I am new to the forum and the RME family.
I need a new PC for my also new FIREFACE UFX+.
I would like to go with windows. Are there any recommendations?

I am considering buying a refurbished HP Z820 Workstation Intel Xeon 16 Core 2.6GHz 128GB RAM 500GB Solid State Drive + 2TB Hard Drive Dual NVIDIA Quadro FX 3800 Graphics.

My questions is in regards to

How many Cores and RAM?
Is 2.6 GHz enough?
Are two hard drives enough? What are the advantages of a third one?

I also have the possibility to build my own.

I am mainly working ITB (at some point want to add Hardware, Mixers…but that is depending only budget and might take up to two years), don’t have hundreds of VST but I need a robust windows system that can handle some load.

Thanks
Pete

A recording system doesn't need that much CPU power, more important is, that the system is agile (less DPC latency) and that CPU cores are not blocked by e.g. bad drivers or mainboard.

Also important is to find out, whether your HW needs some additional tweaking. I had it more difficult, because I wanted to have a general purpose PC for everything (internet, multimedia, gaming, recording). The more devices you have, the more drivers you have installed and active on your system. Then it is more likely to have a bad driver blocking a CPU core for too long. If there is an audio related task scheduled to be executed on a CPU Core where driver code is exclusively running (and can't be interrupted by the task scheduler of the OS for data integrity reasons), then you can get audio drops.

Especially graphic cards for gaming can cause issues. It took me around 2y to find out the root cause, why I still have audio outages from time to time, although my system is not under load, well optimized and passes the LatencyMon test for approx 15-30 minutes (add. note: maybe I should have let it run for several hours). Additionally it ran very well before with Windows 7. What changed were more or less two things, the CPU (little upgrade from E5-1650v3 to v4), little faster ECC DRAM and the graphic card (GTX 980 -> RTX 2070 Super).

The issue was energy saving in the nVidia RTX2070 Super. nVidia itself doesn't provide any software anymore to turn off energy saving in their graphic cards. Simply by occasion I found an interesting article on Steinberg forum, which pointed me to a 3rd party software from a russian developer named powermizer switch, which you can use to entirely turn of power saving in the GPU. After turning off power-saving for the video card as well, all my problems were gone. Since then my system is rock solid again (like with the initial HW and under Win7), but it took me much time and efforts.

For such reasons you have still companies selling turnkey systems for audio and video systems which simply work. All the testing - which can be very time consuming - has already been performed.

If the company is good, then they can also give you recommendations which type of CPU is ideal for your use cases. A rule of thumb is, that CPUs with a high single thread performance will give you advantages for any kind of application. Especially when working with a lot of VST in one track or CPU hungry VSTi. Then everything of one track usually runs on one CPU core and then it's better, to have a CPU with high single thread performance to be able to process everything on that one core quicker and in time not to get audio drops when using smaller ASIO buffer sizes (to still have lower RTL).

I heard that some companies even get special BIOSes from e.g. Supermicro, but I can't tell you by heart what special purpose this has.

I am using still a 8y old supermicro mainboard where I added or replaced some components as I needed, which is really one of the strength of a PC system. The current system you can see here in my blog: https://www.tonstudio-forum.de/blog/ind … mponenten/
Although the mainboard is around 8y old I still can run games like cyberpunk 2077 on Full HD with a good performance.
I was lucky that the combination of E5-1650v4 and RTX2070 Super works very well, because a too old or lame CPU can have impact on the performance of the GPU.

This 8y old system with Xeon E5-1650v4 CPUs is still powerful enough to play an artificial Cubase project which I use for benchmarking my system to find out whether all works well (if I do not get audio drops during playback) with
- 400 tracks and
- 800 Steinberg VSTs
at sample rates up to 96 kHz and by using the smallest possible ASIO buffersize with my UFX+ via USB3.

See this blog article: https://www.tonstudio-forum.de/blog/ind … cks-de-en/

Why are low DPC so cruicial for a good working system ?

