The Z Build(s) 3.0 - Go Big and/or Go Home

[Zarathustra[H]]

https://www.amazon.com/100pcs-KF2510-Female-Connector-Terminal/dp/B09T5PC7FL

Thanks for the link. I'm trying to understand, though, How have you determined that those are the right ones? Like, what are you looking at to make that determination?

I have found that there are many part numbers for the female terminals for these things. You can refer to them as KK254 female crimp terminals, part number 08500113 or engineering part number 2759-(P909)B.

I'm looking for an exact 1:1 part number match to confirm that they are the right thing, but the description for the parts you have linked contains none of these part numbers and instead introduces a new one, KF2510 which I have not seen before.

The organization (or lack thereof) of all of this is really a complete mess. How does this industry even function?

 

[Zarathustra[H]]

Vaguely amusing. I guess this is what you get when you search for "loose" and "female"

 

[pendragon1]

Thanks for the link. I'm trying to understand, though, How have you determined that those are the right ones? Like, what are you looking at to make that determination?

I have found that there are many part numbers for the female terminals for these things. You can refer to them as KK254 female crimp terminals, part number 08500113 or engineering part number 2759-(P909)B.

I'm looking for an exact 1:1 part number match to confirm that they are the right thing, but the description for the parts you have linked contains none of these part numbers and instead introduces a new one, KF2510 which I have not seen before.

The organization (or lack thereof) of all of this is really a complete mess. How does this industry even function?

its different materials used, size in mm, different brands, different gauge support, all sorts of things. afaik, the ones i linked is the standard for female pwm fan headers(2.54mm). which i think is what youre looking for based on the pic i quoted... that KK254 does allude to 2.54mm though.
you can get kits on amazon that include them with the casings: https://www.amazon.com/LUORNG-Female-Connector-Housing-Terminal/dp/B0BJP5Y7KX
yup. knowledge bases i guess.

 

[Zarathustra[H]]

its different materials used, size in mm, different brands, different gauge support, all sorts of things. afaik, the ones i linked is the standard for female pwm fan headers(2.54mm). which i think is what youre looking for based on the pic i quoted... that KK254 does allude to 2.54mm though.
you can get kits on amazon that include them with the casings: https://www.amazon.com/LUORNG-Female-Connector-Housing-Terminal/dp/B0BJP5Y7KX
yup. knowledge bases i guess.

While I greatly appreciate your reference and explanation, I eventually after googling and seemingly going in circles found a reference to the exact Molex part number for the loose terminals for these connectors, it is 0855102*. Digikey did not have them in stock, but Mouser did!*

I got $200 of them for $20. That should give me more than enough to learn from and mess up several in the process.

They did have a warning on their site about shipping delays due to a backlog after a previous weather event, so who knows when I will actually get them. I think I'll have to move on to working on custom pump mounting while I wait for these to arrive.


Interesting side note:

Today I learned (from the Molex spec sheet) that Molex only rates these terminals for 25 insertions. I bet you never knew that about your fan connectors, huh?

Edit:
*If you are coming across this later, I ordered the wrong parts a few times (for various reasons) but eventually got these connector housings, and these terminals.

 

[pendragon1]

Today I learned (from the Molex spec sheet) that Molex only rates these terminals for 25 insertions. I bet you never knew that about your fan connectors, huh?

i thought it was like 30, been a while. but yes most are rated a lot lower than people would expect. they get compressed and dont connect properly after too many uses.
those ones you found are higher gauge, 22-30 vs 30-33 and brass instead of tin, so different number id assume. plus, like i said different brands have different numbers.

 

[Nobu]

Maybe I need something like this?

https://www.amazon.com/Precison-Cutters-Cutting-Electronics-Aluminum/dp/B07VGK8GBL

Yeah good precision cutters might work, or maybe flush cutters, doesn't need to be perfect but the small cutter helps get it close.

Diagonal cutters don't technically cut, it's more of a shearing action. They're good for cutting harder metal, but they mash anything soft, and because of the shape of the cutter it's difficult to get a straight cut and impossible to flush cut.

 

[Zarathustra[H]]

So, progress has been a little bit slow lately, in part because I have been working on two parallel paths, and both have hit obstacles.

While I greatly appreciate your reference and explanation, I eventually after googling and seemingly going in circles found a reference to the exact Molex part number for the loose terminals for these connectors, it is 0855102. Digikey did not have them in stock, but Mouser did!

I got $200 of them for $20. That should give me more than enough to learn from and mess up several in the process.

They did have a warning on their site about shipping delays due to a backlog after a previous weather event, so who knows when I will actually get them. I think I'll have to move on to working on custom pump mounting while I wait for these to arrive.

Interesting side note:

Today I learned (from the Molex spec sheet) that Molex only rates these terminals for 25 insertions. I bet you never knew that about your fan connectors, huh?

First we have the Molex fan connector terminal issue. I finally received the terminals I ordered above. Sadly they were not exactly what I was looking for.

Don't get me wrong, they are the correct parts, and will work, and they no longer have the wings, but I didn't realize that I accidentally ordered the gold plated variety.

My understanding is that it is not a great idea to mix gold plated and tin plated terminals in contact with eachother, as the gold has a nasty habit of - through what is called "tin fretting" degrading, flaking off, and leaving you with a poor connection.

I don't know a whole lot about it, but I presume it is some sort of galvanic effect.

Every fan header I have ever seen is tin, not gold, so it would make sense that I do my best to get a hold of the tin versions of the terminals.

For a while there I was thinking it wasn't going to be possible. I found one Molex part number, but the minimum order quantity everywhere I looked was 96,000 pieces. (just a little bit more than I need)

But then eventually I found an equivalent loose part number that was in stock at Mouser and ordered them. So now we wait... again...

 

[Nobu]

Don't have any tin-lead solder? Could remove the gold and apply solder to the pins where they make contact.

 

[pendragon1]

lol gold not brass eh.

should still be fine for fans.

 

[Zarathustra[H]]

Don't have any tin-lead solder? Could remove the gold and apply solder to the pins where they make contact.

I probably could. But at this point I want to minimize my tinkering with tiny parts. That stuff drives me nuts.

 

[Zarathustra[H]]

lol gold not brass eh.
View attachment 712498
should still be fine for fans.

The center of the conductor is brass, but the plating is gold.

