Building a DIY Music Server

[Nenon]

6 minutes ago, abase said:

..sorry, I did not want to say that your system is not working -  I have a diy system and it works also fine. But without cooling the card overheads in 2 to 3 minutes - I did not expect that for a network card...

 

Those network cards are designed for servers where you have massive airflow from one side of the server to another. These servers have so many fans that they may sound like an airplane when you start them.

 

Servers are racked in cabinets and on the hot side of the cabinet (where the hot air blows out) you have massive A/C units, usually blowing from the floor, that cool down the hot air.

 

As we install between 2 and 6 of these per server, I've had thousands and thousands of these Solarflare cards going through my office. I've tried them on several occasions on my server and always had the same experience - my system sounds good with them in the first 20ish minutes and then it sounds worse than the onboard NIC. I have always attributed this to the higher temperature.

 

So I would really advise considering using these only if you have a proper cooling solution.

 

And even then, I wonder if the Intel I210-IS would end up being a better option. It should not be too hard to try cutting the PCIe power lanes and modding it for external power. If I were still doing DIY servers, I would be all over that and the copper plate @SK8 made.

[Exocer]

@SK8I was going to embark upon a journey to design my own copper plate for the i210 since I’m convinced this is what I will go with long term. But I figure I’d ask if you have any spares or a means to get more produced for the diy community. Do you have any spares? Cheers 😎

[Johnseye]

51 minutes ago, Exocer said:

@SK8I was going to embark upon a journey to design my own copper plate for the i210 since I’m convinced this is what I will go with long term. But I figure I’d ask if you have any spares or a means to get more produced for the diy community. Do you have any spares? Cheers 😎

 

+1

Interested as well

[MarcelNL]

that X2522 as well as the 85something Solarflare cards are designed for use in servers in a serverroom with forced air flow at around 5'C 

 

The heat dissipated by the chip is a measely 10 Watt or less but it adds up and without forced air OR a proper passive cooling solution it'll overheat and shut down.

 

 

A small PC fan on 9 Volts is enough to keep it cool, passive cooling is of course much better (work in progress for me)

[StreamFidelity]

3 hours ago, abase said:

But without cooling the card overheads in 2 to 3 minutes - I did not expect that for a network card...

Enclosure fans are prescribed by the manufacturer. Your passive cooling system looks very massive. I wouldn't expect any problems. Is there enough thermal paste between the chip and the cooling plate? 

Taiko Audio also tested Solarflare NICs in the early days. As long as there was no overheating, the sound was very good. As they did not want to use passive cooling, it was not used. 

For me, the Solarflare X2522 is the best network card you can have in the consumer sector. Extremely low latency and jitter values and a very powerful chip (hence the heat development) make all the difference. 

[abase]

1 hour ago, StreamFidelity said:

Is there enough thermal paste between the chip and the cooling plate? 

.yes it is - so I must thank you for spend so many information on your website. Today I followed your guide to update the firmware...

 

Today the card is running over 10 hours ;-). All my single mode transeivers are not working with the card. At the moment I`m using a multimode tranceiver..
I think that is the reason why I`m not satisfy with the sound...

[Exocer]

8 minutes ago, abase said:

.yes it is - so I must thank you for spend so many information on your website. Today I followed your guide to update the firmware...

 

 

 

Today the card is running over 10 hours ;-). All my single mode transeivers are not working with the card. At the moment I`m using a multimode tranceiver..
I think that is the reason why I`m not satisfy with the sound...

Could be! I always strongly preferred the sound of single mode transceivers.

[Johnseye]

Do you hear a difference in SQ with different NICs (ie. Solarflare, JCAT, onboard, etc) installed when playing music from a local drive? 

 

I have 4 different NICs and have been spending time comparing with both local and streamed sources. 

[StreamFidelity]

20 hours ago, abase said:

All my single mode transeivers are not working with the card. At the moment I`m using a multimode tranceiver..

 

I use singlemode fibers without any problems. But there are differences in quality. You need very good qualitiy transceivers. Some from Startech and Finisar are good. Take a look:

And they do exist: Differences in the quality of LWL (fiber optic cables) – Grigg Audio Solutions

[JackJohnson316]

Which Finisar SFP+ RJ45 Module does the Taiko Extreme use on their network card? Thanks

[SK8]

7 hours ago, JackJohnson316 said:

Which Finisar SFP+ RJ45 Module does the Taiko Extreme use on their network card? Thanks

I bought this one:

 

https://www.amazon.com/gp/product/B00BTWEE5K/ref=ppx_yo_dt_b_asin_title_o00_s00?ie=UTF8&psc=1

 

Have a look on the post #13294 of the Taiko Audio SGM Extreme : the Crème de la Crème thread in the whatsbestforum. A lot of good info there. It seems that the best is to use the module with the SFP+ Switch port, if you have one. 

