I would buy a turnkey system, tell your requirements and wait for an offer.

non ECC based systems here: https://www.da-x.de/
ECC based systems here: http://xi-machines.com/de/einsatzgebiete-audio_o.php

da-x is unflexible when it comes to OS support. They didn't sell me a Win7 based system when Win8 was out, which finally brought me to the point to build myself Xeon based.

If you buy such a system I would still try to request that the components still support Win7 (as a "plan b").

As an option I would request a Win7 / Win10 parallel installation on different SSDs.
Shall it turn out that Win10 has issues you can very easily fall-back.

I do it the other way around as Win10 doesnt meet my personal requirments in terms of privacy and other things.
So I run on Win7 and have a Win10 test system to keep an eye on it.

Methodology to save time (installing all applications and VSTs is very time consuming): I reinstalled Win7 with all applications freshly and used the Win10 upgrade installation (which was very good) so that I didn't have to reinstall all of my applications again.

If you plan to buy an SSD then save money by getting Samsung EVO series disk. Plan it to be big enough, so that OS, Programs and Sample Libraries can be placed on this disk. I choosed an 1 TB Samsung EVO. For Win10 I used an older 512 GB disk and removed some of the very big Superior Drummer 3 sample libraries. The basic libs are sufficient for a test system. I use in my system an older Samsung SSD 830 for Recording and Video projects. The key reason was to re-use existing disks and to distribute i/o load across different disks. For User Bulk Data I use a silent 3TB Seagate 7200 U/min. For backup I use a 10 TB Seagate Enterprise grade disk, which turns off when not being used (also 7200 U/Min).

If you want to use only Win10 then I would still get a 2nd SSD, this would enable you to clone the whole installation to 2nd disk before performing Windows 10 upgrades. With the upgrade from 1709 to 1803/1804 there were severe problems for some systems / types of SSD .. so if such an update creates problems, you can keep this system state for troubleshooting and continue to work on the cloned system.

Your user data should then of course reside on a different disk.

For backup purposes I would plan for a bigger internal disk which is capable to hold disk Images of your 1 or 2 OS deployments (Macrium Reflect) and a mirror of your user profiles, user and project data (freefilesync).
In terms of disk you should plan one internal and one external disk for backup purposes only.

Much cheaper than buying NAS is to connect an extermal disk via USB3.1 gen2 and getting a case for it which supports this as well and on top UASP (USB attached SCSI). I use this combination of USB3.1 controller and external case / disk successfully (on a new board I would expect it to have USB 3.1 gen2):
https://www.heise.de/preisvergleich/sta … 03149.html
https://www.heise.de/preisvergleich/sta … 29902.html
This disk is silent but maybe you want it to have more capacity: https://www.heise.de/preisvergleich/sea … 03248.html

My current setup you can find here: https://www.tonstudio-forum.de/blog/ind … mponenten/

Performance: https://www.tonstudio-forum.de/blog/ind … cks-de-en/

Other things to consider:
1. Performance after microcode upgrades, still well enough, but ...
https://www.tonstudio-forum.de/blog/ind … tre-EN-DE/
2. Some of the reasons why I still prefer Win7: more proven, mature, less changes and has no "hidden" performance issues like Win10 for over 2 years
https://www.tonstudio-forum.de/blog/ind … s-7-EN-DE/

Therefore I tend to suggest to deploy Win7 and Win10, as I do not trust Win10, its a "moving target", too many changes, thus too many mistakes.