The ones I need have plating (or maybe they are just dipped?) that is tin on top of that same brass conductor.

At least that is my understanding.

 

[Zarathustra[H]]

Alright, the second path has been working out how to mount the reservoirs and pumps to the sides of the radiators.

After initially deciding to make my own bracket, and then later changing my mind and ordering Watercool's reservoir brackets, I am now back to making my own, because of the issue I previously mentioned where they sat a bit too high for my tastes.

I bought a 24" long 1/4" sheet of ABS, that I had planned on cutting into strips (70mm wide) to run the length of the side of the radiators, for me to cut custom holes in and mount my stuff.

I didn't want to make a mess indoors cutting though, so the plan was to do it outside, but we had a slushy messy ice storm followed by a week during which the temperatures never climbed above freezing, and I didn't want to work outside in 10°F weather, so I waited.

Today it was a balmy 30°F out, so I decided it was time.

One thing I have learned over the years of tinkering and trying to make stuff, is that even if you have steady hands, it can be surprisingly difficult to cut straight lines linger than what you can do with a single move of a miter saw blade.

If I had a band saw with a backstop, that would be amazing, but sadly middle school shop class lied to me about the tools I was likely to have at my ready disposal in my life.

If you use a jigsaw hand or powered, it becomes squiggly and uneven no matter what you do.

I was going to have to make do with my circular saw. Just for good measure, I went to Harbor Freight and picked up a blade intended for cutting plastics, as I wasn't convinced my wood blade was going to make a clean cut. (something to do with rotational speeds and number of teeth)

To get a straight cut, I decided to do it like I would ripping a plank along the grain, and clamped down a known straight piece of wood, and measured my cut distance to that, with the plan of running the side of the circular saw along it.

It was a pain in the but to get both sides precisely aligned, as I would first align one, and then move to the other side, and my adjustments on the other side caused the first side to move, etc. etc.

Eventually I got it dialed in, and it was time to cut:

I was dreading this a little, but it actually wound up turning out pretty good.

The width of the two pieces wound up being more repeatable than it had any right to be (within a millimeter).

While I was having some success, I decided to also cut a third piece, and then chop it into 8 squares to use as spacers or something, when I go to assemble it all, if I need them.

At least that is a little bit of progress

 

[Zarathustra[H]]

Alright, please forgive the silence here.

This past weekend was my baby sisters wedding shower, and I was volunteered to help entertain the men while the women did their thing.

After that I was sick as a dog. So not much progress.

But, I was hell bent on making at least a little bit of progress without the weekend completely disappearing.

I remembered that I had bought an AMD RX 7600 (non-xt) on eBay to go into the workstation once I pull the 4090. I also remembered that I had done that in January. Whoops. More than 30 days had past, and I hadn't tested it yet to make sure it was in working order. Rookie eBay mistake...

The one I bought was an OEM part out of a Lenovo "Gaming Desktop". It looks like someone bought large quantities of these Lenovo "gaming desktops", pulled all of the GPU's, upgraded them and sold them, so they had like 200 of these RX7600's in essentially new, never used condition for sale. I made a half-lowball offer, and - surprisingly - got it.

While I do frequently buy used Enterprise equipment from decommissioned servers on eBay, I don't normally buy consumer hardware there. Usually it has been abused, and the seller is some kid who knows nothing about ESD bags or filling a package with packing materials. I decided to make an exception as this appeared to be a high volume seller, with good reviews, and I got the feeling they knew what they were doing.

So, my 30 day buyer protection was now up, and I hadn't gotten around to testing it yet.

I popped it into the testbench machine (this is one of the reason I have a testbench machine, to test hardware without having to pull apart my main machine). I got a BIOS splash logo, but then nothing. No text output during post, no BIOS, no display output in either Linux or Windows.

Oh oh.

But then I thought, you know, my testbench uses not only old (Ivy Bridge era) hardware, but also Server hardware. This is so far outside of what an RX 7600 would have been validated on to be silly. Maybe I should test it on newer hardware. But I didn't want to break into the workstation, and I have given away a bunch of my newer and more usable hardware lately, so I didn't have many options.

I pulled out my old Sandy Bridge era i7-3930k and Asus p9x79 motherboard and hooked it up open air for some testing.

This one was worse than the previous one. No image output at all aty any time. Not even a splash screen.

Starting to get more worried that I had been scammed (or at least sent a defective part, that I now would not be able to do anything about)

I decided I needed to test it in something newer. I knew I had a Coffee Lake i5-9500 I had used in an HTPC at one point somewhere, but I couldn't find it. (it must be in a box buried somewhere). So I sighed, and decided I was going to have to test it on the actual workstation.

I'm glad that I decided to build my current build with Koolance QDC's, because that allowed me to pop out the water cooled 4090, pre-fill a length of tubing with two QDC's on it to bypass the GPU and keep the water flowing, and then install the GPU in the desktop.


Apparently I forgot to take before pics, so here is what it looked like a couple of years ago when I removed my old 6900xt:


6900xt installed:

6900xt removed:

GPU bypass and 7600xt installed:

(Apparently it got a little dusty in there when I did my heat pump project last year, and I never noticed. Whoops. I guess that is a risk when you stick the case out of sight in another room.... I'd freak out, and give it a good blow out, but I am going to disassemble the whole thing pretty soon anyway. Everything will get a thorough clean before being put in the rack or put into storage anyway)

Anyway, long story short, it was a success!

In the TRX40 Threadripper board, it fired right up with out any issue and worked as expected.

My conclusion is that AMD's 7000 series GPU's just don't like older systems very much. Not sure what the incompatibility is, but it is real.

I have decided to just leave the 7600xt in the system until the changeover. I'm not going to have time for any big heavy games before then anyway, and if I am going to have any issues with it, I might as well learn about (and do something about) them now.

 

[Zarathustra[H]]

Alright, please forgive the silence here.

This past weekend was my baby sisters wedding shower, and I was volunteered to help entertain the men while the women did their thing.

After that I was sick as a dog. So not much progress.

But, I was hell bent on making at least a little bit of progress without the weekend completely disappearing.

I remembered that I had bought an AMD RX 7600 (non-xt) on eBay to go into the workstation once I pull the 4090. I also remembered that I had done that in January. Whoops. More than 30 days had past, and I hadn't tested it yet to make sure it was in working order. Rookie eBay mistake...

