S4MAX: Brickless S4M w/ 3090 FE and R9 5950x - 800W, 5l, water cooled

[petricor]

News from the postman!

This has just arrived...

...and after doing away with about 80 layers of foam sheets, it starts to very promisingly look like my custom machined parts from China would be here.

And voila:

Meet the non-optional heatsink bezel, updated to allow for air intake, and the 30x0 universal terminal module for the Corsair HydroX water block, all machined in Al6061 Aluminum alloy, sand blasted and anodised.

On the rear side of the heat sink you see (apart from a ton of my fingerprints) the pads designed to compensate for the openings in the S4M's front panel and allow for maximum contact with the PSU, and on the terminal module, the coolant ports interfacing with the Corsair's plexi block.

The air intakes are angled towards the PSU to direct the air flow, and contribute to making the openings less obvious - well, that's if there is no sunshine coming from behind!

This photo shows some of the complexity of the internals of the terminal module - compressing this into a bounding box suitable for both a 3080 and 3090 FE whilst not becoming the bottle neck in the water loop required pretty tight arrangements with down to 0.5mm material thickness between cavities.

Here you see the side facing G1/8 fitting interface - a nice to have but not required for my loop; their actual purpose is to enable the machining paths required to make the internal geometry work, which has to avoid any interconnection between the voids required for the screws and water flow.

The threads appear to be precise and the Festo NPQH G1/8 fittings sit tight. This is how it looks like with fittings and G1/8 blind plugs installed; a highly scientific blow-into-one-end-and-put-a-finger-onto-the-other-test confirms that everything is airtight and that there are no internal interconnections. That’s quite a relief as it has been a tad risky to design things with sub-mm tolerances, but it looks like the machining shop did an amazing job! Let alone the finish... rather happy with the way the parts came out!

A small but important feature of the terminal module a this little notch on the side facing the water block:

It makes space for a little bulge in the nickel-copper block of the Corsair Hydro X, an apparent necessity to accommodate for the screw-in PCB spacer; without that notch there would be no way to get to a water tight fit.

Time to put things together: Two M4x16 and one M3x14 screw should fit the original threads in the plexi block...


...and make for a reliable fit.

The design intent to avoid any projections of the terminal module beyond the plexi block's top surface...

...and the PCB layer (note the standoffs) appears to pan out - so this should be good for a water tightness test and ultimately a test fit onto the 3090!

 

[Vaderator]

A small but important feature of the terminal module a this little notch on the side facing the water block:

It makes space for a little bulge in the nickel-copper block of the Corsair Hydro X, an apparent necessity to accommodate for the screw-in PCB spacer; without that notch there would be no way to get to a water tight fit.

Good catch - I had mine flat at first and got it watertight through force, but had since carved the surface around the bulge.

Do you think you’ll have to sand / polish the watertight faces at all? Can’t tell from the pictures but they look like they have a course grain.

 

[AlexTSG]

Just spent the last hour catching up on Phase 2 of your amazing build.

I was just thinking that Asrock will probably release the X570S version of your motherboard, just as you've got everything sorted out and fitting properly.

Hopefully it will have exactly the same layout (to the millimeter), but with slightly less cooling requirements.

Also, if you're looking for a slim cooling solution for your SSD, I bought and used one of these ceramic heatsinks last month for a build I did for a friend. I have since ordered another 2 for myself:



I used the included thermal pads, but you could probably use something like a Minus Pad 8 for better results.

The final thing that's popped into my head is that Sugru could be useful for some of your power supply related work, as it works as a form of glue and it's electrically insulating.

 

[luca69002]

Hey! just wanted to make a S4M with external rad (MORA3), what would be my best option to have at least a 750w PSU in there ? (brick or brickless)

 

[morj]

Hey! just wanted to make a S4M with external rad (MORA3), what would be my best option to have at least a 750w PSU in there ? (brick or brickless)

Define "best"? Of course, PSUs with 40mm fans are louder than usual. You can go with a large passive or semi-passive "brick" made out of any decent 750w PSU with a solid single 12V rail just by using any DC plug with bypassed +5vSB or an off-the-shelf solution like the one from J-Hack. If space is a factor, you can definitely go brickless with a modded Supermicro PWS-804P-1R like described in this thread.

