine what is a good plan versus what is merely an expensive plan.
  • Need More Input
    Now we come back to the "less is more" part of it.

    Junkie XL, Hans Zimmer, Current Value, and Deadmau5 all use multiple networked computers in their professional workstation setups. A network of slaves tied to one master over LAN is not a niche or esoteric thing to build in this context, but increases in computing power would seem to indicate the number of machines required to achieve the same results will be fewer. This is where I quickly become confused: how many systems, theoretically, should I aim for?

    Some things to keep in mind are that I want to connect my main system to a UAD-2 Apollo 8p to offload even more work when using UA plugins, but I won't be using that all the time. I also plan to use Focal Twin6 be as my reference monitors.

    I am concerned that using the TRS outputs on the Apollo will be detrimental to a clean sound on the Focals, but from what I understand, balanced TRS cables and XLR/XLR cables are internally the same, meaning a shielded TRS/XLR between the Apollo and the Focals should be just fine. I hope?

    I had always heard to use XLR whenever possible, prompting me to ask.

    Here is my first thread over on Level1Techs about my initial ideas for this project (if you want more reading).

    Also, here is my duplicate of this thread over on Linus Tech Tips.

    And here are my two designs for the Mini ITX slave systems on PC Part Picker (yes, I know I don't need to include Optane in the Z370, but who knows I might need legacy drives in a NAS at some point):
    Here are some basic rules I have already mapped out, but need to vet if these are faulty assumptions on my part or if they are valid. They seem to be valid based on research, but I am open to removing my own bias.

    No AMD

    Latency is huge on Ryzen, unfortunately. Because of the NUMA on Threadripper, the per core performance, floating point throughput, and overall Dawbench scores are mighty sad when compared to the 8700K, 7820X, and anything higher from Intel.

    My old system that is long since past dead was an AMD 9950 quad and simply moving from that entire PC tower to a Core i7 MacBook a couple of years later was enough to experience a nearly 4x performance boost over what I thought was a powerful option from AMD at the time.

    The playing field is much more level now, but in terms of audio that demands low latency, Intel easily still wins.

    Preferably X299

    I had contemplated building everything using an 8700K, like the Z370 build. This was because per core performance, number of cores, possible overclock, and performance per dollar all seemed to line up within that one chip.

    But since the price difference between the two PC Part Picker lists is less than $300 (at current prices), I would opt for the X299 option because, for a nominal added cost, I could get more lifespan out of an LGA-2066 design.

    Even if the Z370 option is a better performer due to cost considerations, it is still only preferable at the moment and pursuing it solely on lower cost ignores the maturation of X299 moving forward. At least, that's my rationale for opting to go this route.

    UPDATE:
    Before publishing this thread, I have received info over at the LTT forum that 8700K still has less latency than X299 as of the timestamp this post was published.

    Water Cooling

    I am aware that all systems in this network would benefit from or even require water cooling. I welcome this, having never built a water cooled system before and having taken great pride in seeing previous projects through to completion with my own two hands.

    What Main System?

    I think what I need the most help with aside from how many systems I should build is how to sensibly plan the master system that brings all of this together into one place.

    I do not currently see any benefit whatsoever in trying to run something like a 7980XE, especially since that price tag exceeds the entire build of one of the Mini ITX modules. But you can try to convince me.
I can share more about my concept for the main system in this thread, but for now I'll let this information stand as a "rough draft" of my intentions.

Thanks for reading!

Post

Courtesy of @brob over on the LTT forum, here is a more rational list of parts with adjustments to the processor (do not need unlocked / overclocked for slave PC), cpu cooler (do not need water cooling for slave PC), and RAM (less CAS latency than my pick):

VE Pro Sound Module PC - brob at LTT

Looking forward to designing the main system with everyone in the coming days/weeks so I have a better overview of the overall puzzle and how it will fit together.

Post

You won't really see any benefit going with 3200 memory here. 2666 is the highest spec memory officially supported and going over that is technically overclocking as far as the system is concerned If 2666 happens to be cheaper where you are, save the money. Rest of the Module system spec is solid through.