 

 

[Dev]

I am completely detached from audio these days (as I peruse other projects/hobbies) but I have been using the Startech i210 single port sfp for over a year now with a Cisco AOC cable and am quiet content with the way it sounds. The server is connected to a custom modded Buffalo switch. Previously I was using Jcat Net femto but I always wanted to try a fiber nic which lead me to buy several other vendors - 10gtek and some others from ebay before I bought the Startech. I also have i350 based dual sfp nic but I did not have time to experimented much with it. Both the Startek & 10gtek sounded nice in my system but I marginally preferred the Startech. During my experiments with 10gtek, I have spoken to them at lengths about making a custom one with external +12v power (instead of feeding through pcie) and a better clock but they are only willing to do for large quantity order. If there is enough interest, maybe we can get them to build a custom one. Anyway, I am glad that folks are finding the Startech nic musical as I do.

 

 

 

 

[NewOldman]

has anyone compared Intel x710 with Solarflare Nic

[seeteeyou]

Some audiophiles here in Asia would add those cheap PCIe extenders between the PCIe slot and plain old Gigabit Ethernet PCIe card for the sake of providing isolation. Though they don't need the bandwidth for anything serious and therefore even PCIe x1 would work fine.

 

Of course it's a different story for Solarflare

 

https://www.xilinx.com/content/dam/xilinx/publications/product-briefs/xtremescale-x2522-product-brief.pdf#page=2

Quote

PCIe Gen 3.1x8

 

These extenders aren't meant to work over a long distance at all, therefore the Solarflare NIC would still stay pretty darn close to the chassis and then all that heat might still affect the overall temperature quite a bit

 

https://www.thermaltake.com/tt-premium-pci-e-3-0-extender-300mm.html

https://www.thermaltake.com/tt-premium-pci-e-4-0-extender-300mm.html

https://www.thermaltake.com/tt-premium-pci-e-4-0-extender-200mm-with-90-degree-adapter.html

 

Then we've got something else that's good for a much longer distance, though the entire package would cost "just a bit" too much

 

https://adnacom.com/s31/

https://adnacom.com/product/s31-order/

 

Recently I read a little something about this

 

https://liliputing.com/lenovo-could-ship-first-mainstream-laptops-with-an-oculink-port-in-2024-leaks/

 

I figured a new way to run an eGPU off M.2 using OCulink

https://egpu.io/forums/custom-egpu-chassis/i-figured-a-new-way-to-run-an-egpu-off-m-2/

Quote

I found this out when researching M.2 to 10GbE which is doable with an M.2 to OCuLink adapter and then the HP 1QL49AA.

 

OCuLink: What It Is, Why You Need It, and Where To Get It

https://egpu.io/forums/custom-egpu-chassis/oculink-what-it-is-why-you-need-it-and-where-to-get-it/#post-104553

Quote

Cable length doesn't matter much for degradation until 1.5m, where proper cable offerings get very stiff to keep signal quality, so order to your needs but know excess isn't as coilable as a TB cable, so try to get what you need, not over.

 

So HP 1QL49AA should be this particular NIC

 

https://www.hp.com/ie-en/products/accessories/product-details/18809667

 

If that were good enough for eGPU already, could we also try turning Solarflare NIC into an eNIC then?

 

https://www.aliexpress.com/item/1005005277989277.html

https://www.microsatacables.com/oculink-8i-dual-port-to-pcie-x16-slot-adapter-pcie-4-0

https://www.microsatacables.com/pcie-gen3-8-lane-to-oculink-sff-8612-8i-add-in-cardaic-ocu-1615-8l

[abase]

..in my previous post I mentioned, that I was very unhappy with the sound auf my new server and the network card.

It took four days until I found the cause..
To make room for my new server I placed my Amp on another board – it is a bit heavy and maybe I was not gently enough for this thing ;-)  so the fuse gets a tiny, tiny bit out auf place.

I have tried everything but I did not even guess that there was something wrong with my AMP.
It was working flawlessly – but the soundstage was completely gone.
I was pissed off, that my new hardware could not reach my old Intel Nuc.

One afternoon I decided to put my Susvara in the headphone amp (it is normally connected directly to the power amp) and the soundstage was there…
From that moment I knew that there was something wrong with my AMP. After putting the fuses correct in place it all works fine. And my smile was back to my face…
That is an evident that every tiny bit matters to get the right sound..

[seeteeyou]

 

 

https://t1p.de/gzx75

https://twitter.com/TheBobPony/status/1776012502695923916

https://www.deskmodder.de/blog/2024/04/04/windows-11-24h2-26100-1-ltsc-iso-durchgesickert/

https://dl.bobpony.com/windows/beta/11/26100.1.240331-1435.ge_release_CLIENT_ENTERPRISES_OEM_x64FRE_en-us.iso

https://archive.org/download/26100-ltsc-x64-enus/26100.1.240331-1435.ge_release_CLIENT_ENTERPRISES_OEM_x64FRE_en-us.iso

[AngeloVRA]

Building a DIY chassis for my DIY Music Server

 

The chassis is an often overlooked component of DIY music servers and LPS. The big boys Extreme and Pink Faun have substantial, stiff, fit for purpose chassis..... that look beautiful too!  And there are no similar commercially available chassis.

 

Having been unsatisfied with currently available PC chassis from Streacom, HDPlex and even the Taiko DIY chassis, a few months ago I decided I would design and  3D model my own and go ahead and have it CNC manufactured.