The one I bought was an OEM part out of a Lenovo "Gaming Desktop". It looks like someone bought large quantities of these Lenovo "gaming desktops", pulled all of the GPU's, upgraded them and sold them, so they had like 200 of these RX7600's in essentially new, never used condition for sale. I made a half-lowball offer, and - surprisingly - got it.

While I do frequently buy used Enterprise equipment from decommissioned servers on eBay, I don't normally buy consumer hardware there. Usually it has been abused, and the seller is some kid who knows nothing about ESD bags or filling a package with packing materials. I decided to make an exception as this appeared to be a high volume seller, with good reviews, and I got the feeling they knew what they were doing.

So, my 30 day buyer protection was now up, and I hadn't gotten around to testing it yet.

I popped it into the testbench machine (this is one of the reason I have a testbench machine, to test hardware without having to pull apart my main machine). I got a BIOS splash logo, but then nothing. No text output during post, no BIOS, no display output in either Linux or Windows.

Oh oh.

But then I thought, you know, my testbench uses not only old (Ivy Bridge era) hardware, but also Server hardware. This is so far outside of what an RX 7600 would have been validated on to be silly. Maybe I should test it on newer hardware. But I didn't want to break into the workstation, and I have given away a bunch of my newer and more usable hardware lately, so I didn't have many options.

I pulled out my old Sandy Bridge era i7-3930k and Asus p9x79 motherboard and hooked it up open air for some testing.

This one was worse than the previous one. No image output at all aty any time. Not even a splash screen.

Starting to get more worried that I had been scammed (or at least sent a defective part, that I now would not be able to do anything about)

I decided I needed to test it in something newer. I knew I had a Coffee Lake i5-9500 I had used in an HTPC at one point somewhere, but I couldn't find it. (it must be in a box buried somewhere). So I sighed, and decided I was going to have to test it on the actual workstation.

I'm glad that I decided to build my current build with Koolance QDC's, because that allowed me to pop out the water cooled 4090, pre-fill a length of tubing with two QDC's on it to bypass the GPU and keep the water flowing, and then install the GPU in the desktop.


Apparently I forgot to take before pics, so here is what it looked like a couple of years ago when I removed my old 6900xt:


6900xt installed:
View attachment 714700

6900xt removed:
View attachment 714701

GPU bypass and 7600xt installed:
View attachment 714702

(Apparently it got a little dusty in there when I did my heat pump project last year, and I never noticed. Whoops. I guess that is a risk when you stick the case out of sight in another room.... I'd freak out, and give it a good blow out, but I am going to disassemble the whole thing pretty soon anyway. Everything will get a thorough clean before being put in the rack or put into storage anyway)

Anyway, long story short, it was a success!

In the TRX40 Threadripper board, it fired right up with out any issue and worked as expected.

My conclusion is that AMD's 7000 series GPU's just don't like older systems very much. Not sure what the incompatibility is, but it is real.

I have decided to just leave the 7600xt in the system until the changeover. I'm not going to have time for any big heavy games before then anyway, and if I am going to have any issues with it, I might as well learn about (and do something about) them now.

As a side note, I am really liking this little RX 7600.

I mean, it's no 4090, but it sips power and on the desktop it is every bit as snappy as the 4090.

In Linux, I think it may even snappier than the 4090. It feels especially responsive when using my Windows 10 Virtualbox VM under Linux. Maybe due to AMD's excellent Linux open source drivers.

Back in the day I used to always buy Nvidia because of their excellent binary blob Linux drivers, while AMD's Linux drivers were trash. It's funny how things change.

For shits and giggles I did some RX 7600 performance testing.

If anyone is curious:

RX 7600 Time Spy: 10527 overall score, 10362 graphics score.

RX 7600 Time Spy Extreme: 5395 overall score, 4962 graphics score.

RX 7600 Port Royal: Score: 5449

I also did some testing in Civilization 6 - Gathering Storm.

I normally run that game at ~120fps at 4k native, with all settings maxed on the 4090.

The late game canned benchmark was a little slow, but once I reduced MSAA 8x down to MSAA 2x, the game runs great at 4k on the RX 7600. Civ 6 Gathering Storm is totally playable. ( I probably wouldn't enjoy most more modern titles on it at 4K though.)

But since this is my workstation, and only rarely intended to be used for games, that is not a problem at all.

 

[Zarathustra[H]]

As a side note, I am really liking this little RX 7600.

I mean, it's no 4090, but it sips power and on the desktop it is every bit as snappy as the 4090.

In Linux, I think it may even snappier than the 4090. It feels especially responsive when using my Windows 10 Virtualbox VM under Linux. Maybe due to AMD's excellent Linux open source drivers.

Back in the day I used to always buy Nvidia because of their excellent binary blob Linux drivers, while AMD's Linux drivers were trash. It's funny how things change.

For shits and giggles I did some RX 7600 performance testing.

If anyone is curious:

RX 7600 Time Spy: 10527 overall score, 10362 graphics score.

RX 7600 Time Spy Extreme: 5395 overall score, 4962 graphics score.

RX 7600 Port Royal: Score: 5449

I also did some testing in Civilization 6 - Gathering Storm.

I normally run that game at ~120fps at 4k native, with all settings maxed on the 4090.

The late game canned benchmark was a little slow, but once I reduced MSAA 8x down to MSAA 2x, the game runs great at 4k on the RX 7600. Civ 6 Gathering Storm is totally playable. ( I probably wouldn't enjoy most more modern titles on it at 4K though.)

But since this is my workstation, and only rarely intended to be used for games, that is not a problem at all.

Interestingly enough, another difference I have noticed in going from the 4090 to the RX 7600 is a big quality of life improvement, at least in Linux.

With the 4090, since I use an LG OLED TV as my monitor, whenever the screens were asleep, I'd have to be careful to turn on the TV using the remote first, and then wait until it fully starts up. Only once fully started could I jiggle the mouse to wake it up. If I jiggled the mouse too soon, the TV would never be detected, and would sit there with no input, and once that happened, nothing would help, other than unplugging the HDMI cable and plugging it back in again. A huge annoyance.

This is not the case with the RX7600 at all. The TV is always detected once it wakes up, and I never have to worry about accidentally waking the screens before the TV is done "booting".

I can't remember if this happened in Windows or not. I usually only use WIndows for games, so I boot it up, start up my game, play it and shut it down again. Rarely is it sitting booted with screens asleep in Windows.