 

[petricor]

It has been a while - time for a gradual assembly attempt featuring some of my new parts!

On the upper half you see the 3.5mm bullet connectors I have successfully fried with the 3090- below the new hypothesis:
To the right, some MicroFit 3.0 connectors I intend to use for my own take on @Josh | NFC 's low-profile-connectors, this time for a NVidia 12-pin (thanks for the hint @Vaderator) - to the left, 5.5mm bullet connectors with a rather comfortable 120A rating - should just about do the trick!

The size comparison illustrates the upgrade - and regardless of diameter, the 5.5mm connectors have sprung segments rather than a spring insert with a much bigger potential contact area. The MicroFit contacts, however, don't cease to amaze - the current I'll get through the massive 5.5mm bullet will eventually be consumed by the six tiny pins to the right.

Now, taking things apart:

The S4M is not only an amazingly compact and well-finished case, it's also super nice to work with as pretty much every panel can be taken out without having to disassemle anyhing else. In my case, I take out the front panel to which I have attached my PSU...

...and can start to work on it without removing the PSU from the frame. Reason for choosing a beefier connector rather than adding another set of power leads is that the cables run in front of the PSU's exhaust air grille and reduce the ventilation cross-section - also, they are not the bottle neck... The bullet connectors on the left were those feeding the GPU, and you can see how dark they are compared to those for the CPU, clearly suffering from too much sustained current...

...which becomes even more apparent when comparing them to a new connector.

Annoyingly, the male connectors (unlike the female ones) lose some of their gold plating, exposing the copper underneath, when soldering them... not sure whether I caught a bad batch, or whether I need a beefier solder tip to heat the tin quicker - anyway, they are so over-dimensioned that this should not become an issue.

In preparation for wiring up the GPU, I prepare two wires ending in female 3.5mm connectors that I will use to distribute to six 18 AWG wires each ending in the MicroFit3.0 connectors for the GPU.

The connector seems to be of the right size...

...and after a bit of practice, I get a satisfying result crimping them onto the 18AWG silicone wires I have set aside for this - actually being a bit thicker than what the spec sheet calls for.

To make sure I get no nasty surprises in case a crimp is coming loose (you don't want a vagabonding +12V cable in your build...), I add some solder to the connection - better safe than sorry.

That's how the result looks like with some shrink wrap applied,

making for a surprisingly tight and robust fit in the 3090's 12-pin connector - it almost clicks in and takes quite some force before pulling out.

This is where I leave it for now:

To bend them into the correct angle using @Josh | NFC 's method for low-profiles connectors requires me to strip the GPU, and as I want to lengthen the branch cables correctly when I have routed them around the connector, I park this until I am tackling the GPU.

Along with the PSU, I also upgrade my other wiring to 5.5mm connectors (oops, bokeh-overkill):

From left to right, my "GPU starter cable", the ultimate GPU wires, and the CPU power connector.
The offset female/male pairing of the wires ensures that exposed male pins on PSU side are ground, and that the +12 female ends are reasonably well protected. Offsetting the cable ends by a connector's length makes a short at the PSU near-impossible, even with nothing plugged in; the shrinkwrap around the positive ends is cut with a tiny overstand to protect the connector edges.

With that done, I can fit the PSU back into the front frame - and: Tadaaa....

...attach my new heat sink bezel. This picture shows how the machined pads of the bezel match the perforations in the S4Ms frame...

...which is key to establish direct contact with PSU and for the bezel to contribute to heat distribution.

With some thermal tape for proper contact...

...the bezel goes on and fits precisely (which is always a bit of a relief when using self designed parts).

Four M3 hex screws hold it on the frame...

...with the air slots directed at the PSUs intake side.

Moment of truth:

Bingo!

Clearly that GPU still parks outside of the case - and the reason I haven't stripped it yet is this:

I have a bit of a riser situation.

As the 3090 is shorter than the 2080ti I had in there originally (which is a good thing as it stays clear of the main fan), the PCIe connector needs to move 55mm to the right (you see the notch where it has been before) - and that, in turn, means, that my amazingly thin HDPlex riser cable is 55mm too short.
So, my very naive first take on this is to daisy-chain it with a shorter riser and extend it, and spatially that's a perfect fit...