 

This DIY post is intended to share the design thought process (not perfect), various technical issues, and solutions when making my own DIY chassis.

 

 

Where do I begin?

Firstly, I realized from previous projects that you would need to provide detailed STEP files and drawings  for the CNC facility to make your parts. This necessitates using some kind of 3D CAD/CAM design modelling software. After a bit of time comparing various options, I settled on Autodesk Fusion360.

If you are already proficient in 3D CAD/modelling software app, you’re in luck, else be prepared for a learning curve which is enjoyable though requiring what I’d call “more than a bit of time and effort”.

2D drafting apps such as Front Panel Designer are much easier to learn but are limited to one plane. For example, it is sufficient to design a rear panel until you decide you needed to threaded screw holes for the top of PCIE brackets which is on another plane.

2D drafting apps also cannot simulate assembly of the  various panels that will make up the chassis so you won’t be able to see how they fit together. Nor will you be able to effectively communicate to the CNC facility how the different parts are supposed to fit together. You will end up reordering the same part with various revisions/corrections which will end up costing you “more than just a bit”.

 

 

In my own journey, I designed the chassis in the following order:

-        Rear Panel

-       Top Panel

-       Front Panel

-       Heatsink

-       Bottom Panel (I dreaded measuring for the various motherboard standoff locations) 

 

I limited my chassis use case to :

-       SoTM smb-Q370 motherboard

-       2x stacked SoTM SCLK-EX boards

-       Taiko ATX DC-DC converter

-       34v Regulated DC power supply to the Taiko ATX

-       12v Regulated DC supply to the CPU

 

Rather than starting with a blank slate and researching the various standards for PC cases, I used my existing HDPlex H5 chasssis as a starting point.

 

 

My priorities were :

1.     The chassis must be stiff and sturdy, not requiring further bracing and reinforcement

2.     It must be fit for purpose. It must address EMI shielding and no more additional extensive drilling and thread tapping

3.     Aesthetically nice.

 

Rear Panel

 

 

Above: the rear panel of an HDPlex H5 V3

Design points:

1. The existing panel is flimsy and thin. My previous project which entailed simply replacing it with a 10mm thick panel proved beneficial to SQ, adding solidity and more precise imaging among others. The top of a PCIE bracket is 12mm long, so the rear panel will have to be at least 12mm thick.
2. The area above the PCIE brackets is a wide open space through which EMI can leak in/out. Covering that area would require another 3mm. Total thickness for my rear panel design is thus set at 15mm
3. Neither the bottom panel nor the top panel is screwed onto the rear panel which in my opinion compromises its stiffness. The area underneath the I/O panel is so thin (only 4.6mm as shown in pic)  and further compromises its structural rigidity. I decided that I would add 3mm thickness to that area. This meant that
   • I would have to raise the MB another 3mm. Existing MB standoffs are 7mm.  simple change to 10mm standoffs would accomplish the task. This improves cooling of the bottom of the MB.
   • In addition, I would also have to raise the PCIE slot openings by the same amount 3mm
4. The area on top of the PCIE slots where the PCI brackets are supposed to be screwed into is relatively thin at xx mm. Drilling threaded holes into it would puncture the bottom part which I thought would visible detract from its aesthetics. I decided I would beef it up and make it to thicker at yy mm by reducing the height of the PCIE slot opening by the same amount.
5. The bottom of PCIE brackets are supported from behind by the rear edge of the motherboard except those additional PCIE brackets to the right. Such additional PCIE brackets are usually used to hold connectors for clock inputs and DC power in for various cards, etc. Inserting the plugs usually cause the floppy PCIE brackets to flex inwards and doesn’t give one a solid feeling feedback. If the top of the brackets are secured by a screw, not a big deal. But in some cases e.g., with a Taiko DIY chassis, the bracket is not secured with a screw and the entire bracket can just come loose and fall back inside when one tries to plug in a clock cable or a DC cable.
   • In addition to threaded holes for the top of the PCIE brackets, I added a locking bar to secure the bottom of PCIE brackets
   •  I added 2 of my favored Jaeger chassis connectors for the 34v and 12v DC power supply input.
6. In the HDPlex H5, the thin rear panel is affixed to the heatsinks with externally visible bolts which I didn’t like. I opted to use internal “embedded” L brackets to affix the rear panel to heatsink. 

   • 

   •  L brackets would be unsightly if simply placed on top of the panels. Embedding them into the metal would be more visually appealing and helps torsional and shear stiffness of the joint.

   • Structurally, 1 screw per leg would be enough but just to be sure I would put 2 per leg. The thinner panel, the heatsink would be 10mm thick. Embedding the 3mm L-bracket leaves 7mm which nets me 5-6 thread depth for the screws. This is about 1.5x the diameter of an M4 screw which is perfect for the job.

   • 2x M4 screws with 8mm countersunk holes mean that the L- bracket had to be at least 20mm wide.