 

[Zarathustra[H]]

Alright, the second path has been working out how to mount the reservoirs and pumps to the sides of the radiators.

After initially deciding to make my own bracket, and then later changing my mind and ordering Watercool's reservoir brackets, I am now back to making my own, because of the issue I previously mentioned where they sat a bit too high for my tastes.

I bought a 24" long 1/4" sheet of ABS, that I had planned on cutting into strips (70mm wide) to run the length of the side of the radiators, for me to cut custom holes in and mount my stuff.

I didn't want to make a mess indoors cutting though, so the plan was to do it outside, but we had a slushy messy ice storm followed by a week during which the temperatures never climbed above freezing, and I didn't want to work outside in 10°F weather, so I waited.

Today it was a balmy 30°F out, so I decided it was time.

One thing I have learned over the years of tinkering and trying to make stuff, is that even if you have steady hands, it can be surprisingly difficult to cut straight lines linger than what you can do with a single move of a miter saw blade.

If I had a band saw with a backstop, that would be amazing, but sadly middle school shop class lied to me about the tools I was likely to have at my ready disposal in my life.

If you use a jigsaw hand or powered, it becomes squiggly and uneven no matter what you do.

I was going to have to make do with my circular saw. Just for good measure, I went to Harbor Freight and picked up a blade intended for cutting plastics, as I wasn't convinced my wood blade was going to make a clean cut. (something to do with rotational speeds and number of teeth)

To get a straight cut, I decided to do it like I would ripping a plank along the grain, and clamped down a known straight piece of wood, and measured my cut distance to that, with the plan of running the side of the circular saw along it.

It was a pain in the but to get both sides precisely aligned, as I would first align one, and then move to the other side, and my adjustments on the other side caused the first side to move, etc. etc.

Eventually I got it dialed in, and it was time to cut:

View attachment 712502 View attachment 712503

I was dreading this a little, but it actually wound up turning out pretty good.

The width of the two pieces wound up being more repeatable than it had any right to be (within a millimeter).

While I was having some success, I decided to also cut a third piece, and then chop it into 8 squares to use as spacers or something, when I go to assemble it all, if I need them.

View attachment 712504

At least that is a little bit of progress

Alrighty, here's a little bit of progress.

I'm actually a little embarrassed how much time this took me, but it has been a game of trial and error, and then one of repeated assembly and disassembly to see if the hole I just made lines up and everything fits, only to do this over and over and over again

So I used the strips of ABS I cut last week to put together my first reservoir/pump bracket.

Now, these were supposed to be enough for both of them, but it didn't turn out that way.

The original plan was to attach one strip to each radiator using the screw holes present in them (four M4's, one M3 and and another M3 for a little alignment feature, which is quite nice. At least I think those are the measurements on the screws. Not 100% on that. I just pulled my screws out of my metric machine screw kit, and used the ones that fit.

But I ran into a little issue. (Or at least a refinement of my plan in my head as the project is underway.

The pumps I am using have the ability to use an inlet on top, and an outlet 90 degrees to the side (or both inlet and outlet on the side). In the interest of making each radiator loop as short and limited in bends as possible, I thought it would be best to use the top inlet, and the side outlet going straight into the radiator using a rotary male-to-male fitting.

Only one issue. That makes the pump stick out too much. (seen here with my first failed pump spacer bracket experiment. My smallest hole saw - 1" - was too big)

And seeing that the G1/4 threads are straight into the plexi, I didn't feel comfortable having all of the weight hanging on that lever arm. It just seems like a crack risk waiting to happen.

Additionally, the big hole is there to allow for cable routing. I plan on attaching a 1" liquid-tight conduit there. (It doesn't need to be waterproof. I just want to use it for cable management.) But there really isn't much space between the bracket and those pipes. Not even sure the threads of the liquid-tight fitting could be able to mount there without interfering with them, so I wanted to add another layer there too.

So I decided to go fro a three layer design. Base layer attaches to the radiator, and also gets some pressed in threaded inserts to help attach the radiator and pump. Second layer is smaller than the base layer, and sits on top of it, acting as a lid for some of the larger holes, gets a hole designed for the liquid-tight fitting, and acts as a spacer for the pump bracket.

For threaded inserts, I tried these in a 6-32, UNC size. They are specifically designed for plastic. No heat or ultrasonic necessary. You just drill a 3/16ths hole, and press them in. First time you put a screw in them, they expand to sit extremely tight in the hole.

Here is a test piece I inserted one in to make sure it worked well before possibly butchering my main bracket pieces.

I get the impression there is a specialty tool to press them in, but I don't have it. I tried a few methods to press them in, including my benchtop vise on my workbench (worked, but overkill) a screw style wood clamp with pieces of wood on either side (annoying, too many loose pieces) and finally, just pressing it in with a channel lock. This was easy, and just worked.

The first time you screw a screw into one of these to expand it, it is really tight. The friction actually makes both the screw and the insert very hot. So I just grabbed the longest 6-32 UNC screw I could find (only like 2") and screwed it all the way in and out several times in a row. The parts got nice and hot. My thought process is that this softens the plastic and re-hardens it around the insert. No idea if that is actually true or not, but I figured it was worth trying.

Here are the three layers attached to the radiator:

For what it is worth, this is a work in progress. I'm not even bothering wiping down figerprints or smudges until it is all done, and I take the final pics. I'm going to touch these things a million times, and that is going to happen.

While installing the pump is a real pain (I'm sure auto mechanics would call it a real engineers special) It does work, and everything fits. Time to drill holes for the reservoir, and press in some more inserts.

Here are the thee finished layers:

They are full of classic "early prototype jank", but most of that is hidden when installed, so who cares?


Here is a closeup on the pump mount.

Here we have the reservoir attached:

Here's where the tubes from the res to the pump will go. I hope they don't interfere with the conduit too much...

There is more space up top in this configuration than what was originally planned. I might even fit the QDC's now, but I am not sure I want to anymore. I'm not sure how much flow restriction the Koolance QD4's actually add, and I don't want to slow down the flow...

Oh, and here is the conduit fitting. We will have to see how much interference between the tubing and the conduit there winds up being.

I am not attaching any tubing yet, as I wound up having to order more ABS to make the second bracket, and when it gets here, I am going to disassemble the whole thing again, and use the existing parts as templates for the second bracket, so it is going to need to be completely disassembled yet again...

