Critique my updated Cascade rig

[Ruiner]

That's one of 2Fresh's Caprice cores, with twin delta 120's (not very powerful, but silent at 7v) and an eheim 1048.
I haven't measured water temps yet, but my cpu temp folding is 50C per the horribly inaccurate in socket diode. Specs in sig.

[AntiBling]

Whats the copper thing in the line between the pump and block for? If its a union, why? Seems like that needs to go unless its in there for a reason.

One thing... the duct tape. If you dont care how it looks, thats cool, but duct tape dries up eventually. You could retape it with aluminum tape at least, if you really dont want to build/buy a shroud. That would last longer. I'd use a real shroud but thats just me. How far are the fans from the rad?

Another thing... Wont air get caught in the radiator with it mounted that way? That may be your biggest problem.

[Ruiner]

Originally Posted by AntiBling

Whats the copper thing in the line between the pump and block for? If its a union, why? Seems like that needs to go unless its in there for a reason.

One thing... the duct tape. If you dont care how it looks, thats cool, but duct tape dries up eventually. You could retape it with aluminum tape at least, if you really dont want to build/buy a shroud. That would last longer. I'd use a real shroud but thats just me. How far are the fans from the rad?

Another thing... Wont air get caught in the radiator with it mounted that way? That may be your biggest problem.

The copper union is to allow a gentler curve to the cascade. The lumen is no narrower than the other barbs in the system. That's cheap vinyl tubing and doesn't bend in 2 directions well.
The duct tape is hopefully temporary. The shrouds that 2fresh sells cover the whole rad (including the barbs) and won't fit in a setup like this. He may make new ones or I could grab some sheet aluminum and snip some myself. The fans are 1" or so from the rad.
I was wondering about the possibility of an air trap in the rad. I could always drill a tiny bleed hole at the top and cover it with a short sheet metal screw and washer. How much of an issue is that?

[AntiBling]

I guess its not an issue if you dont mind bleeding it as necessary. Id rather engineer it so it doesnt need bleeding. You could put a bleed on it in many different ways, but when you can just turn the rad over and replace a few tubes it hardly seems worth it. BTW I also started out using the vinyl tubing, but after screwing around with it I switched to clearflex from Cooltechnica. Sometimes saving money on the tubing is more trouble than its worth.

Hows this...

Turn the rad over.

Turn the pump around so the res inlet is attached to the GPU outlet.

Pump outlet to rad inlet.

Rad outlet to CPU inlet.

You may not need your union that way either even with vinyl tubes.

[Cathar]

Biggest issue to me there seems to be a total lack of adequate air-flow intake into the radiator.

[Ruiner]

The metal frame of the front of the case is cut to the size of the heater core. The plastic trim/fairing over that is pretty well vented, so I think inflow is ok. FWIW, socket diode temps are 3C lower with the fans 12v.
I have to measure the water temps next.

[Varsis]

what worries me themost is my little maze2 and blackice PRIME, remember, the OLD ONES. gets 46c load when i have the fans on SUPER low voltage to the point where they almost wont turn on.
and i have 29 ta 30c ambient air.
gotta love hawaii......rawr
seems to me that the heater core isn't getting enough air, or is getting shit for flow. or has a HUGE airpocket in it.
i'd vote on the last of them, seeing as its trying to suck water thru it, stead of push some thru it, impeller type pumps are known for not being able to suck ANYthing, though they do push rather alright.

[Cathar]

Originally Posted by Varsis

what worries me themost is my little maze2 and blackice PRIME, remember, the OLD ONES. gets 46c load .

46C, measured how?

Remember that no two boards report anyting remotely alike, even the same model board from the same manufacturer, and believe me, I've been through three Abit NF7-S boards and they each reported differently. Even various BIOS updates will alter the reported temperatures.

Heck, on my latest test setup I see ~51C load temperatures on my AthlonXP CPU. Mind you, that's running BurnK7 at 2850MHz/2.15v Vcore, so the heat load there is somewhat higher than "average".