there are lots of things to watch out; and some of my thoughts:
1) PCIex lanes layout - not every slot is running lanes from cpu, some even conflict with nvme controller, so make sure to consider what devices and cards you'll use and if you'll have enough PCIex lanes and PCIex slots
2) many recent motherboards have some power-saving features, which may even prevent HDSPe cards from being initialized in OS or perform incorrectly, so doublecheck these features are disabled
3) most motherboards have poor power design of PCIex slots, so it's very likely for electrical interference on wired inputs/outputs with let's say graphics card or cpu phase changes (not essentially a problem for you particularly, since raydat has optical connections)
4) ecc and xeon is in my opinion waste of money and inferior performance result
5) high-end chipsets offer additional functionality, which may introduce some latency, longer boot etc. but I'd say it's not a noticeable problem
6) for audio workflow, high IPC (instructions per clock) matters more than amount of cores, 7820X is decent choice, even 7800X would be good enough
7) speaking of RAM, I definitely encourage you to get 4x8GB (32GB in total) as the cpu supports quad-channel memory and also it will save you couple more years before need to upgrade, and especially since you plan to use virtual instruments I wouldn't hesitate to call 32GB a reasonable minimum RAM config
8) Windows is "the problem" regarding latency, reliability, even the whole DPC is "problem" of Windows NT design, no such thing on MacOS as far as I know - prepare to spend some time optimizing the OS, it'll be worth it, but gonna be some hassle
9) overall your planned parts list is reasonable, I have similar specced rig for about 3 years (i7-5820K, ASUS X99-S, 32GB DDR4 2666MHz Kingston, HDSPe AIO, EVGA 980Ti Classified, EVGA G2 750W, 500GB 850EVO) and it let me down only once - when after BIOS upgrade I had to manually re-initialize RAM sticks

Compare Performance / Single Thread Performance / Base / Turbo / # of cores
https://www.cpubenchmark.net/cpu.php?cp … mp;id=3098
https://www.cpubenchmark.net/cpu.php?cp … mp;id=3038

Builtin Intel Graphics is not a favourite of mine, better a dedicated GPU.

You wont reach Turbo speed with neither of the two, its desirebale to avoid clock changes.
So you will end up with maybe 200-300 MHz more compared to Base Clock.

Maybe 2 cores more are better for you, as the clock will be high enough with approx 3,8 GHz.

Socket 2066 will give you double memory BW, so maybe the i7 7820x.

I would take a supermicro board if possible.

Supermicro is a company selling enterprise grade stuff. Server, workstations. If I have an issue, I get even as end customer usually a reply to my mail on the next workday or I can even call them by phone and get a support technician.

I made very good experience with a server board from them. I know that a manufacturer of turnkey systems for audio also choosed mainboards from them.

I heard last recently that Gigabyte has bad support and I read already strange things about issues with their boards / bios.
I had already a MSI board with bad design and one which was not ideal.
Asus is IMHO the best company still when it comes to delivery of "consumer" grade stuff.

Can you run this project without any plugins at lower buffersizes?
Maybe a program made for live mixing instead of reaper will give better results? Doesn't waves have that, I think it does.
Does removing the send fx like reverb completely from the project allow lower latency? If yes there might be other solutions for reverb etc which do not influence performance.

@vinark

If I can good remember, I believe I already did the test and the answer was "yes".
I have to test it again the next time I will be in rehearsal to be sure that the glitches come from the processing of plugins.

No go. Reaper is mandatory because it's a really good piece of software, very stable, affordable, very good maintained and we are experienced with it.

What would be these solutions?!?
The RVerb and TrueVerb plugins are pretty lightweight, don't they?!?

@ramses

thanks for the link about the hardware. Maybe we will try a Supermicro MoBo with Xeon CPU combination. I have to spoke with the members of the band if we could spend more money for it...

I get same results if I replace all the Waves plugins by the Reaper built-in VST and JS plugins. In our opinion, the Waves plugins sound better because they bring some nice colorations to the sound and they are for us easier to get set right.

We don't need the Izotope for live purposes. These plugins are very heavy on CPU.
I think the Izotope are very good in the production process at the master stage.
What we here discuss concerns only effect processing in a live context.

I more and more think that the ASUS mainboard is not quick enough for real time processing.
It's a consumer grade X390 chipset, mostly designed for gaming and overclocking capacities.

This you could eventually find out with the help of LatencyMon, although I have to say that I am not
very familiar with the values that I got approx 2 years ago with LatencyMon under Win10.
Because of changes in Windows they had to change the measurement methodology which
delivers other results, not anymore down to 1 microsecond, more smth in the range of 100 microseconds,
so to say factor 100 less accurate (IMHO).

Consumer Mainboard does not necessarily mean that it does not work reliable.
The problem is simply that there are chipsets / products on the market which perform well,
and some that doesn't.