Post

I have just discovered this three part tutorial by Cory A. Robbins.
Vienna Ensemble Pro Setup in Cubase - Part 1 Organization and Connections

In that video at the beginning, he describes using (more or less) this server: Dell PowerEdge R710 Server
His had two 6 core Xeons in it, 128GB RAM, and 4 x 450GB SAS. The one linked has 8 x 300 GB SAS in it.

My thoughts on using old servers went something like this:
  • Wow, Kontakt actually works on that hardware?
  • It's cheap, but would I get burned by pricier server parts when servicing a clunky refurb unit?
  • I'm going to be stacking these inside of rack mount cases anyway and those are near $100 each, empty...
  • I don't like having to use 4U rack mount cases to fit my i7 coolers, this looks like a space saver with less headache...
  • The videographer is a day job IT professional, so he has a more informed purchase decision than Joe Public...
  • Why not just get some of these and save that money for the main system instead?
But then, I'm sure there are reasons not to buy these. Like, Xeons (even dual socket configs) being really weak compared to single socket Core i7 8700's. Or parts being iffy when dealing with refurb units. I'm sure there are a billion ways I could talk myself out of this, but it's very compelling and I'm wanting to say yes because of how much of a turn key it would be by comparison. It's over $500 in savings per system at fixed costs.

I think the case to be made here is that this option is a greater value, professionals are already using these units and it's working very well for them, and the parts are actually rated to last longer than the consumer market. So, I guess it makes more sense to go for a small batch of these and only focus on the details of the main system design.

What do you think? Is this an equivalent, worse, or better idea?

Post

Kaine wrote:You won't really see any benefit going with 3200 memory here. 2666 is the highest spec memory officially supported and going over that is technically overclocking as far as the system is concerned If 2666 happens to be cheaper where you are, save the money. Rest of the Module system spec is solid through.
Thanks @Kaine,
I had these same thoughts regarding no need for OC memory, especially with it being the most expensive part in the parts list. I don't actually know what would be the best way to do this at this stage of considerations.

Refurb servers are something I hadn't considered until now, but that is how Junkie XL runs his. They look like they're just refurb units with his instances installed on them, and there are documented use cases for just such a rig on that kind of hardware. That would mean all I need to do is design the main system after having quickly gathered my prefab sound module systems.

I think the servers make more sense due to higher max memory, stability, longer lasting parts, and the low level of workload these things will be tasked with. It suddenly makes very little sense to build my own units when all they will be doing is running audio and MIDI with a whole slew of samples sitting in RAM.

Post

Quick update to this build thread, I have been conversing over on the LTT board and I have narrowed the list of parts for the sound module PCs that I'll be running my Kontakt instances on, etc.

The list, as it currently stands, is as follows:
VE Pro Sound Module PC - ATX 2

The CPU and memory were chosen because the 8700 does not need to be K series (unlocked) for this purpose, and RAM of this speed yields better performance than simply max spec as indicated on the motherboard.

The motherboard was also selected for its stability and to allow for 64GB memory instead of 32GB on Mini ITX. Mini ITX is also more expensive on average than ATX due to tighter tolerances during design and fabrication.

There will possibly be one more revision to place this build into either a 3U or even 2U rack mount chassis, but if that doesn't work then this will become the final build for my library machines.

Main system has yet to be designed. It's next on my to-dos.

Post

WhizzingGosh wrote:and RAM of this speed yields better performance than simply max spec as indicated on the motherboard.
I'll repeat what I said before... no, it doesn't. Something just about every DAW builder has discovered over the years of testing.

Within an audio dedicated system the bottle neck isn't memory speed, it's ASIO. You're not pushing the memory in the slightest when dealing with real-time audio buffers.

Sure, it gives you more performance for video editing, graphic rendering, gaming or running databases all day long, but I'm assuming that isn't your usage scenario?

We sell 3200MHz and up and I'm pushed to include it on my systems, but point blank refuse as I view it as a waste of money for the average audio user. If I saw any benefit to it, i'd include it in a heartbeat!

Anyhow, remember to spec extra replacement fans if you care about noise, I've never heard a rack case that didn't require an upgrade in that department.

Post

Kaine wrote:
WhizzingGosh wrote:and RAM of this speed yields better performance than simply max spec as indicated on the motherboard.
I'll repeat what I said before... no, it doesn't. Something just about every DAW builder has discovered over the years of testing.