   • 

   •  

7. Datasheet for the Jaeger connectors show that the connector hole is 22 mm and 4 holes for M3 screws arranged in a 28 mm circle
   • 

8. The existing PCIE slot openings were only 13.2 mm wide which prior experience adding BNC chassis connectors to PCIE brackets show was just a bit too narrow. Thus this was expanded to 15.5mm which is also the space needed to be able to insert a Telegartner RJ-45 connector. Standard PCIE brackets are 18.5mm wide so we will have 1.5 mm latitude on each side.

9.  

   For stiffness, the top panel will be screwed onto the rear panel at 4 points, 1 at each end and another 2 over the IO panel. The bottom panel will similarly be screwed into the rear panel at 4 points, albeit asymmetrically, one at each end and 2 under the PCIE slot area.

10. Aesthetic choice was to fillet (round) the left and right edges of the rear panel at a diameter of 10mm. Similarly the I/O panel cutout was filleted at 2mm and the PCI slot dividers at 1mm

Top Panel

 

For the top panel, I simply followed Taiko’s lead with the Extreme that the vent holes would actually be tubes if the panel was thick enough... Ergo! EMI waveguides. This reduces EMI leakage as only the “straight” waves get through at full force. The waves that hit the walls of the tube are greatly diminished after they have bounced off the walls of the tube several times. This theory appealed to me.

Problem was:

-       How to draw the holes in a concentric circle pattern

-       With all the holes the same distance from each other

This required a bit of thought. It is not the only method but here is what I did (short version)

-       Assume we want 3mm holes with center spacing of 6mm

-       Draw concentric circles with each circle 6mm larger in radius than the previous.

-       For each concentric circle. Knowing the radius, you can calculate the circumference. Divide that by 6mm and you will get the number of holes you need to place around that circle to get close to 6mm spacing.

-       In my case, it was 6 holes for the 1st circle, then 12, 19, 25,31. 37.44 and so on. The top panel has over 2,500 holes.

 

Tedious, but here’s the result... 10mm Top Panel

 

The left and right edge areas will be covering the top of the heatsinks. And we need to open up that area so the heat can be expelled upwards from the heatsink fins.

My aesthetic choice for this is a set of triangular holes that form a truss-like structure. Easier choice would be circular or rectangular vent openings.

 

For the top panel. I decided to use counterbore M4 screws which would yield a stronger bond than when using countersunk screws. For aesthetics, there are no screws over the top of the front panel.

 

HEATSINKS

 

There are 4 design parameters to consider with heatsinks:

1.  The length of the heatsink front to back. My chassis design objective to have internal depth of 340mm. This shall be the same as the length of the heatsink

2. The thickness of the base of the heatsink determines its ability to spread the heat to other fins. The HDplex heat pipes are 6mm which will require 5.6mm deep grooves which will leave 4.2mm cover if we use 10mm as our base thickness

 

 

3.  Fin Thickness and Fin Spacing: After much research, I settled on 2.5mm fin thickness at 7.5mm center spacing which leave a 5mm vent space between fins. This netted me a whopping 44 fins. For comparison, HDPlex has 17 fins and Streacom FC-10 has 20 fins. The 5mm space between fins meant that the mounting plates that press the heatpipes firmly into the groove can no longer be screwed from the outer side of the heatsink. By Using allen head screws, the mounting plates can be screwed in from the inside.

4.  Fin Height: My target was for my chassis to have an internal height of 100mm so the hetsink base will have the same 100mm height. However, to aid the flow of heat convection currents, I wanted 3 mm space between the top of the fins and the top panel. Same goes for the bottom panel. Hence the fin height will be net 94mm high.

 

For best heat radiation characteristics, black would be my best choice for color.

The heat pipe grooves were also located much higher up in heatsink so that all the grooves would be at the same level or higher than CPU heat block to let gravity aid in the flow of cooled liquid back to the CPU.

Another aesthetic choice was to fillet round the exterior corners of the fins by 5mm diameter.

I decided to put an L bracket at all of the 8 internal corners so that if either the top panel or bottom panel is removed, the chassis is still fully structurally sound.

 

 

Bottom Panel

 

The critical part here are the locations of the standoffs for the motherboard.

There are several online sources for the positioning of motherboard mounting points for all the different sizes of MB.

 

Here is a pic of my layout of MB standard mounting points. These set of points will then have to be positioned onto the bottom panel so that the I/O panel fits neatly into the rear panel I/O cutout and PCIE cards line up with the rear panel slot openings and that there is no gap between the rear panel and the PCIE brackets.

 

In my case, I just took guidance from my existing HDPlex chassis and positioned them at the same locations. Its easy to make a mistake here. I made a lifesized printout of my bottom panel design and overlayed my motherboard over it to make sure the mounting points all lined up.

 

 

 

 

 

 

Unlike the Top Panel, I wanted to be able to screw in the bottom panel into the underside of the Front Panel so the front of the bottom panel was extended out 12mm, leaving an ample 3mm cover.

 

Power Switch: My HDPlex power switch was located on the front end of the right heatsink. Since my new heatsinks only had 5mm space between fins, I decided to locate the 12mm diameter power switch near the front edge of the bottom panel. It would be a stealth location and there would be a soft blue light glow under the chassis from the switch.

