Very specific NUC needed

I am going to mod a NUC. Like insanely mod it..

But where to start.. There are so many and i have some fairly unique requirements.

I want absolute minimum number of systems. I want just the exact systems needed for a very stripped down audio server.

I believe less parts on the board is better. So I dont want ANY extra stuff.

1 x 2.5 ( or 10 gbps ) Gbps ethernet OR 1 SFP+ port. I DO NOT want 2 ports.

1 x SATA port

1 x PCIe M.2

I would like dual or more channel RAM slots to 32GB

1 x USB I dont need USB 3, would prefer 2. Need driver support and maybe a known good chip for audio use

VERY minimal graphics, very minimal HDMI. Command line 640x480 is all that is needed.

CPU socket - intel. very lowest CPU needed for audio roon core / UPnP use.

Ideally, all the above with a high end board. This is not about cost savings, its about having the least amount of electronics on the board.

I want to look over the board, so I need pics. As I intend on doing a LOT of mods to it I need to see it. A lot of NUCs its hard to find board level pics. Besides clock chips I also need to look at layout and I need to look at the PWM regulators for a number of reasons.

For OS.. It needs to run some flavor of Linux, it needs to be able to run roon rock and maybe even FreeBSD. I want to be able to do it for roon or for UPnP use. I am also interested in AES67/Dante/Ravenna and so I need to look at any NIC issues or CPU requirements. It would be good if it can do PTP. I think tho I need a serious high powered rig for AES67/Dante/Ravenna use.

You all look to be experts at all this and maybe you can suggest a solution..

Short of all that, maybe you can suggest something close to this outline that is a known good NUC for this use.

21 minutes ago, Xymox said:

I am also interested in AES67/Dante/Ravenna and so I need to look at any NIC issues or CPU requirements. It would be good if it can do PTP. I think tho I need a serious high powered rig for AES67/Dante/Ravenna use.

I have years of experience with this piece you are interested in. 

I've had issues with 10GbE and Ravenna because while the cards say they support PTP, the driver doesn't. You'll have to really research both the NIC and the driver before settling on hardware and an OS. 

Ravenna / AES67 don't require that much horsepower. As you increase sample rate and channel count, in addition to using DSP like room correction, then some horsepower is needed. 

Ravenna on Linux certainly works, but I hope you're a savvy Linux user interested in fighting with the installation and troubleshooting. 

I would focus on a 1 GbE NIC if you want Ravenna. 

Thank you. Wow Revenna experence as the first post :) from the founder.  This is a savvy set of people here :)

My 30 years of *nix experience is, oddly, on FreeBSD. But pretty much the same idea. Its a *nix. Yes ive done all manner of brain vexing things on these OSs. Yes I have spent a week, full time, on single issue before. I still prefer it to WIndows tho.

The driver support for the nic / chip is indeed key. The issue is, man, there are so many moving parts to nail down. Its like a rubics cube. I also would need the right driver for a USB chip that is best for audio use. ALong with everything else.

Hmmm... Overall I dont think I want a music server doing DSP. I think that is best left to external hardware made for that use. IMHO..

Why I *think* I want to limit CPU horsepower is I know that a higher power CPU socket will end up with higher power PWM regulators. These devices spew wideband noise from a few Khz to low Mhz. The lower the power the regulator is, the lower the radiated noise will be. Its pretty much physics with these. Its also easier to fully filter a lower current PWM reg then it is a high current design. Its also easier to shield these parts of the overall output in radiated RF is lower. SO I would like to stick to a socket / CPU that has just enough horsepower for its intended use. But this is hard to figure out without real testing.

I think the lowest noise on all the busses and chip outputs is best achieved by a minimal part count and lowest power to achieve the desired results. Phase noise and buss jitter have a lot of noise that comes from these power supplies and really poor ( if any ) matching of regulator and chip impedance. This poor matching results in the chip outputs on busses being sloppy and noisy. Also each trace on the board acts as a antenna picking up, and transmitting, noise. This ends up as jitter on everything.