Within an audio dedicated system the bottle neck isn't memory speed, it's ASIO. You're not pushing the memory in the slightest when dealing with real-time audio buffers.

Sure, it gives you more performance for video editing, graphic rendering, gaming or running databases all day long, but I'm assuming that isn't your usage scenario?

We sell 3200MHz and up and I'm pushed to include it on my systems, but point blank refuse as I view it as a waste of money for the average audio user. If I saw any benefit to it, i'd include it in a heartbeat!

Anyhow, remember to spec extra replacement fans if you care about noise, I've never heard a rack case that didn't require an upgrade in that department.
@Kaine, thank you for clarifying these details for me.

The original context for my memory selection comes from this post over at Linus Tech Tips, which references this benchmark article.

It didn't seem correct to my purposes, especially when I have heard that you don't want to be doing any overclocking when it comes to audio work. It makes sense that this gives preference to system stability and per core performance over fast for fast's sake, which seems like it would give rise to a lot of dropouts due to pushing the processor too hard for the tasks performed.

Your advice sounds more reasonable than strictly going for something that is good for gaming or rendering. The main computer may be a bit of a different story, as I would like it to also be efficient at video editing and other visual work as a secondary function to being my main DAW machine. Maybe I should segment that into two different systems with one audio and one visual. I don't want to build more systems than I really need to.

So let me ask you a couple things beyond just RAM.
  1. Refurb Servers
    Is this a more sensible option or should I stick with building to suit as I am currently planning?
  • Single Machine
    Is this entire plan outdated compared to the processing power of a single decent workstation build, or is there still some merit to pursuing networked systems using VE Pro 6?
There is a lot to figure out here. For example, is Optane a benefit in the context of Kontakt sound module PCs/servers, even if they have SSD instead of spinning drives? I would think the answer is no, but again, there's a lot to figure out.

Post

WhizzingGosh wrote: It didn't seem correct to my purposes, especially when I have heard that you don't want to be doing any overclocking when it comes to audio work. It makes sense that this gives preference to system stability and per core performance over fast for fast's sake, which seems like it would give rise to a lot of dropouts due to pushing the processor too hard for the tasks performed.
It's the other way around in that if you push a system too hard in that regard it won't drop out, it'll just blue screen.

Out of the box chips are designed to stagger their cores, so on an 8 core for instance:

2 core mights turbo to 4.5
2 core mights turbo to 4.4
2 core mights turbo to 4.3
2 core mights turbo to 4.2

I tend to test systems with them mildly overclocked and with all those cores set to the max turbo limit, so they end up locked off at the maximum 2 core speed rating. The reasons is, is that for audio handling (ASIO again) a machine is only efficient as the workload that can be done within a buffers time frame. If one core is running 10% lower than the highest core, then you run the risk of that core overloading and taking down the rest of the cores with it. A general rule to keep in mind is that the CPU for audio is only really as effective as it's weakest core.
WhizzingGosh wrote: Your advice sounds more reasonable than strictly going for something that is good for gaming or rendering. The main computer may be a bit of a different story, as I would like it to also be efficient at video editing and other visual work as a secondary function to being my main DAW machine. Maybe I should segment that into two different systems with one audio and one visual. I don't want to build more systems than I really need to.
Nor do you need too. My earlier RAM comments were simply regarding saving money if you don't have the need to do work that would take advantage of it. If you're going to be doing some video editing along the way, then sure go for the upgraded RAM.
  1. Refurb Servers
    Is this a more sensible option or should I stick with building to suit as I am currently planning?
  • Single Machine
    Is this entire plan outdated compared to the processing power of a single decent workstation build, or is there still some merit to pursuing networked systems using VE Pro 6?
The biggest problem with the servers is the noise. If you can set them up as network attached nodes in other rooms, then go for it. Just make sure you understand exactly what you're buying, as the difference between say a V2 and V4 Xeon chip can sometimes be pretty large as far as performance can be concerned.
There is a lot to figure out here. For example, is Optane a benefit in the context of Kontakt sound module PCs/servers, even if they have SSD instead of spinning drives? I would think the answer is no, but again, there's a lot to figure out.
Optane has proven to be of little use to anyone so far, since they've failed to really scale up or bring the prices down since launch. The biggest gains for it will be large databases and mail servers, once they get the capcity up and the pricing down. 4k video editing scratchfiles and that kind of thing will pay off, but for audio not really. The bottleneck is the ASIO buffer and unless you find yourself satuating the SSD's then you'll see little to no benefit going with anything faster.