 

Mounting holes for chassis feet: I added 4 M5 threaded holes at the 4 corners of the bottom plate to affix my favored Arya Lab Revopods.

 

Vent holes: With the bottom panel, I didn’t need the geometric pattern of the Top Panel. I settled for a simple rectangular distribution of 3.5mm vent holes

 

Thickness: for structural rigidity and "EMI waveguides", the bottom panel was also set at 10mm thickness.

 

 

Front Panel

 

The least complex part is the Front Panel. Yet arguably it has the most visual impact. Some would say the Top Panel has more visual impact.

 

Since the rear panel was 15mm, I decided to balance it out and also have a 15mm Front Panel. Design aesthetic choices:

-       I made up a logo that appealed to me and drilled it through the entire thickness of the front panel with a facility to add an LED power-on indicator if I opted to.

-       I made an elliptical groove across the lower third of the front panel and envisioned it to have a matte fine sandblasted finish while the large flat surface will have a matte fine-brushed natural aluminum finish, a two-tone texture to go with two-tone color of natural aluminum panels with matte black heatsinks.

-       Similar to the rear panel, the left and front vertical edges of the front panel was filleted/rounded with a 10mm curve. The curve and sides of the front panel will also have a matte fine-sandblasted finish.

 

Assembly Simulation

 

 

 

Having the facility to simulate assembly of the chassis and inspect/measure will save you $$$ in discarded prototypes due to errors easily overlooked when looking at individual panels.

You would also need to send the assembly model step files to the CNC facility so they can assist in checking the parts for fit.

 

The above are 3D drawings of the chassis design.

Here are some pics of the chassis as actually realized.

 

 

Above: Front and rear panels attached to bottom plate. Visible are standoff locations for motherboard, SCLK-Ex and Taiko ATX DC-DC converter as well as threaded holes for feet.

 

Above: Heatsinks in place

 

Above: 15mm rear panel

 

  

 

    

Weighs a hefty 13 kg. The front panel alone weighed in at 4kg

 

 

 

 Above with its matching chassis for Linear Power Supply

 

 

 

 

Above : pictured with full installation of:

- SoTM SMB-Q370 motherboard

- Intel Optane

- Taiko ATX DC-DC converter

- Two SCLK-ex PCBs stacked on top of each other

- SoTM SNI-1G lan card

- SoTM USBx10G

- Two tX-UsbHubInt (for double reclocking of USB signal

- 4x BNC clock input connectors for my Habst clock cable from a Mutec Ref10 120SE

 

Above: Heat pipes re-bent to fit the new height of the groove. All heat pipes are now above or at same level as the CPU heat spreading block.

 

And there she is!

 

 

 

 

And there it is. My personal dream chassis for my music server finally realized after several months of work on it.

Aside from the visual appeal, I had a rare opportunity to listen for SQ differences between the 2 chassis having simply transplanted the entire server 100% without any new wire or cable or new connectors. The sound I got was just like the chassis: Solid body yet airy, more elegance in the treble, smooth yet textured, more natural timbre, and lastly more stable and precise rendering of the 3D space. Usual caveats apply: in my system, to my ears.

Was it due to the stiffness of the chassis?

Or was it due to better EMI shielding?

Others would probably say it's simply expectation/confirmation bias.

I don’t know and I can't say, but I love it.

 

It was a lot of work but I believe I wouldn’t have gotten the results I got in another way.

 

Next will be building up the chassis for its matching LPS.

 

Many thanks to everybody who post here and keeps on sharing their tips and techniques and makes this such a wonderful forum.

 

[flkin]

This is the best DIY streamer I’ve ever seen in this thread. Wow!

 

Both case work and thought on the internals too. Well done Angelo! 👏👍

 

Please show us your power supply once it’s done too.

[Todd H]

Seriously impressive. Just wow.

[Exocer]

The difference in quality between this and the H5 is quite astonishing. I happen to have a set of these chassis as well. The LPS chassis will be used for my final LPS build (for the foreseeable future), while the server chassis will be used for my previous SOTM build. I can’t wait to fire this up again in its new chassis to listen for the improvements mentioned above by Angelo.

 

@SK8 special shout out to you as well. I won’t speak on Angelo’s behalf but as far as I’m concerned your chassis design really made me want to build my own chassis. I don’t have the time to come up with this on my own, so yeah, chatting with Angelo 1:1 eventually resulted in what you see here. Angelo’s interpretation of his dream chassis.

 

 

 

As you can see, it isn’t completely done yet but I couldn’t help not posting about it. To prospective buyers of this chassis, I am quite certain you will be amazed at the quality and SQ benefits. Thanks for bringing this to fruition @AngeloVRA. It is a game changer to say the least.

[The Computer Audiophile]

5 hours ago, AngeloVRA said:

Building a DIY chassis for my DIY Music Server

 

The chassis is an often overlooked component of DIY music servers and LPS. The big boys Extreme and Pink Faun have substantial, stiff, fit for purpose chassis..... that look beautiful too!  And there are no similar commercially available chassis.