Temperature comparisons are meaningless between different systems, and depending on various circumstances, even meaningless between air/water setups on the same system. May as well try to pick what the weather will be on a particular day next month by looking inside your toilet bowl.

[j813]

Comparing Air to Watercooling. Even if it's in the same PC, parts, etc.?

[Cathar]

Originally Posted by j813

Comparing Air to Watercooling. Even if it's in the same PC, parts, etc.?

Yes. Most motherboard's CPU temperature sensors are significantly affected by the pattern of air-flow coming off a HSF, and with water-cooling most of that air-flow is gone.

[Varsis]

Originally Posted by Cathar

46C, measured how?

Remember that no two boards report anyting remotely alike, even the same model board from the same manufacturer, and believe me, I've been through three Abit NF7-S boards and they each reported differently. Even various BIOS updates will alter the reported temperatures.

Cathar, got a little thermal probe shoved right up next to the core with a bunch of goop
. seems to be pretty acurate with in .1c

1.8vcore.
runs 48c full load full fans with 1.85v

[Cathar]

Originally Posted by Varsis

Cathar, got a little thermal probe shoved right up next to the core with a bunch of goop
. seems to be pretty acurate with in .1c

1.8vcore.
runs 48c full load full fans with 1.85v

Sorry, that's about the worst way to measure CPU temps. Completely non-repeatable with remounts as removing the CPU to clean it results in minor variations in contact. It's very much subject to air-flow patterns, what's going on inside the waterblock above the probe, and the edge of the core is perhaps the coolest bit to measure.

Find out what your water temps are. I suspect that they may be a lot higher than is desirable, especially if from 7V to 12V is yielding a 3C drop in water temps.

Really need to focus on the temp deltas though: CPU die temp minus water temp, and water temp minus radiator intake temp. That's about the only real way to figure out what's going with the block, as opposed to what's going on with the radiator.

[Varsis]

N how do you suppose one should measure acurate core temps?
a prob under neeth wont do it, drilling a hole in the waterblock and putting one right over the core would kill cooling efficiency.

[Cathar]

Originally Posted by Varsis

N how do you suppose one should measure acurate core temps?
a prob under neeth wont do it, drilling a hole in the waterblock and putting one right over the core would kill cooling efficiency.

With an external probe, there's no such thing as "accurate". You can achieve good consistency though with an external probe, just so long as you don't expect it to be accurate.

By that I mean that it is possible to setup an external thermal probe in such a way that it is able to reliably detect relative differences (with respect to itself) between different cooling scenarios.

Sticking a thermal probe "out there" on the edge where it is subject to both variations in air-flow and thermal contact with the CPU.

An under-the-socket probe is much maligned, but perhaps about the only way to get repeatable external results, so long as adequate measures are taken, such as sticking insulation around it in the socket, around the socket, and behind the motherboard, ensuring it directly touches directly behind the CPU with thermal paste.

That won't give you an accurate measurement, but it will give you a fairly repeatable one, which is the best one can hope for.

Probes hanging out "in the wind" in even very minor ways can be compromised very easily by any form of air-flow.

[Ruiner]

update: here's a pic of the rad behind the plastic facia.

I put a bleed screw in the top of the rad to get any air out, and made doubly sure by tipping the case on it's side. I'm pretty sure it's bled.



With the facia in place and the fans at 6v, water temp is 7C warmer than air. At 12v the gradient is 6C.
Removing the fascia drops the gradient to 2C at 12V and 3C at 6v. Thus the facia is costing me about 4C.....time to mod it for better flow. Cathar was right.

The water/socket diode temp gradient is 16C. Not accurate, but a bit high. I may have to reseat and/or use something better than artic alumina.

[alexwai]

Originally Posted by Cathar

Really need to focus on the temp deltas though: CPU die temp minus water temp, and water temp minus radiator intake temp. That's about the only real way to figure out what's going with the block, as opposed to what's going on with the radiator.

Cathar

Can you tell me more about the delta between "CPU die temp minus water temp, and water temp minus radiator intake temp"?