I have looked at a LOT of servers from the Taiko to Melco to just all sorts of servers and I just dont see good design. So I have decided to do it myself.

I guess I am not fixated on a NUC. The small size is nice tho as short PCB traces can be good if the rest of the board is well laid out.

On top of everything, I need bulletproof reliability, hahaha..

OMG.. This all seems impossible. I should probably just grab some random device and mod the **** out of it and see what it sounds like.

Oh, side note.. I will be driving it with a R-core based linear that does remote sensing and noise cancelation. Most likely I will have to create a new higher current version of my current supply anyway, so, I can move into other voltages. BUT. I need it to be 1 voltage rather then a ATX set of voltages.

36 minutes ago, Xymox said:

Thank you. Wow Revenna experence as the first post :) from the founder.  This is a savvy set of people here :)

 

My 30 years of *nix experience is, oddly, on FreeBSD. But pretty much the same idea. Its a *nix. Yes ive done all manner of brain vexing things on these OSs. Yes I have spent a week, full time, on single issue before. I still prefer it to WIndows tho.

 

The driver support for the nic / chip is indeed key. The issue is, man, there are so many moving parts to nail down. Its like a rubics cube. I also would need the right driver for a USB chip that is best for audio use. ALong with everything else.

 

Hmmm... Overall I dont think I want a music server doing DSP. I think that is best left to external hardware made for that use. IMHO..

 

Why I *think* I want to limit CPU horsepower is I know that a higher power CPU socket will end up with higher power PWM regulators. These devices spew wideband noise from a few Khz to low Mhz. The lower the power the regulator is, the lower the radiated noise will be. Its pretty much physics with these. Its also easier to fully filter a lower current PWM reg then it is a high current design. Its also easier to shield these parts of the overall output in radiated RF is lower. SO I would like to stick to a socket / CPU that has just enough horsepower for its intended use. But this is hard to figure out without real testing.

 

I think the lowest noise on all the busses and chip outputs is best achieved by a minimal part count and lowest power to achieve the desired results. Phase noise and buss jitter have a lot of noise that comes from these power supplies and really poor ( if any ) matching of regulator and chip impedance. This poor matching results in the chip outputs on busses being sloppy and noisy. Also each trace on the board acts as a antenna picking up, and transmitting, noise. This ends up as jitter on everything.

 

I have looked at a LOT of servers from the Taiko to Melco to just all sorts of servers and I just dont see good design. So I have decided to do it myself.

 

I guess I am not fixated on a NUC. The small size is nice tho as short PCB traces can be good if the rest of the board is well laid out.

 

On top of everything, I need bulletproof reliability, hahaha..

 

OMG.. This all seems impossible. I should probably just grab some random device and mod the **** out of it and see what it sounds like.

 

Oh, side note.. I will be driving it with a R-core based linear that does remote sensing and noise cancelation. Most likely I will have to create a new higher current version of my current supply anyway, so, I can move into other voltages. BUT. I need it to be 1 voltage rather then a ATX set of voltages.

 

 

Now you’re bring back some memories. One of my early music server builds was FreeBSD on an Alix board :~)

https://www.pcengines.ch/alix.htm

Cool !...

I think the *nix you start with is just sorta home.. Linux to me seems sorta, well, not as good, hahaha.. I am biased I suppose. .

I still run a FreeBSD server and it runs Unbound ( DNS ) and does rDNS on my local network. I always run a rDNS server. Unbound even does prefetch and I do DNS over TLS ( DoT ).. So much faster DNS and so much more secure then using a ISP or even 1.1.1.2

Sadly, I dont think its possible to run what i need to tho in FreeBSD.

11 minutes ago, Xymox said:

Sadly, I dont think its possible to run what i need to tho in FreeBSD.

Agreed :~(

This will be fun to watch happen!  Kind of exciting.  

FYI Intel is basically out of the NUC business these days.