...and it does power up - but unfortunately not even get me to a bios screen.

Next bus stop: Shielding. Perhaps it's just some undesired interference (and yes, I have switched the PCIe x16 interface to 3.0)...

Applying some space-age aluminum foil, I wrap my cable for some neutrino-proof shielding...

...and BAM, one step further:

I get a boot screen.

But that's it - hang upon switching graphics modes at windows logon, so no HDPlex for me

Looks like I'll have to ponder over this a bit further - @Vaderator's great find of the pretty amazing Louqe Cobalt cable would not cut it in my case as the 3090 is a bit wider than the 3070/80, and the Cobalt's GPU connector is rather deep; I need something that puts the connector more or less flush at the case's rear for the GPU to fit. Or it's back to ultra-low profile RAM... but that would be a step back, and that's not what we are doing here.

The @LINKUP 4.0 prototype cable I am using for my dry runs with the 3090 is too tall on the GPU connector end at 24mm (that's measured normal to the GPU's PCB):

I only have about 15 to spare - so I need something that has pretty much the HDPlex's geometry...

...on a >=260mm cable.
Plus, it's a rather exotic right-angle-type (270 degrees).

PCIe 3.0 would be well-sufficient... any ideas, proposals, one-off engineering samples, prototypes, custom builds or wireless interfaces are most welcome!

 

[riba2233]

It has been a while - time for a gradual assembly attempt featuring some of my new parts!

On the upper half you see the 3.5mm bullet connectors I have successfully fried with the 3090- below the new hypothesis:
To the right, some MicroFit 3.0 connectors I intend to use for my own take on @Josh | NFC 's low-profile-connectors, this time for a NVidia 12-pin (thanks for the hint @Vaderator) - to the left, 5.5mm bullet connectors with a rather comfortable 120A rating - should just about do the trick!

The size comparison illustrates the upgrade - and regardless of diameter, the 5.5mm connectors have sprung segments rather than a spring insert with a much bigger potential contact area. The MicroFit contacts, however, don't cease to amaze - the current I'll get through the massive 5.5mm bullet will eventually be consumed by the six tiny pins to the right.

Now, taking things apart:

The S4M is not only an amazingly compact and well-finished case, it's also super nice to work with as pretty much every panel can be taken out without having to disassemle anyhing else. In my case, I take out the front panel to which I have attached my PSU...

...and can start to work on it without removing the PSU from the frame. Reason for choosing a beefier connector rather than adding another set of power leads is that the cables run in front of the PSU's exhaust air grille and reduce the ventilation cross-section - also, they are not the bottle neck... The bullet connectors on the left were those feeding the GPU, and you can see how dark they are compared to those for the CPU, clearly suffering from too much sustained current...

...which becomes even more apparent when comparing them to a new connector.

Annoyingly, the male connectors (unlike the female ones) lose some of their gold plating, exposing the copper underneath, when soldering them... not sure whether I caught a bad batch, or whether I need a beefier solder tip to heat the tin quicker - anyway, they are so over-dimensioned that this should not become an issue.

In preparation for wiring up the GPU, I prepare two wires ending in female 3.5mm connectors that I will use to distribute to six 18 AWG wires each ending in the MicroFit3.0 connectors for the GPU.

The connector seems to be of the right size...

...and after a bit of practice, I get a satisfying result crimping them onto the 18AWG silicone wires I have set aside for this - actually being a bit thicker than what the spec sheet calls for.

To make sure I get no nasty surprises in case a crimp is coming loose (you don't want a vagabonding +12V cable in your build...), I add some solder to the connection - better safe than sorry.

That's how the result looks like with some shrink wrap applied,

making for a surprisingly tight and robust fit in the 3090's 12-pin connector - it almost clicks in and takes quite some force before pulling out.

This is where I leave it for now:

To bend them into the correct angle using @Josh | NFC 's method for low-profiles connectors requires me to strip the GPU, and as I want to lengthen the branch cables correctly when I have routed them around the connector, I park this until I am tackling the GPU.