I made myself this experience with an MSI mainboard approx 6-8y ago.
I could optimize everything, OS and BIOS, but still I got kernel latency timer results not better than 60-100 microseconds.

Then I bought the same board with the letter "A" behind. Same board/chipset, only with a few improvements in terms of SATA6 and USB3 support.

The same installation (I swapped only the board, kept the same installation on same disks) of all sudden had kernel latency timer values down to approx 2 - 20 microseconds.

There are 2 different measuring tools. The DPC Latency Checker tells you everything with spikes up to 1000 microseconds is fine (more or less), The truth is, the more occupied the CPU is by other priotized taks the less agile / responsive it is to serve audio in time with lower ASIO settings and increasing system load.

As a rule of thumb I can only tell, to optimize BIOS and System to get DPC latency as low as possible and that you get a stable / steady CPU clock, that does not change as even clock changes introduce a little latency.

On Win7 I have the information and tools to build such a system. On Windows 10 the problem 2y ago was, that you can not use LatencyMon anymore to measure kernel latency timer which is in my opinion the best measurement, as you get values down to 1 microseconds and by this kind of higher accuracy in your measuring.

Am not sure whether this changed now with latest versions of Win10 and LatencyMon. 2y ago you could only measure on (I had the feeling) process scheduler level and with a resolution over 100 microseconds. Which is simply factor 100 worse.

This is one of the things why I personally tend to stay still on Win7 for a recording system, as I have the experience to differntiate, whats ok and whats not ok.

And there are other things which make me think about whether Windows 10 is already mature enough, like I put together in a blog article 9 month ago:
https://www.tonstudio-forum.de/blog/ind … s-7-EN-DE/

At a certain point also I will have to change to Win10, but I want to move this point as long as possible to the future.

BTW .. last time I bootet my test system I had under Win10 of all sudden issues with audio loss, which was not the case before. This is still not solved. I am not 100% sure whether it came with the latest patches / updates to 1809.
Therefore I will have to restore my golden image of 1804 (or 1803) and look whether it is still ok there and retry the upgrade. The audio loss I got when simply playing back FLAC file via MusicBee using RME ASIO driver, usually no issue.

Which brings me to another point ? I would propose that you buy a 2nd SSD/DISK to perform a Win7 parallel installation and use Win7 for recording purposes, shall your mainboard still has Win7 drivers.

Otherwise I can still recommend my mainboard and CPU (use the v4 it's better in terms of producing less heat), because then you can still run Win7 and its proven. The bad thing today is, that with newer CPUs Intel tells, no Windows 7 support anymore and this draconic decision I personally regard as very bad for customers esp. for us in the recording area, as I see still a certain demand being able to deploy on Win7 simply to have a peace in mind as the deployment cycles of Win10 seem to be too tough even for Microsoft, you can see the decreasing quality definitively here:
https://www.heise.de/newsticker/meldung … 82651.html

To sum up, possible solutions
- BIOS WIN tuning
- Win7 parallel installation and comparing with Win10 on same HW
- Trying to make use out of LatencyMon for the measurements (for Win7 I recommend to take the old version 4.02 to ensure, that you get really compareable values, as I am not 100% sure, what changes they made in the tool for Win8/10 compatibility), you can get the older version from my dropbox: https://www.dropbox.com/s/imcbu6lcac5vg … 2.exe?dl=1
MD5 checksum here: https://www.dropbox.com/s/9h6uoqgcaqqvd … e.md5?dl=1

Good luck!

The E5-1640v4 is very nice.

This Broadwell-EP CPU runs approx 8+°C cooler because of 14nm,
compared to the 22nm with the v3 version (Haswell-EP).

- up to DDR4 2400 (v3: 2133)
- higher passmark performance (higher CPU clock)
- when executing  AVX/AVX2 instructions, then the v4 CPU falls down to the same clock like the v3 CPU (3.5 GHz)
- otherwise you get a 200 MHz higher base clock for all other applications / corews where AVX/AVX2 instructions do not run

Cubase uses AVX2 commands. So when Cubase runs full power on all cores, then you will see with
HWMonitor, that most if not all cores run then at 3.5 GHz (v3+v4 CPU).