Testing started: opinions welcome

[pHaestus]

I have settled upon a course of action that is now set in stone. Most of the design (and for that matter the equipment) is thanks to Bill Adams.

I do not like CPU as heat source, but I want to get waterblocks tested before Socket A is dead. So I am, for now, going to test out waterblocks using the calibrated AMD internal diode/MAX6655. I cannot get the same size error bars as Bill, and I really need to focus efforts on ADC for some of the equipment so I can log everything more effectively. There are a few problems with my testbed, but I am willing at this point to live with them just to get some real results. Here is how I am testing waterblocks:

On the inlet and outlet side I have 3/4" PVC crosses that have 1/2" NPT threads on them. On two opposite sides are hose barbs so liquid flows straight through. On the other two sides are the connections for a YSI 400 series dual linear thermistor and a reducer so that 1/4" OD tubing can be plumbed in for my digital pressure gauge. Here is a pic of one of the crosses:



I am still fiddling with the pressure gauge; I will probably be a while before I get that properly calibrated by messing with it.

My setup has, of course, gotten a bit oversized and out of control:



Yes it is increadibly cobbled together right now and I am not totally satisfied with it. But it seems to be possible to get good numbers from 0.5GPM to 2.5 GPM so I will go with it and make adjustments as I get a chance. Highlights of the picture are the reservoir (that needs to be replaced with a larger one), the Little Giant 3E-12NYS industrial pump, the large heatercore (it's a one pass from a 1970 Blazer), the crosses, the Digitec 5810 thermometers (red LEDs) and the Omega digital pressure gauge on the right, the GPI flowmeter (it's a paddle wheel type), and then the needle valves and gate valve to tweak flow rates. I am using a big 3/4" needle valve, and when it is supposedly fully closed it still allows about 0.8GPM through. Not sure what is the deal with that. I did have the gate valve and the needle valve on parallel channels but I was not able to get down to 0.5 gpm that way. I have a second needle valve that I guess I will try out this evening. The 0.5 GPM measurement is a bit tricky at the moment because I am using the gate valve to adjust (which sucks).

My strategy is to post preliminary results in this thread, and to post updated pictures and setup as it invariably changes over the next few weeks. Consider all numbers and figures here as "in progress" and not necessarily 100% accurate. When I have good data then you will see it in the form of Part2 of Procooling's waterblock roundup. No sense in waiting months to see that though since certain testing is of immediate interest (Maxe3 vs. Maze4?).

I have a die simulator on the way (just the copper hunk) from Andrew at www.millennium-thermal.com . This is an identical die sim to the one that JoeC at overclockers uses for Athlons. I don't care too much for the big copper socket on this simulator, and I may try to have it milled off and then just cover the top and sides with the phenolic resin that Bill uses. I will need to either get some Minco heaters or a cartridge heater as well pretty soon. When I have that die sim ready, I will retest the socket a blocks on it. I am going to do ALL P4 and Opteron blocks on a die sim from the start and just use heatspreaders (anyone have a P4 heatspreader?) and the different mounting pressure specs to simulate the different chips.

[pHaestus]

The little giant industrial pump must be putting out enough EMI to interfere with the CPU's diode reading. With the pump running, the diode temperature is 41C (no power to system) while water is 27.03C for inlet and outlet of waterblock. Turn off the pump and water temps stay the same but diode temperatures drop to 26.9C (I haven't been running the system that long though and would expect the diode temp to rise to water temp over a long enough time.

So how can I shield this pump? I tried moving it away as far as possible but no go. I also tried plugging the pump and the PSU that is powering the diode reader into separate surge protectors. Made no difference.

Edit: ANother observation is that if I turn off the surge protector to shut off pump then the interference remains. I have to pull the pump's plug out of the socket for the temperature to return to normal. Weird

[hara]

I may be speaking out of my ass, but, how about housing it in a brass tank, maybe even submerging it.

[Les]

With only a rudimentary knowledge of electrics cannot offer any help. However would remind that I had problems with pump interference when the cpu was unpowered -
http://forum.oc-forums.com/vb/showth...0&pagenumber=2

[pHaestus]

I just swapped pumps and am seeing if the problem repeats itself with a different one. I held an L20 right at the diode reader and the CPU and didnt notice any interference.