Post

Kaine wrote: Out of the box chips are designed to stagger their cores...

I tend to test systems with them mildly overclocked and with all those cores set to the max turbo limit, so they end up locked off at the maximum 2 core speed rating. The reasons is, is that for audio handling (ASIO again) a machine is only efficient as the workload that can be done within a buffers time frame. If one core is running 10% lower than the highest core, then you run the risk of that core overloading and taking down the rest of the cores with it. A general rule to keep in mind is that the CPU for audio is only really as effective as it's weakest core.
Been a bit of time since I posted here, had to think things over.

Based on news that has come out of Computex, AMD is looking more and more compelling in terms of system design, especially when considering TDP and bootable NVMe RAID.

Because of the difficulties in deciding between the two platforms, I think it really comes down to driver support and system stability overall.

The biggest concerns I have regarding AMD as an option are as follows:
  • Increased Latency from NUMA
  • Forced System Reboot When CPU Stressed
  • Random Crashes When Running DAWs
  • Lack of Audio Interface Driver Support / Unstable ASIO Drivers
  • No TB3 Support / Only "Slow" Connections
What draws me to AMD is the lightning fast response that bootable NVMe RAID would give me for workflow and loading / saving projects. I think it would be a fantastic thing to have nearly zero wait time in that regard. Also, it costs less and uses less TDP, meaning running the systems also costs less, not just the parts. Obviously there are the much larger numbers of PCI-e lanes, larger max capacity for RAM (eventually), DIMM.2, and non-bottlenecked M.2 slots to consider when sizing up Intel's clumsy and expensive looking VROC.

Let's rewind and revisit the purpose of this build:
  • The main system should be the strongest machine, capable of doing a lot of real time processing, such as VSTi that require calculations to achieve their sound. It's basically going to be the "brain" of everything. It's where the actually composing / scoring is done. I would like the flexibility to also do some visual work. Basically Adobe CC with an emphasis on video editing and after effects.
  • The sound module systems are going to be primarily used for loading many instances of Kontakt and orchestral libraries. By their very nature these libraries are more or less sample based and require a lot less CPU to simply stream out audio as opposed to generating it.
While the Core i7 8700 (non OC version) is a very good value, it seems as though it's overkill given the context of what the sound module units will actually do. If I need to spread out entire libraries of VSTs across certain systems, like a Waves bundle, then I would probably build those specific systems with the 8700 in them for better performance. But otherwise, I don't think it's necessary.

When speaking in terms of socket TR4, so 1950X and its successor, could the latency of NUMA be mitigated by using virtualization to effectively split up resources into "several" (say, 4) machines in one sound module, effectively quadrupling my return on investment as opposed to much more costly Intel systems?

Can I quietly water cool rack mount cases with slight modifications for running loops? Is that as much of a nightmare as it sounds? Water cooling is a requirement on socket TR4, based on everything I've ever seen, heard, or read.

I want to say yes to AMD, I really do. The price to performance ratio is very tempting. But I have been burned in the past, it seems to good to be true, and the DAWBench numbers still look very lean compared to what Intel chips can deliver. Still, those benchmarks are aimed at doing everything on a single machine. My use case is a bit different. It all boils down to this:

I need my main system to be able to run as much as I want it to, within reasonable limits, taking Cubase Pro 9.5's "offline processing" into account for added optimization of project files.

I need my sound module systems to be reliable and tailored to the tasks they are meant to do. I'd rather not build something with more beef than the job requires whenever possible, but not penny pinch it to death either.

I require all systems to be designed with longterm use in mind. Dependable, serviceable, upgradable, salvageable / harvestable.

I think the Core i7 8700 PCPartPicker list is still a valid starting point, but I guess I need to know if AMD is even worthy of consideration given my intentions. I have seen people praise Ryzen, but there is virtually no real world user experience regarding Threadripper 1 or 2 as an audio production machine specifically for use in composing with large template files.