Anyway, I was happy to finally get through the design of the bracket.

 

[Zarathustra[H]]

Looks like I should have bought the Motherboard for phase 2 (the game system) while I could.

I have only been able to find two x870 or x870e motherboards that have two Ethernet ports and at least one of them is 10gig.

The Asus ProArt X870E (1x 10gig + 1x 2.5gig) and the Gigabyte AI Top (2x 10gig)

The Asus ProArt board has been out of stock for months.

The Gigabyte AI TOP was available at an obscene $799 when I started this, but since then it has been subject to retailer scalping. I can only find it at over $1200 now. I didn't want to pay the $799 original price tag, and I am sure as hell not going to pay over $1200 for it.

Plan B was to find a motherboard with just a single 10gig port, and add in a cheap gigabit or multigig nic in a chipset PCIe slot. Boards with a single 10gig slot are also either unavailable, or only have two PCIe slots (and if you use both, the primary drops down to 8x.)

It may have to be Plan C I go with then. A board with a more normal single multigig NIC on board, and an available minimum of 4x slot off of the chipset (that does not drop the primary PCIe slot down) that can be used for a 10gig NIC. There aren't too many available boards with 4x chipset slots, but there are some. The MSI x870e Carbon (I'm tempted to call it al carbon. Mmm Tacos) seems like it might be a decent option here. the Gigabyte x870e Aorus Master might be another one. While being just under $500 didn't used to seem like a good deal, by comparison, maybe it is. Going to continue poking around and seeing what else I can find.

This used to not be possible. At least not affordably. All of the 10gig NIC's I could easily get my hands on were decommed server NIC's, and thus required 8x Gen2 slots. In my experience, while 4x Gen2 should be more than enough bandwidth for a single 10Gig port, at least the intel NIC's I've used that were 8x Gen2 in the past behaved very very strange when given anyhting less than 8x lanes.

It looks like there are a handful of options now for 4x Gen3+ cards now. Most of them seem based on the Aquantia AQ107 chip. I have never used an Aquantia NIC. Hopefully they work well.

Edit:

Hmm.

I just realized something.

If I am going to use the 3U Sliger case with the sideways mount GPU (like I was planning in order to make the GPU fit), I'm not sure how easy it will be to use these other PCie slots, as the GPU will pretty much block them all.

it looks like there is only one upright slot, and it is in the location where I'd expect to find the primary GPU slot, meaning, it will be blocked by the riser connector...

Unless - since my GPU with water block is now two slot, I get yet another riser, and try to mount it sideways as well right next to the GPU. That might be a little sketchy...

I may have to hold out for other motherboards or rethink my case strategy and go with something 5U or taller so the GPU can fit standing up...

 

[pendragon1]

niche demands, niche price... yipes! get what you can and add them in.

 

[Zarathustra[H]]

niche demands, niche price... yipes! get what you can and add them in.

Yeah, 5u+ cases aren't exactly super common or cheap either. Silverstone has a few, but they are ridiculously expensive.

I'm not familiar with RackChoice, but they have one that is not terribly priced that may meet my needs.

It's probably better to not have to deal with the complications from a riser anyway, especially as GPU's start getting into Gen5 speeds where signal issues become more and more difficult...

Of course it throws off my rack spacing plan, and may require me to move things again, but can't do much about that... I should still be able to fit a 5U and 4U cases where I had intended to fit the 4U and 3U cases, but there won't be any spacing between them, so it will be a little less flexible.

 

[pendragon1]

thats a decent looking 5u without breaking the bank

 

[Nobu]

Central computer has a few for not stupid prices. Didn't look to see if they were in stock or special order though.

 

[Zarathustra[H]]

Central computer has a few for not stupid prices. Didn't look to see if they were in stock or special order though.

I was not familiar with that retailer.

I will have to look them up. It's always good to have another decent one to turn to. Looks like they are primarily in the bay area, but they do ship.

Thanks for that heads up.

 

[Zarathustra[H]]

Another weekend, another limited amount of progress.

I spent some time trying to duplicate my brackets for the second radiator. No pics here. There is nothing new to show off, and I have a little bit of trimming left to do before it will work just right, and I ran out of daylight before i got that done...

I'll say this. I'm an engineer. I am pretty good at being an engineer. But there is a difference between designing a place for a hole on a drawing, and actually making that hole.

I've always had a lot of respect for good fabricators/machinists. I am not one of those. This project has only reinforced my respect for those who are good at it.

So, I am not quite yet at the level of "manufacturing maturity" to be able to claim that I can make "replaceable parts". Each of these two brackets required significant fine tuning to get them to work, and replacing one of the layers with one from the other bracket would result in them just not fitting at all.


I also spent some time working on the C-channels I had originally bought to reinforce the side panel (which ultimately wound up being unnecessary).

I am re-using them as mounting bars for the box that will contain the fan controllers, power supply (and possibly some pumps).

I wound up cutting them down to 19" and then marked the cage nut holes on them:

Then I went to drill the holes using my old shitty drill press.

Last winter I had a mouse move in to my shed where I have my main workbench. I eventually wound up killing it with bait. I knew it built a nest in there somewhere, but I just couldn't find it. Well, I found it today:

The little fucker was living inside of the top of my drill press. Got to hand it to him (or her, no idea) that was a pretty protected space. At least until someone went to turn on the drill press.

So, anyway, brief break to stop, clean out the top of the drill press and inspect for any chew marks or other damage, and then I drilled the holes.

I measured the screws that go into the cage nuts on the rack. They all seem to be 1/4". In order to give myself a little tolerance to play with and make everything fit, I decided to go up a size to 3/8".

This wound up looking pretty good:

And of course, I had to do a test fit on the back of the rack, to make sure they will work:

More to come...


I also did some thinking. Phase 1 of this project is to just move the workstation into the rack, but phase 2 involves building the new game machine.

My local MicroCenter has both 9800x3d's and 9950x3d's in stock now. They also have the motherboard I think I have settled on for that build. The MSI x870e MAG Tomahawk.

I have decided on this one because it both has a 4x PCIe slot off the chipset I can use for a 10gig NIC and it is one of the few x870 or x870e motherboards that allow you to disable the 40Gbit USB ports and instead use those 4x Gen5 lanes off the CPU for a second CPU connected m.2 port.