[BillA]

is the pump case independently grounded ?
(separate from the white leg of the AC line)

EDIT:
emi shielding is done best with a TOTALLY closed box of mu metal
- so I'm told

[pHaestus]

OK it is definitely an issue just with the 3E-12NYS; I just plumbed in a Laguna pump and temp of unpowered CPU is right where it should be. Unfortunately 1.5 GPM is all it can muster wide open. Ok another odd observation is that the Little Giant doesn't cause interference when it is plugged in and close to the testbed but not plumbed into the loop. What the hell? One thought; I didn't have the motherboard in a grounded case before and now I do. I will put the Little Giant back in again now I guess.

I tried putting a ground wire from the pump to a running PSU cover that was on a separate protector. No difference. Why would it cause problem when in the loop but not when sitting right next to the system?

[ezlid]

This might work for a simple pump EMI shield. Get a large coffee can and cut openigs for the hoses and power lead and put a gound wire to the case or something tied into earth ground. If it doesn't work, at least you didn't spend any $$$ on it. If it works you can make a permanent metal enclosure.

Also you might want to get some Romex shielding from Home Depot or Lowes. It's the metal flexible conduit used in homes. Run your pump AC cord in the Romex and it will reduce EMI from the cord. Gound it to the same point as the coffee can.

Finally try the proven fix everything tool. Duct Tape! I saw a roll in your pic.

Hope one of these works and thanks for the work you are doing.

[Les]

An ungrounded case was the root of my problem.

[BillA]

measure the resistance between the coolant and the motor leads (AC wires)
seems that something is shorted

[ezlid]

You're getting replies before I can even finish one. At least people are interested.

This one might sound off the wall but - Try grounding the water.

Solder a ground wire to one or more of the brass barbs. Again, use a single point ground. Oh and ground the rad, too.

[joemac]

Your pump is interfering only with the thermal diode? I doubt that (hold the flames) let me explain.

Your CPU has millions of transistors. Well I don’t know for sure but I bet the thermal diode is not really a diode but a transistor itself. To say that the pump is only interfering with this one unit seems highly unlikely as I am sure there are hundreds more of the same components on the same C.P.U, MB etc.

(To the PhD’s out there I know a transistor is an array of diode in the most basic sense, but I am trying to keep it simple).

[pHaestus]

When the system is not powered then only that single transistor is relevant.

Also I managed to somehow kill the diode reading altogether from that chip now. I pulled it out and looked it over but can't find anything wrong with it. Nonetheless it always reads -1C now on either of the MAX6655's external ports. A transistor in diode mode works fine on both. Guess I should see if the CPU itself is dead but honestly I don't much care if it still functions as long as the diode isnt working. Guess I can desolder the wires and swap it into my 8k3a and see if it reports a gibberish temperature or not.

Crap. Oh well I am getting pretty adept at the CPU pin soldering and calibration procedure now at least.

[pHaestus]

Just ordered another 1700+ TBredB from newegg; I will see if I can find the time to put the 1600+ into my main system and then solder wires onto the TBredB this week.

[pHaestus]

Well I just tested the CPU with a DMM in diode tester mode and the diode isnt working. I don't know exactly what could have happened to cause this; the system the CPU was in wasn't even powered. It must have shorted out somehow but not sure exactly why.

Oh well I guess I can start over this week soldering onto and calibrating my 1700+ TBredB. Joy.

[joemac]

Two things that that come to my mind are:
1: test equipment
2: too much heat when soldering,

[pHaestus]

It has been running fine for over a month now so I don;t think it had anything to do with the soldering. I guess it must have had something to do with applying current to it while I had a short in the loop (as discussed above) but I really dont know if that is reasonable or not.

[Groth]

Static discharge? That carpet looks dangerous to me.

Do you know if your eval. board has any isolation provisions? A floating ground might give your reader the voltage difference needed to punch through the diode's insulation.

[Since87]

It would be a good idea to check the resistance between each diode pin and a Vcc pin and a Ground pin. This might give some indication of how the damage occurred, and consequently the cause. Knowing whether the CPU otherwise works could also help in knowing what happened.

Have you checked all the relevant wall outlets with one of those plug in testers that tell you that the wiring is right? I'd check the surge suppressor outlets as well. Incorrect outlet wiring could be consistent with the power strip being off and the pump still conducting a voltage into the water.