 

Having been unsatisfied with currently available PC chassis from Streacom, HDPlex and even the Taiko DIY chassis, a few months ago I decided I would design and  3D model my own and go ahead and have it CNC manufactured.

 

This DIY post is intended to share the design thought process (not perfect), various technical issues, and solutions when making my own DIY chassis.

 

 

Where do I begin?

Firstly, I realized from previous projects that you would need to provide detailed STEP files and drawings  for the CNC facility to make your parts. This necessitates using some kind of 3D CAD/CAM design modelling software. After a bit of time comparing various options, I settled on Autodesk Fusion360.

If you are already proficient in 3D CAD/modelling software app, you’re in luck, else be prepared for a learning curve which is enjoyable though requiring what I’d call “more than a bit of time and effort”.

2D drafting apps such as Front Panel Designer are much easier to learn but are limited to one plane. For example, it is sufficient to design a rear panel until you decide you needed to threaded screw holes for the top of PCIE brackets which is on another plane.

2D drafting apps also cannot simulate assembly of the  various panels that will make up the chassis so you won’t be able to see how they fit together. Nor will you be able to effectively communicate to the CNC facility how the different parts are supposed to fit together. You will end up reordering the same part with various revisions/corrections which will end up costing you “more than just a bit”.

 

 

In my own journey, I designed the chassis in the following order:

-        Rear Panel

-       Top Panel

-       Front Panel

-       Heatsink

-       Bottom Panel (I dreaded measuring for the various motherboard standoff locations) 

 

I limited my chassis use case to :

-       SoTM smb-Q370 motherboard

-       2x stacked SoTM SCLK-EX boards

-       Taiko ATX DC-DC converter

-       34v Regulated DC power supply to the Taiko ATX

-       12v Regulated DC supply to the CPU

 

Rather than starting with a blank slate and researching the various standards for PC cases, I used my existing HDPlex H5 chasssis as a starting point.

 

 

My priorities were :

1.     The chassis must be stiff and sturdy, not requiring further bracing and reinforcement

2.     It must be fit for purpose. It must address EMI shielding and no more additional extensive drilling and thread tapping

3.     Aesthetically nice.

 

Rear Panel

 

 

Above: the rear panel of an HDPlex H5 V3

Design points:

1. The existing panel is flimsy and thin. My previous project which entailed simply replacing it with a 10mm thick panel proved beneficial to SQ, adding solidity and more precise imaging among others. The top of a PCIE bracket is 12mm long, so the rear panel will have to be at least 12mm thick.
2. The area above the PCIE brackets is a wide open space through which EMI can leak in/out. Covering that area would require another 3mm. Total thickness for my rear panel design is thus set at 15mm
3. Neither the bottom panel nor the top panel is screwed onto the rear panel which in my opinion compromises its stiffness. The area underneath the I/O panel is so thin (only 4.6mm as shown in pic)  and further compromises its structural rigidity. I decided that I would add 3mm thickness to that area. This meant that
   • I would have to raise the MB another 3mm. Existing MB standoffs are 7mm.  simple change to 10mm standoffs would accomplish the task. This improves cooling of the bottom of the MB.
   • In addition, I would also have to raise the PCIE slot openings by the same amount 3mm
4. The area on top of the PCIE slots where the PCI brackets are supposed to be screwed into is relatively thin at xx mm. Drilling threaded holes into it would puncture the bottom part which I thought would visible detract from its aesthetics. I decided I would beef it up and make it to thicker at yy mm by reducing the height of the PCIE slot opening by the same amount.
5. The bottom of PCIE brackets are supported from behind by the rear edge of the motherboard except those additional PCIE brackets to the right. Such additional PCIE brackets are usually used to hold connectors for clock inputs and DC power in for various cards, etc. Inserting the plugs usually cause the floppy PCIE brackets to flex inwards and doesn’t give one a solid feeling feedback. If the top of the brackets are secured by a screw, not a big deal. But in some cases e.g., with a Taiko DIY chassis, the bracket is not secured with a screw and the entire bracket can just come loose and fall back inside when one tries to plug in a clock cable or a DC cable.
   • In addition to threaded holes for the top of the PCIE brackets, I added a locking bar to secure the bottom of PCIE brackets
   •  I added 2 of my favored Jaeger chassis connectors for the 34v and 12v DC power supply input.
6. In the HDPlex H5, the thin rear panel is affixed to the heatsinks with externally visible bolts which I didn’t like. I opted to use internal “embedded” L brackets to affix the rear panel to heatsink. 

   • 

   •  L brackets would be unsightly if simply placed on top of the panels. Embedding them into the metal would be more visually appealing and helps torsional and shear stiffness of the joint.

   • Structurally, 1 screw per leg would be enough but just to be sure I would put 2 per leg. The thinner panel, the heatsink would be 10mm thick. Embedding the 3mm L-bracket leaves 7mm which nets me 5-6 thread depth for the screws. This is about 1.5x the diameter of an M4 screw which is perfect for the job.

   • 2x M4 screws with 8mm countersunk holes mean that the L- bracket had to be at least 20mm wide.