I don't know if power stages and overclocking capability is any good (or if that stuff even matters anymore these days) but it seems like it is in the direction of what I want.

Thing is, this is not an official MicroCenter bundle. But the last time I bought a bundle from MicroCenter, they allowed me to customize it a bit and still gave me a bundle discount. I'm going to see if they let me do that with these.

Things I am not yet settled on:

9800x3d vs 9950x3d:
They seem to trade blows in games, which makes me think that the cheaper one (9800x3d) would be the smartest choice. That said, I could get a 9950x3d and just disable the non-X3d CCD, and essentially get a higher clocked (5.7ghz vs. 5.5Ghz ) 9800x3d.... And this would give me more flexibility. If I ever need to use it for anything that requires more cores, I can just enable them again...

In the past I wouldn't have cared this much about the CPU as I game at 4k, but the title I want to enjoy when this is all done is Stalker2, and it is crazy heavy on the CPU. If I don't wan' to use frame gen (which I don't) even the 9950x3d may not avoid minimum drops under 60fps. I generally see the 9800xrd getting better average frame rates, but the 9950x3d often gets better 0.1% minimums, which might be significant.

The 9950x3d is - of course - about $200 more expensive, but after the money I am dumping into this project, what's another $200? CPU's just aren't the significant part of a build (cost wise) like they once were...

Which RAM to buy:
I want to try for DDR5-6400 with a 1:1 MCLK:UCLK ratio, but this will involve a mild overclock of the uncore parts of the chip, which may or may not work based on silicon lottery.

I might just go with DDR5-6000 CL30, and try tho overclock it up there, instead of just trying to buy DDR5-6400 directly (in case the uncore overclock fails) Not sure.

Also not sure how much I need. I haven't thought much about RAM use in games for a long time. My workstation has had 64GB in it since 2014, so I am not even sure how much RAM a game-centric machine might need.

16GB is probably a bare minimum in 2025, but you can't even buy 8GB modules in DDR5. 32GB (2x16) is probably the standard recommendation right now. Not sure if I want to step it up to 48GB (2x24) especially if that makes me sacrifice latency.

Anyway, the reason I mention this is, I am not quite ready for phase 2 yet, and I am reticent about buying things too early and then not having a chance to test them before my return window (15 days at MicroCenter) closes. But on the flip side, it would really suck if by the time I got to phase 2, they were sold out...

So, I have some thinking to do about which RAM and which CPU I want, but I may just drive to MicroCenter and do a defensive Phase 2 buy in the next few days.

Appreciate anyones thoughts on the choices above.

 

[Zarathustra[H]]

There's been another little setback. Not directly related to this build, but it has taken some time away from it.

I should have taken pictures when working on this, but I was frustrated and just plain forgot.

You see, ever since the pandemic and hitting my 40's, my sensation of time passing just hasn't been what it used to be.

When I built my current loop in late 2019 right before everything shut down, I was of course aware - as I have been in all of my years of water cooling - that the coolant should be drained, flushed and replaced on occasion.

I usually consider 1-2 years to be appropriate.

Well, last year as I started planning this build it struck me. Holy shit. I haven't changed the coolant in 5 years.

I inspected it. It looked pretty clear. I decided it would have to do for now, as I was already planning the replacement. If the coolant looks this good after 5 years, I'm sure it will be fine for another couple of months.

Well, when I removed the GPU two weeks ago, it was perfectly clean on the inside. No gunk at all. I just gave it a lengthy flush (using my spare pump res from my first custom loop) and decided to think about whether or not I wanted to open and clean the block or not.

I continued using the system with the bypass cooling only the CPU.

Until last night. When I suddenly had no flow across the CPU. My flow meter was reading 0.0.

I did some troubleshooting, and concluded the CPU block was clogged.

So, if you - like me - are careful with loop cleanliness, EK Cryofuel will apparently last just fine for 5 years and ~10 weeks.

After that it rapidly degrades to the point where it completely obstructs a block in just two weeks.

At least that is my experience in my loop with the batches of Cryofuel I had.

It felt stupid to spend time taking everything apart so soon before breaking the system apart, and redoing it, but I need it for work, so it had to happen.

So I immediately repeatedly drained and flushed the loop. Took all sensitive parts out and cleaned them, and then flushed the whole loop with a small amount of dawn (but apparently not a small enough amount to avoid massive amounts of bubbles )

Once I properly flushed out the dawn, I chased it with a a flush of 6% household vinegar diluted with approximately 80% distilled water.

Then drained, filled, flushed, drained filled flushed, over and over and over again until I was happy I had both reasonably cleaned the loop, and gotten enough of the vinegar (and dawn) out of the system.

Then it was time to open the Heatkiller IV Pro block up and give it a good scrubbing.

Except those screws were stuck. Like really stuck. I even broke off an allen wrench trying to open it. I tried with my impact driver and everything. Only thing I accomplished was to strip the screws.

Now I thought I was really screwed. In defeat, I grabbed my Speed-Out kit and drilled and reversed out the screws.

Great. It's apart. I can clean it. I thoroughly washed the block, lubed up the seals, but now couldn't re-assemble it due to not having the eight little screws it needed.

Watercool will sell you replacements, but I don't have time to wait for them to ship from Germany. Modmymods had them in stock. I ordered them, but their shipping wouldn't get them here on a reasonable timeline either.

Grabbed the old calipers, and started counting threads.

Looks like they are 8mm long M3x0.5 screws, with a tapered "flat" style head.

This screw is harder to find than you might think. After a lot of searching, I figured the quickest way to get some screws would be to order them from BoltDepot.com. They had the right screws with the right type of head (just Torx drive instead of internal hex drive).

They apparently have a location about a 45 minute drive from where I live, so I oped to pick them up in person to speed things up. Only downside is that these bolts are stainless. So they are not an exact metal match. Still, they are not in a fluid contacting location, so I think they'll be fine, but I figure when I rebuild the system into the rack-mount case, I will likely have received the exact Watercool screws, and will re-asssemble it using those, just in case.

I reassembled the block. Pressure tested it for leaks. Followed up with an actual water test as a sanity check to make sure it wasn't leaking and then assembled it back into the loop.

I have now filled the loop again and am doing the whole loop leak test. It is mostly distilled water, but I found a bottle of PrimoChill Liquid Utopia that must have come with th etubing I bought, so I decided to use that for a biocide/corrosion inhibitor.