All of our systems (no matter whether Windows or Apple) are not real-time computer systems. Unlike normal programs / applications / Services - which can be priorized and managed by the task scheduler - driver code (low level routines) need to run as long as coded in the driver itself for data integrity reasons. Some drivers simply rob/steal too much time and can block audio related tasks with near-realtime demands for too long.
I am pretty sure that some driver developers might let their drivers run a little bit longer, so that benchmark values look better than the competition. But if they run for too long, then this blocking of CPUs is counter productive for all other processes, especially audio/video related processes with real-time requirements.

This makes it more likely that an audio related job is not being finalized in time and then you get audio drops. The task scheduler can not prevent this, as it can not know, how long the low level routine will run and reorganize job distribution across different CPUs.

From audio application / DAW perspective you can only mitigate this by e.g. using bigger ASIO buffer sizes. Then the transfer of audio data can happen in larger chunks of data, so the system can work more efficient on audio. By this also the amount of interrupts and context switches on a system decreases. Also CPU load decreases. But you will get a higher Latency (RTL, round trip latency through USB/FW/TB) when increasing the ASIO buffer size which you can only do, if you do not need to monitor/listen with near-realtime performance. So to say when recording, but not when playing through VSTi's.

When working with VSTi then the RTL (round trip latency) should be below 10ms.
This you can only achieve by using ASIO buffersizes <= 128 samples at single speed.
In some cases also 256 samples will work, because when playing with a master keyboards, you save one A/D conversion.

So if you want to use VSTi, then you need to ensure, that your audio projects run with an ASIO buffersize of 128 (max 256) at single speed (44.1/48) to stay under 10ms RTL. Or you need to use some tricks like freezing of tracks to reduce the realtime load.

If you work with VSTi, then I would use a CPU with not only a lot of cores, but with a high single thread performance.
Usually this you can achieve if
- the CPU cores run very efficient (from design)
- the CPU cores run at high clock speed
The more CPU cores in a CPU, the lower the clock speed.
So you need a good tradeoff between number of CPU cores and clock / single thread performance.

You also need to know how your DAW works.
Does it have limitations with a certain thread like Logic has (or had). There is always one thread under fire and of the one CPU core is not fast enougn, then you have a limitation there.
Other DAWs can only utilize up to 12 or 16 Cores.
Latest changes in Windows resulted in a certain limitation of the max number of threads that can be proirized by using MMCSS. With Windows 7 the number of threads were nearly unlimited, at least 65.000 still worked.
With Windows 10 the number of threads was below 16 (I do not remember anymore).
For that reason Cubase needed a different coding/approach and for a certain time it was even required to apply a registry change which was (funny funny) under non-disclosure and Stainberg had to ensure that only customers where it had been validated that they have a certain CPU with at minimum ~16 core, that they get this registry fix.

Usually this is all not a big problem, if you have the luck, that you picked luckily proper components.

But if you base a business on these components for mixing and mastering, then I would better use such a tested turnkey system from a company that you can trust in that area.
Some of these companies work with consumer HW, some of them use Enterprise HW and Xeon CPUs up to multi-core systems. I would keep it simple and use a Single CPU system, that is powerful enough not to get into issues with Multi-CPU systems, if the application or your use cases do not benefit from a Multi-CPU design.

As you work ITB you might need a high Single thread performance. How high depends on the number of tracks, VST that you use and how complex / CPU hungry they are.

Nowadays with quick SSDs its not really required to have muliple SSDs. Its not like with harddisks anymore where you have long access times because of the head positioning. Also the throughput is much higher.

I would use a number of SSDs and HDs for operational reasons:
1. to be able to separate the operating system with applications and registry settings from user data.
OS, Apps, Registry forms a unit, you better backup and restore this to have the same timestamp.
And if you need to restore the OS, you do not want to loose your data.
So its better to have a separate SSD for OS and one for user data.
2. to have internal and external backup. If something bad happens in your PC you have a good chance to have the 2nd backup still in good shape if it is fully decoupled from your PC, e.g. is you backup over network using a NAS.

Although still quite expensive, I would use 10 Gigabit NICs to reduce the transfer time over LAN.
But if you need a cheap solution:
a) one internal 10GB disk, backup over Sata is still quite nice and usually faster than over 1 Gbit LAN.
b) one external backup using USB3. I would look for USB adapter and external cases supporting UASP as they work more efficiend when you backup many small files.