   • 

   •  

7. Datasheet for the Jaeger connectors show that the connector hole is 22 mm and 4 holes for M3 screws arranged in a 28 mm circle
   • 

8. The existing PCIE slot openings were only 13.2 mm wide which prior experience adding BNC chassis connectors to PCIE brackets show was just a bit too narrow. Thus this was expanded to 15.5mm which is also the space needed to be able to insert a Telegartner RJ-45 connector. Standard PCIE brackets are 18.5mm wide so we will have 1.5 mm latitude on each side.

9.  

   For stiffness, the top panel will be screwed onto the rear panel at 4 points, 1 at each end and another 2 over the IO panel. The bottom panel will similarly be screwed into the rear panel at 4 points, albeit asymmetrically, one at each end and 2 under the PCIE slot area.

10. Aesthetic choice was to fillet (round) the left and right edges of the rear panel at a diameter of 10mm. Similarly the I/O panel cutout was filleted at 2mm and the PCI slot dividers at 1mm

Top Panel

 

For the top panel, I simply followed Taiko’s lead with the Extreme that the vent holes would actually be tubes if the panel was thick enough... Ergo! EMI waveguides. This reduces EMI leakage as only the “straight” waves get through at full force. The waves that hit the walls of the tube are greatly diminished after they have bounced off the walls of the tube several times. This theory appealed to me.

Problem was:

-       How to draw the holes in a concentric circle pattern

-       With all the holes the same distance from each other

This required a bit of thought. It is not the only method but here is what I did (short version)

-       Assume we want 3mm holes with center spacing of 6mm

-       Draw concentric circles with each circle 6mm larger in radius than the previous.

-       For each concentric circle. Knowing the radius, you can calculate the circumference. Divide that by 6mm and you will get the number of holes you need to place around that circle to get close to 6mm spacing.

-       In my case, it was 6 holes for the 1st circle, then 12, 19, 25,31. 37.44 and so on. The top panel has over 2,500 holes.

 

Tedious, but here’s the result... 10mm Top Panel

 

The left and right edge areas will be covering the top of the heatsinks. And we need to open up that area so the heat can be expelled upwards from the heatsink fins.

My aesthetic choice for this is a set of triangular holes that form a truss-like structure. Easier choice would be circular or rectangular vent openings.

 

For the top panel. I decided to use counterbore M4 screws which would yield a stronger bond than when using countersunk screws. For aesthetics, there are no screws over the top of the front panel.

 

HEATSINKS

 

There are 4 design parameters to consider with heatsinks:

1.  The length of the heatsink front to back. My chassis design objective to have internal depth of 340mm. This shall be the same as the length of the heatsink

2. The thickness of the base of the heatsink determines its ability to spread the heat to other fins. The HDplex heat pipes are 6mm which will require 5.6mm deep grooves which will leave 4.2mm cover if we use 10mm as our base thickness

 

 

3.  Fin Thickness and Fin Spacing: After much research, I settled on 2.5mm fin thickness at 7.5mm center spacing which leave a 5mm vent space between fins. This netted me a whopping 44 fins. For comparison, HDPlex has 17 fins and Streacom FC-10 has 20 fins. The 5mm space between fins meant that the mounting plates that press the heatpipes firmly into the groove can no longer be screwed from the outer side of the heatsink. By Using allen head screws, the mounting plates can be screwed in from the inside.

4.  Fin Height: My target was for my chassis to have an internal height of 100mm so the hetsink base will have the same 100mm height. However, to aid the flow of heat convection currents, I wanted 3 mm space between the top of the fins and the top panel. Same goes for the bottom panel. Hence the fin height will be net 94mm high.

 

For best heat radiation characteristics, black would be my best choice for color.

The heat pipe grooves were also located much higher up in heatsink so that all the grooves would be at the same level or higher than CPU heat block to let gravity aid in the flow of cooled liquid back to the CPU.

Another aesthetic choice was to fillet round the exterior corners of the fins by 5mm diameter.

I decided to put an L bracket at all of the 8 internal corners so that if either the top panel or bottom panel is removed, the chassis is still fully structurally sound.

 

 

Bottom Panel

 

The critical part here are the locations of the standoffs for the motherboard.

There are several online sources for the positioning of motherboard mounting points for all the different sizes of MB.

 

Here is a pic of my layout of MB standard mounting points. These set of points will then have to be positioned onto the bottom panel so that the I/O panel fits neatly into the rear panel I/O cutout and PCIE cards line up with the rear panel slot openings and that there is no gap between the rear panel and the PCIE brackets.

 

In my case, I just took guidance from my existing HDPlex chassis and positioned them at the same locations. Its easy to make a mistake here. I made a lifesized printout of my bottom panel design and overlayed my motherboard over it to make sure the mounting points all lined up.

 

 

 

 

 

 

Unlike the Top Panel, I wanted to be able to screw in the bottom panel into the underside of the Front Panel so the front of the bottom panel was extended out 12mm, leaving an ample 3mm cover.

 

Power Switch: My HDPlex power switch was located on the front end of the right heatsink. Since my new heatsinks only had 5mm space between fins, I decided to locate the 12mm diameter power switch near the front edge of the bottom panel. It would be a stealth location and there would be a soft blue light glow under the chassis from the switch.