It seems to be working well. Fluid is clean, flow is good. I'm just letting it flush through and get rid of air pockets overnight. Going to try powering it up tomorrow.

Let this be a cautionary tale about keeping up with regular maintenance, and not allowing yourself to get old to the point where time loses all meaning. ( can't effing believe it is 2025...)

 

[Zarathustra[H]]

Another weekend, another limited amount of progress.

I spent some time trying to duplicate my brackets for the second radiator. No pics here. There is nothing new to show off, and I have a little bit of trimming left to do before it will work just right, and I ran out of daylight before i got that done...

I'll say this. I'm an engineer. I am pretty good at being an engineer. But there is a difference between designing a place for a hole on a drawing, and actually making that hole.

I've always had a lot of respect for good fabricators/machinists. I am not one of those. This project has only reinforced my respect for those who are good at it.

So, I am not quite yet at the level of "manufacturing maturity" to be able to claim that I can make "replaceable parts". Each of these two brackets required significant fine tuning to get them to work, and replacing one of the layers with one from the other bracket would result in them just not fitting at all.


I also spent some time working on the C-channels I had originally bought to reinforce the side panel (which ultimately wound up being unnecessary).

I am re-using them as mounting bars for the box that will contain the fan controllers, power supply (and possibly some pumps).

I wound up cutting them down to 19" and then marked the cage nut holes on them:

View attachment 717416

Then I went to drill the holes using my old shitty drill press.

Last winter I had a mouse move in to my shed where I have my main workbench. I eventually wound up killing it with bait. I knew it built a nest in there somewhere, but I just couldn't find it. Well, I found it today:

View attachment 717422

The little fucker was living inside of the top of my drill press. Got to hand it to him (or her, no idea) that was a pretty protected space. At least until someone went to turn on the drill press.

So, anyway, brief break to stop, clean out the top of the drill press and inspect for any chew marks or other damage, and then I drilled the holes.

I measured the screws that go into the cage nuts on the rack. They all seem to be 1/4". In order to give myself a little tolerance to play with and make everything fit, I decided to go up a size to 3/8".

This wound up looking pretty good:

View attachment 717423

And of course, I had to do a test fit on the back of the rack, to make sure they will work:

View attachment 717424

More to come...


I also did some thinking. Phase 1 of this project is to just move the workstation into the rack, but phase 2 involves building the new game machine.

My local MicroCenter has both 9800x3d's and 9950x3d's in stock now. They also have the motherboard I think I have settled on for that build. The MSI x870e MAG Tomahawk.

I have decided on this one because it both has a 4x PCIe slot off the chipset I can use for a 10gig NIC and it is one of the few x870 or x870e motherboards that allow you to disable the 40Gbit USB ports and instead use those 4x Gen5 lanes off the CPU for a second CPU connected m.2 port.

I don't know if power stages and overclocking capability is any good (or if that stuff even matters anymore these days) but it seems like it is in the direction of what I want.

Thing is, this is not an official MicroCenter bundle. But the last time I bought a bundle from MicroCenter, they allowed me to customize it a bit and still gave me a bundle discount. I'm going to see if they let me do that with these.

Things I am not yet settled on:

9800x3d vs 9950x3d:
They seem to trade blows in games, which makes me think that the cheaper one (9800x3d) would be the smartest choice. That said, I could get a 9950x3d and just disable the non-X3d CCD, and essentially get a higher clocked (5.7ghz vs. 5.5Ghz ) 9800x3d.... And this would give me more flexibility. If I ever need to use it for anything that requires more cores, I can just enable them again...

In the past I wouldn't have cared this much about the CPU as I game at 4k, but the title I want to enjoy when this is all done is Stalker2, and it is crazy heavy on the CPU. If I don't wan' to use frame gen (which I don't) even the 9950x3d may not avoid minimum drops under 60fps. I generally see the 9800xrd getting better average frame rates, but the 9950x3d often gets better 0.1% minimums, which might be significant.

The 9950x3d is - of course - about $200 more expensive, but after the money I am dumping into this project, what's another $200? CPU's just aren't the significant part of a build (cost wise) like they once were...

Which RAM to buy:
I want to try for DDR5-6400 with a 1:1 MCLK:UCLK ratio, but this will involve a mild overclock of the uncore parts of the chip, which may or may not work based on silicon lottery.

I might just go with DDR5-6000 CL30, and try tho overclock it up there, instead of just trying to buy DDR5-6400 directly (in case the uncore overclock fails) Not sure.

Also not sure how much I need. I haven't thought much about RAM use in games for a long time. My workstation has had 64GB in it since 2014, so I am not even sure how much RAM a game-centric machine might need.

16GB is probably a bare minimum in 2025, but you can't even buy 8GB modules in DDR5. 32GB (2x16) is probably the standard recommendation right now. Not sure if I want to step it up to 48GB (2x24) especially if that makes me sacrifice latency.

Anyway, the reason I mention this is, I am not quite ready for phase 2 yet, and I am reticent about buying things too early and then not having a chance to test them before my return window (15 days at MicroCenter) closes. But on the flip side, it would really suck if by the time I got to phase 2, they were sold out...

So, I have some thinking to do about which RAM and which CPU I want, but I may just drive to MicroCenter and do a defensive Phase 2 buy in the next few days.

Appreciate anyones thoughts on the choices above.

Figures.

I went to Microcenter last night, and while they still have plenty of 9950x3d's and 9800x3d's, they are out of the MSI MAG x870e MAG Tomahawk motherboard.

They had 50+ of them according to the website when I last checked, but apparently they included them in a bundle, and apparently Microcenter's AMD bundles move.

Well, I don't exactly need it yet. I Still have a bit of work ahead of me. Maybe I'll wait and see if it comes back into stock.

I did confirm that they cannot mix and match bundles though, so I'd have to pay their full price since 9950x3d (or 9800x3d) + MSI MAG x870e MAG Tomahawk was not an official bundle (they must have bundled the motherboard with something else.) so I guess there is no benefit to getting the motherboard straight from them. I can get it somewhere else, and only take advantage of the fact that MicroCenter has the CPU's in stock.

 

[Zarathustra[H]]

As an aside,

For shits and giggles, I decided to check eBay to see if there were any cheap 64C/128T Threadripper 3990x CPU's available. I don't really need one, but I have habit of when a platform starts getting older, just picking up the top CPU and dropping it in for shits and giggles.