Along with the PSU, I also upgrade my other wiring to 5.5mm connectors (oops, bokeh-overkill):

From left to right, my "GPU starter cable", the ultimate GPU wires, and the CPU power connector.
The offset female/male pairing of the wires ensures that exposed male pins on PSU side are ground, and that the +12 female ends are reasonably well protected. Offsetting the cable ends by a connector's length makes a short at the PSU near-impossible, even with nothing plugged in; the shrinkwrap around the positive ends is cut with a tiny overstand to protect the connector edges.

With that done, I can fit the PSU back into the front frame - and: Tadaaa....

...attach my new heat sink bezel. This picture shows how the machined pads of the bezel match the perforations in the S4Ms frame...

...which is key to establish direct contact with PSU and for the bezel to contribute to heat distribution.

With some thermal tape for proper contact...

...the bezel goes on and fits precisely (which is always a bit of a relief when using self designed parts).

Four M3 hex screws hold it on the frame...

...with the air slots directed at the PSUs intake side.

Moment of truth:

Bingo!

Clearly that GPU still parks outside of the case - and the reason I haven't stripped it yet is this:

I have a bit of a riser situation.

As the 3090 is shorter than the 2080ti I had in there originally (which is a good thing as it stays clear of the main fan), the PCIe connector needs to move 55mm to the right (you see the notch where it has been before) - and that, in turn, means, that my amazingly thin HDPlex riser cable is 55mm too short.
So, my very naive first take on this is to daisy-chain it with a shorter riser and extend it, and spatially that's a perfect fit...

...and it does power up - but unfortunately not even get me to a bios screen.

Next bus stop: Shielding. Perhaps it's just some undesired interference (and yes, I have switched the PCIe x16 interface to 3.0)...

Applying some space-age aluminum foil, I wrap my cable for some neutrino-proof shielding...

...and BAM, one step further:

I get a boot screen.

But that's it - hang upon switching graphics modes at windows logon, so no HDPlex for me

Looks like I'll have to ponder over this a bit further - @Vaderator's great find of the pretty amazing Louqe Cobalt cable would not cut it in my case as the 3090 is a bit wider than the 3070/80, and the Cobalt's GPU connector is rather deep; I need something that puts the connector more or less flush at the case's rear for the GPU to fit. Or it's back to ultra-low profile RAM... but that would be a step back, and that's not what we are doing here.

The @LINKUP 4.0 prototype cable I am using for my dry runs with the 3090 is too tall on the GPU connector end at 24mm (that's measured normal to the GPU's PCB):

I only have about 15 to spare - so I need something that has pretty much the HDPlex's geometry...

...on a >=260mm cable.
Plus, it's a rather exotic right-angle-type (270 degrees).

PCIe 3.0 would be well-sufficient... any ideas, proposals, one-off engineering samples, prototypes, custom builds or wireless interfaces are most welcome!

Adt gen3 risers have shorter connectors (r33ul, r33sl etc)

 

[FreshGnar]

Moment of truth:

Bingo!

Clearly that GPU still parks outside of the case - and the reason I haven't stripped it yet is this:

I have a bit of a riser situation.

As the 3090 is shorter than the 2080ti I had in there originally (which is a good thing as it stays clear of the main fan), the PCIe connector needs to move 55mm to the right (you see the notch where it has been before) - and that, in turn, means, that my amazingly thin HDPlex riser cable is 55mm too short.
So, my very naive first take on this is to daisy-chain it with a shorter riser and extend it, and spatially that's a perfect fit...

...and it does power up - but unfortunately not even get me to a bios screen.

Next bus stop: Shielding. Perhaps it's just some undesired interference (and yes, I have switched the PCIe x16 interface to 3.0)...

Applying some space-age aluminum foil, I wrap my cable for some neutrino-proof shielding...

...and BAM, one step further:

I get a boot screen.

But that's it - hang upon switching graphics modes at windows logon, so no HDPlex for me

Looks like I'll have to ponder over this a bit further - @Vaderator's great find of the pretty amazing Louqe Cobalt cable would not cut it in my case as the 3090 is a bit wider than the 3070/80, and the Cobalt's GPU connector is rather deep; I need something that puts the connector more or less flush at the case's rear for the GPU to fit. Or it's back to ultra-low profile RAM... but that would be a step back, and that's not what we are doing here.