 

Mounting holes for chassis feet: I added 4 M5 threaded holes at the 4 corners of the bottom plate to affix my favored Arya Lab Revopods.

 

Vent holes: With the bottom panel, I didn’t need the geometric pattern of the Top Panel. I settled for a simple rectangular distribution of 3.5mm vent holes

 

Thickness: for structural rigidity and "EMI waveguides", the bottom panel was also set at 10mm thickness.

 

 

Front Panel

 

The least complex part is the Front Panel. Yet arguably it has the most visual impact. Some would say the Top Panel has more visual impact.

 

Since the rear panel was 15mm, I decided to balance it out and also have a 15mm Front Panel. Design aesthetic choices:

-       I made up a logo that appealed to me and drilled it through the entire thickness of the front panel with a facility to add an LED power-on indicator if I opted to.

-       I made an elliptical groove across the lower third of the front panel and envisioned it to have a matte fine sandblasted finish while the large flat surface will have a matte fine-brushed natural aluminum finish, a two-tone texture to go with two-tone color of natural aluminum panels with matte black heatsinks.

-       Similar to the rear panel, the left and front vertical edges of the front panel was filleted/rounded with a 10mm curve. The curve and sides of the front panel will also have a matte fine-sandblasted finish.

 

Assembly Simulation

 

 

 

Having the facility to simulate assembly of the chassis and inspect/measure will save you $$$ in discarded prototypes due to errors easily overlooked when looking at individual panels.

You would also need to send the assembly model step files to the CNC facility so they can assist in checking the parts for fit.

 

The above are 3D drawings of the chassis design.

Here are some pics of the chassis as actually realized.

 

 

Above: Front and rear panels attached to bottom plate. Visible are standoff locations for motherboard, SCLK-Ex and Taiko ATX DC-DC converter as well as threaded holes for feet.

 

Above: Heatsinks in place

 

Above: 15mm rear panel

 

  

 

    

Weighs a hefty 13 kg. The front panel alone weighed in at 4kg

 

 

 

 Above with its matching chassis for Linear Power Supply

 

 

 

 

Above : pictured with full installation of:

- SoTM SMB-Q370 motherboard

- Intel Optane

- Taiko ATX DC-DC converter

- Two SCLK-ex PCBs stacked on top of each other

- SoTM SNI-1G lan card

- SoTM USBx10G

- Two tX-UsbHubInt (for double reclocking of USB signal

- 4x BNC clock input connectors for my Habst clock cable from a Mutec Ref10 120SE

 

Above: Heat pipes re-bent to fit the new height of the groove. All heat pipes are now above or at same level as the CPU heat spreading block.

 

And there she is!

 

 

 

 

And there it is. My personal dream chassis for my music server finally realized after several months of work on it.

Aside from the visual appeal, I had a rare opportunity to listen for SQ differences between the 2 chassis having simply transplanted the entire server 100% without any new wire or cable or new connectors. The sound I got was just like the chassis: Solid body yet airy, more elegance in the treble, smooth yet textured, more natural timbre, and lastly more stable and precise rendering of the 3D space. Usual caveats apply: in my system, to my ears.

Was it due to the stiffness of the chassis?

Or was it due to better EMI shielding?

Others would probably say it's simply expectation/confirmation bias.

I don’t know and I can't say, but I love it.

 

It was a lot of work but I believe I wouldn’t have gotten the results I got in another way.

 

Next will be building up the chassis for its matching LPS.

 

Many thanks to everybody who post here and keeps on sharing their tips and techniques and makes this such a wonderful forum.

 

Members of this community continue to amaze me everyday. I absolutely love this!

[Nenon]

@AngeloVRA - very nice work indeed. Is it big enough to accomodate a dual CPU ASUS Sage motherboard?

[Nenon]

1 hour ago, Exocer said:

To prospective buyers of this chassis, I am quite certain you will be amazed at the quality and SQ benefits.

So this is a commercial product, not a DIY project?

[Exocer]

1 minute ago, Nenon said:

So this is a commercial product, not a DIY project?

Good question - I heard through the grapevine that people wanted to buy one of these (even someone I’ve built an lps for reached out to me directly about availability) and given the way we do things in the diy community, a group buy naturally came to mind. Why? Because producing only one of these chassis is WAY more expensive than having multiple built up. It would be a win-win for all involved.

[AngeloVRA]

16 minutes ago, Nenon said:

@AngeloVRA - very nice work indeed. Is it big enough to accomodate a dual CPU ASUS Sage motherboard?

Hi @Nenon,

Thanks for the compliment.

I didn't focus the design on the dual CPU Asus Sage. Main reason is that my server is based on an SoTM Q370 motherboard and  I didn't have an Asus Sage.

It is just a bit bigger than an HDPlex.

Internal dimensions are: HWD = 100 x 370 x 340mm

So the dual CPU ASUS Sage should fit :

   - will have to add a couple more mb standoffs

   - Not sure how or where to position the Taiko ATX DC-DC

   - certainly won't fit a ULPS inside... unless I redesign it for that specific use case.