I think it would be a good option for you to cross check, what a turnkey system for audio would cost.
If you get a good company, then this is absolutely hassle free and you get most performance out of your RME gear.

It might be more beneficial for you to get a turnkey system with components of today (not an old system) because the competition between Intel and AMD lead to more performance increases (sadly by sacrificing energy efficiency in many cases).

Regarding Windows 11: important ... keep in mind, that Windows 10 will be ceased in 2y, then you need to switch to Windows 11. Windows 11 needs more CPU performance because of their new internal security design. This is the reason why older CPUs are not supported anymore, they fear the power is not sufficient anymore (IMHO crap design, normal consumers do not have such a high security demands like companies or data center, this doesn't justifiy having to decommission good working only 5 or 6y old computers.

And if Microsoft will block the registry tricks, which still enable a Win11 installation without these CPU hungry security features, then your system needs to be more performant. Why I wouldn't recommend to buy old HW which could turn out to be too weak for processing audio on top of this performance sucking security design..

So better invest into a tested turnkey system which supports Windows 11 and gives you enough performance reserves.

Thanks Steve, I rephrased it here and there a little. Thanks for the flowers and good luck with your tuning.
One reminder, although it's nice when LatencyMon reports "your system is well suited for audio processing" ...

There is still a big difference whether your measured performance values are under 50 with rare / small spikes up to perhaps 100 microseconds or wherther your machine runs very often quite below the 1000 microseconds threshold, where LatencyMon will report, that your system is not well suited for processing audio.

The lesser the latency is, the more likely it is for you, that you do not get audio loss even when using small ASIO buffer sizes.
Along with smaller ASIO buffersizes you get smaller chunks of data, that you can transfer in one sweep.
This means the load on a system increases (CPU load, interrupt load, amount of context switches) because the CPU has to get and put "less data" "more often"  from/to I/O ports and application. This is only possible, if you have an agile system and if the DPC latency is only very low on your system.
At a certain point the system will be unable to be quick enough to react on interrupts and to process the data, then you get data loss, which can be mitigated by using bigger ASIO buffer sizes, but at another point even this might not be sufficient anymore, especially on a loaded system where all the tasks run concurrently.

And do not mix the internal DPC latency with the RTL (round trip latency) for
- converter latency for A/D
- time for transfer over USB
- processing in the application/daw
- time for transfer over USB
- converter latency for D/A

This is fix and depends only on the
- efficiency of the ASIO driver
- how low the converter latency is
and what sample rate and ASIO buffer size you choosed for the project.

The DPC latency is the measure for agility of your system, how well it is possible to transport audio without loss even under load ... If audio loss happens you need then to either increase the ASIO buffersize or to lower the system load, by e.g. freezing tracks for VSTi in the DAW ...

I also suggest looking at what audio pc builders are using as these test with latencyMon. In the uk a good one is SCAN. I’m wary about chucking a load of parts together now as I see so many people with problems. I used to build my own but some of these makers are not a great deal more than the parts and you get a guarantee as well.

Surely with any modern Windows build there’s not a huge amount of room for real-time audio becoming a problem - components these days are becoming more reliable and quicker. However, one of the main reasons I went for RME is because they’ve got solid drivers with proper error checking etc, so I know when a system upgrade is due it’s just going to work.

Getting ultra low latency can be especially taxing, but these devices are very low latency regardless. I’ve accepted that computers are not perfect, especially since I’ve been working with them for so long

Wow! Thanks ramses for your great post! And thanks to everyone else who contributed in that thread! I truly appreciate it!

The topic is a bit overwhelming for me…can someone provide me with a good company that will do customization (and won’t break the bank)?

Than I can sort out all the information into a practical approach.

Sorry for my late answer, but I couldn’t find my post anymore…it seems I posted in the wrong sub forum. Sorry for that!

I mainly use CUBASE 11 respectively CUBASE 12 as my main DAW. And typically I using around 50 - 100 VI‘s at a time. In addition I use WAVELAB for mastering. But not on a professional level, meaning I don’t make my living from music production.

Greetings