I figured since I am going to be rebuilding it anyway, why not now?

And holy crap. No they are not cheap yet. These 6 year old CPU's are still going for between $1,700 and $2000 used on eBay. That is absolutely nuts.

Even the 24C/48T 3960x I have are still going for almost $600... Crazy.

I guess we won't be doing a drop-in upgrade after all

 

[Tsumi]

That's quite the saga.

Regarding loop maintenance- my system with only distilled water and silver killcoil was taken apart 2 years ago after at least 6 years in service without changing the coolant and at least 10 years with only changes of the GPU block (I use QDCs so I don't have to drain the system when changing components, I don't even remember doing a coolant flush but probably did at one point when replacing my motherboard). The Heatkiller 3.0 I took apart only had a little bit of gunk in it, less than 10% of the fin area, everything else was fine.

9950X3D vs 9800X3D- is there any definitive proof that the 9950X3D v-cache CCDs are better binned than the 9800X3D? The 9800X3D is unlocked, unlike the previous X3D chips, and should be able to reach the same clock speeds unless the 9950X3D has better binning. There's a chance that AM5 will get another CPU generation with more cores on a single CCD- that might be worth getting a 9800X3D now to upgrade later.

 

[Zarathustra[H]]

I have a few updates I have not had a chance to write up yet.

It's been a crazy busy time full with responsibilities, so I have not been able to dedicated as much time to the project as I would have liked.

Meanwhile, with all the tariff crap going on, I figured maybe I should start buying the parts for phase 2 (the game build), as I may not be able to get them at a reasonable price (or at all) later. I had been putting this off, as return windows on these items are often limited, and I didn't want to be stuck with a bad (or counterfeit) part and run out of time without testing it, but I guess it is now or never, so I ahve to start pushing hard on buying things.

I have now ordered:
- MSI MAG x870e Tomahawk
- Samsung 9100 Pro 4TB
- A tower heatsink for the SSD (I hear Gen 5 SSD's get hot)
- Two 32GB sticks of CL26 DDR5-6000
- Rackchoice 5U Rackmount case
- Asus 10gig NIC.
- Heatkiller IV Pro All Copper (nickel plated) AM5 CPU Block
- Thermal Grizzly AM5 Contact Frame

but...

Last week there were 25+ 9950X3D's in stock at my local Microcenter. They had been in stock for weeks, so I figured I had plenty of time.

Now, nothing. Nothing at MicroCenter. And nothing from any of the big retailers (Newegg, Amazon, B&H) online.

It's annoying. MicroCenter has some in places that are like a 12+ hour drive away (Illinois, North Carolina, Ohio, etc.) but anything near me is gone.

I'm hoping they get restocked, or I guess I'll have to use a 9800x3d.

As for the CPU block, they were allout of stock here in the U.S. I have no idea what happens if I - in our tariff times - order straight from Germany like I did before. They used to not stop individual small packages in customs at all, but who knows what happens now. I just ordered and will hope for the best.

I went with the MSI motherboard as it is one of a very few x870 or x870e boards that allows you to disable the silly 40gbit USB I will never use and use those lanes for a second CPU lane m.2 slot instead without dropping the GPU down from 16x to 8x

The 9100 Pro will be a secondary SSD mostly for the game library. (The primary boot drive will be the 800GB Optane p5800x)

As for the RAM. I went back and forth between DDR5-6000 and DDR5-6400.

With a little luck it is possible to run FCLK in "Auto" at ~2100 Mhz, and then have UCLK and MCLK in a 1:1 ratio at 3200Mhz, for DDR5-6400, but this is not a guarantee. It depends a little on the silicon lottery. So buying DDR5-6400 seemed like a risk, as I may not be able to hit it 1:1, and then - having paid a little extra for higher clocked RAM - I'd be forced to manually figure out RAM timings when I drop it to 6000, which would be a pain.

Also, the best timings on DDR5-6400 I could find in stock anywhere was CL30, but I found DDR5-6000 at CL26 (which is equivalent to ~CL27.7 at DDR5-6400 clocks), so I figured I'd go with the DDR5-6000 and try to overclock it to DDR5-6400. Even if I have to drop the timings a little to make it stable, I should be ahead of the official DDR5-6400 sticks.

The only CL26 DDR5-6000 sticks I could find in stock were stupid rainbow light varieties. I don't want that nonsense. And I don't want to have to install a special app to disable that nonsense, but I guess I'll never see it as it will be deep inside the rack. Performance is more important.

While I had originally intended to go with the Sliger 3U case with the GPU riser, I realized this would make it impossible for me to use the other PCIe slots, so I opted for a 5U case instead where everything can go in without risers. The RackChoice seemed to be the best priced option, and is reciewed pretty well.

Now for that 9950X3D.

If anyone has any leads from reputable retailers (online, or brick and mortar in the New England area) I'd love to hear about them...

 

[Zarathustra[H]]

Lesson learned.

Most rack cage nuts are not 1/4" 28TPI. The cage nut machine screw may fit in a 1/4" 28TPI test thread at Home Depot, but once you order (and wait for) your specialty screws of exactly the length you want, they will not fit in the cage nuts.

Apparently the cage nuts are M6x1.0. Worth noting here. An M6x1.0 screw looks almost exactly the same as a 1/4" 28TPI screw, but has an approximately 0.4mm smaller diameter, so the 1/4" screw just won't go in.

Apparently they are close enough that the M6 screw will thread in a 1/4" 28TPI hole (albeit not as tightly as designed, this should have been my hint) but vice versa just won't work.

Can you tell which is which?

 

[Tsumi]

It would be great if the world just settled on the metric system. Standard should just become the specialty item for older things and get phased out slowly over time.

 

[vegeta535]

Man by the time you actually finish your builds it is obsolete and time to upgrade lol.

 

[Zarathustra[H]]

Man by the time you actually finish your builds it is obsolete and time to upgrade lol.

The goal here is to spend time getting the cooling system to the point where it is overkill and future proof. And then I can just swap out the blocks with QDC's, so it is "never obsolete".

But yeah, I anticipated this would be slow moving, and it has been slower moving than I originally anticipated.

I have had some stuff going oi in the background I just haven't hand time to update the thread with yet. I'll try to do that in the next couple of days.