Help me decide how best to not be able to decide.

Post

It still seems the AMDs aren't as battle tested as the Intel's. Maybe contact Cory A Robbins and see how his setup is running after a few months of use and if he has any new advice. Try building the main master computer, get that working smoothly with Cubase, VE and the Apollo, and the visual editing software. At that point you'll have a better feel for how much more power you'll need from your networked slave computers.

Your situation reminds me of an article I read about marketing/shopping analysis. Granted this is an oversimplified example and I don't in any way mean to trivialize your build dilemma (I recently built a modest DAW and feel your pain). If a consumer is presented with too many options they more fear picking what they see as the wrong option than the correct option and wind up unable to make a choice. Your situation is much more intense and complicated than choosing between several cans of chicken soup but the end result is the same. Good luck.

Post

WhizzingGosh wrote: What draws me to AMD is the lightning fast response that bootable NVMe RAID would give me for workflow and loading/saving projects.
How big are the projects, that that'll make a difference? Also note, if by this you mean the sample libraries then be warned that once the queue depth gets ramped up (easy enough if we're talking sample libraries like Kontakt/vsl etc...) then an NVMe will crawl to close to SSD speeds as they don't do well with small files and long queue lengths.

Admittedly, i've not tried that with a Raid 0 which I think is what you're suggesting?
WhizzingGosh wrote: When speaking in terms of socket TR4, so 1950X and its successor, could the latency of NUMA be mitigated by using virtualization to effectively split up resources into "several" (say, 4) machines in one sound module, effectively quadrupling my return on investment as opposed to much more costly Intel systems?
In theory, if your virtualization instances were locked to a given core + memory channel, I guess it could be possible. I'm not sure if you can do that successfully in the real world, however, so one to research.
WhizzingGosh wrote: Can I quietly water cool rack mount cases with slight modifications for running loops? Is that as much of a nightmare as it sounds? Water cooling is a requirement on socket TR4, based on everything I've ever seen, heard, or read.
You can attempt to water cool anything. Even if a dedicated AIO won't work, you could roll your own with pipes and a pump. It'll just be case permitting and you might need a Dremel.

Post

Right now I see AMD more aimed at "computer nerds" like myself... :)
I mean, if any problem/odd behavior happens can you identify the culprit
and fix/remedy the situation?

AMD is toooooooo picky about memory and seems to benefit from faster
RAM with tighter/custom settings, the ultimate optimum performance
is achieved with only 2(single rank) RAM sticks(Treadripper is 4 sticks)
and today this means 8GB per stick, the 16GB sticks are dual rank and do
not achieve the same speed(maybe still good enough), fill all the Ryzen
4 memory slots or Treadripper 8 and it will even get slower...
Some software will benefit from faster RAM with tighter settings and I do
not know if this includes audio stuff, maybe low ASIO buffers...
https://forums.anandtech.com/threads/ry ... t-39391302
https://www.overclock.net/forum/13-amd- ... m-am4.html
https://www.io-tech.fi/artikkelit/ddr4- ... rituskyky/
https://www.computerbase.de/2018-04/amd ... _steroiden

A custom water cooling can be the ultimate silent solution if you
place the pump/radiator/fans in another room... :love:

Post

Pictus wrote: A custom water cooling can be the ultimate silent solution if you
place the pump/radiator/fans in another room... :love:
I had a collegue who used to keep his radiator living on his office balcony.

A Zalman Reserator in a vented cupboard isn't a bad idea either... if they still make them.

Post

Good, if it is a cold place must not forget to use some antifreeze fluid.
Can place the radiator in the swimming pool or pond. :lol:
Have to make sure the pump has enough pressure.

They do not make anymore, but can still find external stuff
at https://www.watercoolinguk.co.uk/cat/Ex ... s_652.html

Found a video of the Aquaduct 720 XT Mark V :love:
(Jump to 15 minutes)
https://youtu.be/wAt2Bqt74So
Cheaper to make our own with eBay copper radiators...
Or any machine/vehicle copper radiator...

Post Reply

Return to “Computer Setup and System Configuration”