Bill's guess, "measure the resistance between the coolant and the motor leads (AC wires) seems that something is shorted", seems the most plausible explanation to me at the moment. Some precautions I'd suggest:

1. Ezlid's suggestion, "Solder a ground wire to one or more of the brass barbs." is a good one. I would suggest connecting to a brass or copper fitting on each side of the pump. (intake and outlet) Tie those two fittings to the ground of the PC which has the temp diode in it. If the pump manages to electrify the water in the impeller housing, the grounded fittings will keep current from getting to the CPU.

2. Dedicate a surge supressor to the pump alone. Pumps can put out nasty spikes when they are powered off. A quality surge suppressor will be nearly as good at preventing outgoing spikes as it is at preventing incoming spikes. I had a Danner Mag5 that would cause my printer to reboot every time I unplugged it. That Little Giant may be much worse.

3. I'd guess that a non magdrive pump, with a seal between the water and circuitry that can fail, would be required by law to have a GFCI circuit built in, but I don't know this. Even if it does have one built in, it may not be working. Probably a good idea to add an external one.

The thought that initially occurred to me, was that that you most likely have a big loop in your setup that can easily generate signals in response to the pump's magnetic field. (This isn't consistent with the pump being plugged in but powered off, and still affecting the temp measurement. It could result in bad measurements and damaged components when the pump is running though.)

The loop is:

diode->
diode reader->
ground of PC connected to diode reader->
ground of PC containing 'metered' CPU->
'metered CPU'->
diode

The best way to 'collapse' this loop, is to cut the earth ground connection to the PC the diode reader is connected to. Then connect the chassis of the PC the diode reader is connected to, to the chassis of the PC the diode is in. This ground connection should be routed alongside the diode reader cabling. Ideally twisted with it. To be safe, both PC's should be plugged into the same surge suppressor. (If a picture would be helpful let me know.)

The issue here, is that a changing magnetic field can induce a voltage in that loop, and that voltage would be applied between the temperature measurement diode, and the rest of the CPU that the diode is in. I'd guess that it may not take much voltage between the diode and the rest of the CPU to do damage. Generally the diode reader would be on the same motherboard as the CPU and inducing a damaging voltage between diode and CPU would take an enormous magnetic pulse. A big loop like you probably have, could easily pick up a voltage from the magnetic field of a powerful pump.

[pHaestus]

I am really tempted to skip the "soldering on CPU" step and instead just pump water held at different temperatures through the cooling loop with an unpowered CPU to generate a calibration. That would certainly make the inevitable death of CPUs much more easy to deal with...

[BillA]

simple is always better

do keep fussing with the pump
as I also have 2 of them, I'm quite interested
BTW, my pump (and chiller) have always been run off a separate circuit breaker at the mains

[pHaestus]

You know the mostly likely explanation is IMO that the outlet is reversed. Lord knows everything else in the finished basement is done half assed and incorrectly. I will get a plug checker tonight

[Groth]

pump water held at different temperatures through the cooling loop with an unpowered CPU to generate a calibration

That's how I did mine. Ice and a 500 watt immersion heater; insulated socket and insulated water block. The bad part is, you have to repeat it everytime you pull out the CPU.

The plug checkers with the three lights are handy, but limited. It'll tell you if you have an open ground or the hot in the wrong spot. It won't say if the neutral and ground line are shorted, as is common in older houses.

With nothing active on the circuit, use your DMM to measure AC between the neutral and ground lines. It shoud be zero. Now put a large load on that outlet (hair driers are perfect), and measure again. If zero, you need a real ground wire. If ~0.5 V , fine. If 1.0 V or more, your wiring is correct, but sucks.

If you are using two or more outlets for your set up, also check between the grounds of the outlets. Should be zero with or without a load.

I've seen a mis-wired ground and floating neutral produce a 40V shower...

[KnightElite]

Originally posted by pHaestus
Well I just tested the CPU with a DMM in diode tester mode and the diode isnt working. I don't know exactly what could have happened to cause this; the system the CPU was in wasn't even powered. It must have shorted out somehow but not sure exactly why.

Oh well I guess I can start over this week soldering onto and calibrating my 1700+ TBredB. Joy.

It decided it wanted to emulate my old 1600+ .

Anyway... looking good pH. It has definitely progressed somewhat since the last time I saw it .

[Ice Czar]

Ive got 2x3E-12NYS as well, and one of the things you might want to check is that the oil that fills the motor housing isnt contaminated with water