The @LINKUP 4.0 prototype cable I am using for my dry runs with the 3090 is too tall on the GPU connector end at 24mm (that's measured normal to the GPU's PCB):

I only have about 15 to spare - so I need something that has pretty much the HDPlex's geometry...

...on a >=260mm cable.
Plus, it's a rather exotic right-angle-type (270 degrees).

PCIe 3.0 would be well-sufficient... any ideas, proposals, one-off engineering samples, prototypes, custom builds or wireless interfaces are most welcome!

I came across this linkup reverse angle cable
Unfortunately, to fit your scenerio best it would need to be flipped 180 for the cable to be underneath the gpu and even then a slot would probably need to be cut in the back panel for the gpu side connector to protrude out of. I do suppose this particular cable might work if you route over the gpu. Then again it might not, as I'm not sure of the new gpu dimensions you have to work with. Merely shooting from the hip here, I actually didnt even realize reverse angle cables like this existed so figured I'd share. Good luck!

 

[Vaderator]

What you are looking for is going to be tough. I'm really surprised the HDPLEX one doesn't work. The 3090 should not require PCIE 4.0, maybe check to make sure resizable bar isn't enabled in your bios too?
Does the long HDPlex cable work when you don't daisy chain them? You might try a shorter PCIE 4 extension that terminates by connecting to the HDPlex cable.
My other thought is you might check with Gury on discord, I've heard he has a custom PCIE 4 cable in the works. This was his PCIE gen 3 cable that looks like it might just be on the edge of working:

here's a crazy view - it looks like it doesn't even terminate as a pcb on the plug-in side:

While you are at it check out his entire thread - he also has the lowest profile PCIE connector I've seen:
https://smallformfactor.net/forum/t...kless-1080ti-mini-case-breaking-record.10462/

 

[Vaderator]

One last thought - on your motherboard, you might be able to reclaim a few mm if you remove the metal enclosures surrounding your usb and sound ports. I've removed them once just to take a peek, and I don't think they were absolutely necessary.
Also, aluminum HVAC tape works nice as a EM shield and you can get it in black

 

[Valantar]

Any chance of asking if HDPlex can whip up an extra long cable? Guess that depends on the MOQs from their suppliers, but at least they have the contacts, production facilities, parts, etc.

 

[BaK]

Or maybe Li-Heat, worth a try asking @yawacool!

 

[lozza_c]

Define "best"? Of course, PSUs with 40mm fans are louder than usual. You can go with a large passive or semi-passive "brick" made out of any decent 750w PSU with a solid single 12V rail just by using any DC plug with bypassed +5vSB or an off-the-shelf solution like the one from J-Hack. If space is a factor, you can definitely go brickless with a modded Supermicro PWS-804P-1R like described in this thread.

Thanks. I understand Flex which seems straightforward, though I'm still getting my head around the other possible brickless options. If you were to go this route (i.e. something like an HDPlex or Meanwell) my understanding is something a little like this:


To my mind this is distinguished from Flex in that you have a choice of connector entries to use, and then wire this up internally to the HDPlex or Meanwell, and then onto a plugin DC unit like the J Hack or G Unique.

I may be completely wrong though. I'm guessing the difference between a C6 and C13 is wiring gauge and the amount of power it can deliver. The C13 'kettle lead' is quite ubiquitous here in the UK.

 

[morj]

Thanks. I understand Flex which seems straightforward, though I'm still getting my head around the other possible brickless options. If you were to go this route (i.e. something like an HDPlex or Meanwell) my understanding is something a little like this:


To my mind this is distinguished from Flex in that you have a choice of connector entries to use, and then wire this up internally to the HDPlex or Meanwell, and then onto a plugin DC unit like the J Hack or G Unique.

I may be completely wrong though. I'm guessing the difference between a C6 and C13 is wiring gauge and the amount of power it can deliver. The C13 'kettle lead' is quite ubiquitous here in the UK.

There's a dedicated thread on 12V PSUs (and specifically Meanwell), I think it's better to discuss this topic there. I haven't researched Meanwell and I don't know how it handles the trickiest part (standby power, on/off states, idle power consumption), but I'm sure it can be found there too.

 

[infoberg]

I may be completely wrong though. I'm guessing the difference between a C6 and C13 is wiring gauge and the amount of power it can deliver. The C13 'kettle lead' is quite ubiquitous here in the UK.

