Intel Xeon 3647 woes.

BrainEater

[H]ard|Gawd
Joined
Jul 21, 2004
Messages
1,183
Ugh....

As some of you may know ,in the last year or two , I have put together a watercooled dual Xeon rig ;
see https://hardforum.com/threads/project-snowballed.1950723/

The specs are there.
-----

I have a problem..
The machine ran really well , until a couple weeks after I installed the waterblocks.
I have lost 2 channels of my CPU2 ram.

I've switched modules around etc....The 2 channels in question show up blank in BIOS.
I know der8auer had the same issue , when he installed waterblocks...
------

If I was gonna guess , I'd say the waterblocks are not rigid enough, and cpu pins are not connecting. The socket p setup is not cool.

Thoughts ?
 
Last edited:
If I was gonna guess , I'd say the waterblocks are not rigid enough, and cpu pins are not connecting. The socket p setup is not cool.

You are probably correct.
Maybe not enough socket pressure (or even too much..)

Can you try to go back to air cooling?
 
That's definitely on the list.
I think I will try to reseat CPU2 and see what happens.

My big concern is that the northbridge is damaged....from what I understand , der8auer never did get those two channels back....
 
Yes absolutely.
All of my memory sticks test out fine.It's the two dimm slots that are not working.
 
Ok.
After some playing around and some research, I am not gonna fuck with it right now , it *still* crushes 4k60 HEVC ......that's all I care about.
All of the ram is fine.It appears to be the cpu/socket mating issue....Everything still works, It's just not fully connecting.

I have 10 of 12 memory channels, I'll call that a win (for the moment)
-------

My next step is to find a dead socket P mobo and see if I can't make a new socket......get rid of the 2 stupid alignment pins and put proper bolts in.

Any thoughts ?
I'm honestly scared to mess with the waterblocks as they sit now.

:ROFLMAO:
 
Ok, self Necro here.
------

I do know this was a common problem with the socket 3647 chips when watercooled. The CPU pins didn't all connect to the ram slots without a super rigid cooler.
Machine still runs great (typing on it now)
--
This machine is now back on air, oem coolers. The ram slots are still missing =/

So I'd like to repair this.

I'm curious where to start.
It's CPU, mobo or ram. (BIOS corruption ? been redone but..)
I'm going to assume the Xeons are probably fine, but where would y'all start ?
 
Last edited:
Ok, self Necro here.
------

I do know this was a common problem with the socket 3647 chips when watercooled. The CPU pins didn't all connect to the ram slots without a super rigid cooler.
Machine still runs great (typing on it now)
--
This machine is now back on air, oem coolers. The ram slots are still missing =/

So I'd like to repair this.

I'm curious where to start.
It's CPU, mobo or ram. (BIOS corruption ? been redone but..)
I'm going to assume the Xeons are probably fine, but where would y'all start ?

Reseat the CPU, and while in there check for bent pins.
 
Yep, that was my first thought too.
Been done, all pins are flawless so are all the lands.
-
This machine runs flawlessly except for the missing ram. Every CPU test runs perfectly, , except for the missing ram. 4k video encode it crushes, Except for the missing ram.
I've swapped the modules around, no joy, RAM seems fine.

I don't think there is an actual problem to fix right now, it's some damage I need to fix from the past.
If you were a high end supermicro Xeon rig and did'nt have all your ram pins connected, what would you kill ?
hehe
I'm definitly leaning towards motherboard but, $$
 
BrainEater , are you using plastic or metal or motherboard standoffs?

I've an AMD epyc system with 8 channel memory (full populated) where there was one missing ram. Found out that the motherboard standoff was touching a pin where that missing ram channel was, so I put an insulator tape (e.g. kapton tape) to prevent any electrical shorting/interference. Mine is air cooled, so it is relatively straight forward to remove the motherboard off the case and put them on a table to test the system first before putting it back. I do agree that it is more difficult with water-cooling system.

Here is one thread in question discussing this issue on epyc system: https://www.reddit.com/r/homelab/comments/zi0xph/epyc_not_detecting_all_ram/

Can you perhaps locate where the motherboard standoffs are in relation to the RAM pins on the back of the motherboard first if you decide to remove the system off the PC case?
 
  • Like
Reactions: bwang
like this
BrainEater , are you using plastic or metal or motherboard standoffs?

I've an AMD epyc system with 8 channel memory (full populated) where there was one missing ram. Found out that the motherboard standoff was touching a pin where that missing ram channel was, so I put an insulator tape (e.g. kapton tape) to prevent any electrical shorting/interference. Mine is air cooled, so it is relatively straight forward to remove the motherboard off the case and put them on a table to test the system first before putting it back. I do agree that it is more difficult with water-cooling system.

Here is one thread in question discussing this issue on epyc system: https://www.reddit.com/r/homelab/comments/zi0xph/epyc_not_detecting_all_ram/

Can you perhaps locate where the motherboard standoffs are in relation to the RAM pins on the back of the motherboard first if you decide to remove the system off the PC case?
ah yes, a real classic on Supermicros. the middle standoff on the top row shorts out the topmost DIMM slot.
 
Never heard that one before.
-----
It's not the standoffs.

The machine ran 100% until I put on the waterblocks about a year later....

This is a heatsink rigidity issue. I know Der8auer had the exact same issues when he was doing socket P(3647) Xeon OC runs.
Now that I'm back on OEM heatsinks, I'd like the RAM back.
 
Last edited:
Back
Top