Yes, the only relevant difference between C6 and C14: C6 is rated up to 2,5A and C14 up to 10A. 2,5A @ 220V means you are safe up to 550W, so both are fit for a 200W PSU.

Update: actually it would be 240V @ 2,5A, so up to 600W

 

[lozza_c]

There's a dedicated thread on 12V PSUs (and specifically Meanwell), I think it's better to discuss this topic there. I haven't researched Meanwell and I don't know how it handles the trickiest part (standby power, on/off states, idle power consumption), but I'm sure it can be found there too.

Yes, the only relevant difference between C6 and C14: C6 is rated up to 2,5A and C14 up to 10A. 2,5A @ 220V means you are safe up to 550W, so both are fit for a 200W PSU.

Thanks both. Still learning as I go

 

[petricor]

Nice sketch!

Not too dissimilar from what I have in my last build. Used a C6 as more compact and it only needs to handle 360W- had to pick a slightly beefier C14 for this one here, for the aforementioned reasons.
NB that if you ever take your machine to a place with 110V AC, your power supply is likely to be able to deal with it (pretty much all present-day models can), but your connector may not, as the current doubles: So if you want to cover the full spec of your PSU, pick a connector that can handle twice what you'd need in Europe.

There's a dedicated thread on 12V PSUs (and specifically Meanwell), I think it's better to discuss this topic there. I haven't researched Meanwell and I don't know how it handles the trickiest part (standby power, on/off states, idle power consumption), but I'm sure it can be found there too.

The Meanwells work like HDPlex PSUs: Always-on for the main transformer, with standby power and other voltages generated at the plugin- that's the part that switches off, the main brick remains powered. It's no big deal, really: I use my old build a with wifi power strip and can kill the power feed via app if I don't use the machine. Nice extra: By setting the bios to boot-on-power, I can start it up remotely.

 

[petricor]

Thanks for the suggestions first of all - the G-Flex cable is amazing but turning the wrong way, unfortunately... but these flat "film" cables are definitely the way to go for a small build.

As I need to fold my riser twice to make the turn I need, I will (at least - plus bending radii) triple the thickness of any cable right over the CPU where I have the least vertical clearance, so cable thickness is a real concern.

So, with the HDPlex being pretty much what I need, I did some more research into why the HDPlex wouldn't daisy-chain and believe to be onto a solution now:

Looking at my improvised extension cables, at first glance, it looks like two daisy-chained risers with one connector in the middle.


Taking a closer look, however, it's really four risers daisy-chained, resulting from the particular female-female design of the HDPlex risers, using adapter PCBs between the female connectors:

And that may just be one too many, as each connection adds resistance to the link. In this case, we have six connections between board and GPU, vs just two with a normal cable.

In turn, this means, that combining the long and folded piece of flat cable I need, with a conventional male-female cable to extend it, should reduce the number of connections from six to four as I can ditch one of the adapters - so I give this layout here a test:

And BAM - I have a picture!

The black riser...

...will obviously need to be replaced by a ~50mm long one, but this could be a pretty standard issue, straight cable with no exotic angles or thickness requirements - there is ample space between the x16 socket on the board and the CPU.

Doing the counter-probe...

...with the short riser at the board and the long black one connected to the GPU...

...also gets me a working solution, so it looks like I have a POC for my flatpack x16; running benchmarks comparing the single 4.0 cable versus the daisy-chained 3.0 solution I cannot see any performance penalty, so the connection appears to be conforming to spec...

...and squeezing the folded HDPlex cable into position illustrates how this appears to also pan out spatially.

Looks like I'm good to tackle the GPU!

 

[robbee]

If you need to buy a new riser anyway, it's worth noting that there are also male to male risers from adt link: http://www.adt.link/product/R33_Jump.html

 

[Valantar]

If you need to buy a new riser anyway, it's worth noting that there are also male to male risers from adt link: http://www.adt.link/product/R33_Jump.html

That looks like a good idea for reducing the number of connections even further - though it would be troublesome if they ever need to do a BIOS reset and connect a 3.0 GPU to get into the BIOS, as they would otherwise likely be able to use the short riser for that purpose. Though I guess at that point disassembly is required no matter what, given the density of this build.