What does it imply if, let say, the delta T between CPU die and water, and water and radiator intake are both small? I tendto think that it has to be somethings to do with the radiator's efficiency. Is it correct to believe that the radiator is not doing its do well?



Thanks

[Etacovda]

I have a thermal probe mounted next to my cpu in the manner you described; its not 'accurate' but onto itself, its not too bad. My resolution is low, however; but if you're talking about what should be a huge difference then its fairly acceptable way to do things. One thing i have learnt from here (as said above, and personal experience) no two set ups are comparable. Theres too many variables.

Granted, my set up is external (flat test bed, no case etc) with little air flow though, and I have epoxied and taped the sensor onto the cpu with thermal paste too. Theres very little external interference with the probe, barr perhaps mosfet heat. My temps seem to line up with what graphs i have seen around here; sure, I dont expect them to be accurate, persay, but relative to an extent. I'm only looking at delta T's at this stage anyway. If I had the money to spend, I'd be buying the set up that Phaetus recommended to me, but at the moment, this will do.

I guess its how far you want to take it... the better question is, whats your ambient temps? water temps? what temps do you expect to be getting?

[Tempus]

The whole issues of precision, accuracy, and repeatablitily is one thats been plaguing this group since the beginning. Its seems nearly impossible to get decent numbers that truly stack up against someone else's.

I will say that I find decent repeatability depending on what changing I'm measuring. But, since I don't have a "true" number to compare my measurements against, all I'm really saying is that most of my measurements are "in the zone" of my expectations. So, yeah, not exactly scientific.


It there a decent solution? Is it possible to build a GOOD socket adapter with accurate thermal monitoring that is a cut or 10 above the crap on the diode? How effective is water temp monitoring given the turbulence and variations in flow still just give a ballpark on the actual temps?


*sigh* Glad I'm a programmer. I'd spend waaayyy to long trying to find some decent solution.

[Ruiner]

More updates:
Ventilated front fascia and new fan controller/temp monitor, plus flipped radiator (up/down).

[pHaestus]

Looks better. Consider using some aluminum tape to make a 1-2" shroud for fans instead of duct taping fans directly on the radiator. Will lower noise and improve cooling

[Ruiner]

The modded fascia got my dT air/water to 4C with the fans at 7v.

The fans are held onto the rad (with screws right between the fins), spaced 1" away (those are only 2.5cm thick fans). It's secure but, yeah the duct tape looks like crap. My squeeze bottle res is pretty ghetto too. I'll be putting a cold cathode between the rad and the fascia soon for some bling.

I also swapped the 80mm mosfet and 50mm NB fans for one 92mm@7v. It works well and is quiet.

I've noticed that at Vcores above 1.85, the PSU gets fairly hot with the fan turned down as quiet as I prefer. I've dropped my Vcore/speed to 1.85/2365MHz. The power management on this cheapo Shuttle board is not the greatest.

I'm curious as to what the stronger/louder Swifty pump might get me. 2C cooler perhaps?

[satanicoo]

Not even close. More like 0.1C acording to PHs WB and Cathar pump's pressure tests.
Maybe more overclock.

[Ruiner]

Originally Posted by satanicoo

Not even close. More like 0.1C acording to PHs WB and Cathar pump's pressure tests.
Maybe more overclock.

That's with just the one block, no?
What about with a more restrictive rig, i.e. with the GPU block in the loop?

I think I'd prefer the quieter 1048, though.

Current plans are to flip the 120mm PSU fan to the outside/bottom of the PSU case, to improve airflow at low fan speeds. It's the hottest/loudest thing in the case now, and the fan speed creeps up under thermistor control. As is, the PSU sinks are in a dead space, and too close to the fan.

I may also fab a proper shroud, or spring for a pair of the 6x6 plastic ones.

[Ruiner]

I dug around for pressure drop info and block and pump curves and did some not-so-WAGing.
I found Cathar's flow estimates for the cascade in a 1/2" system with a pro-core: 1.15 gpm for eheim 1048 and 1.6 for mcp600.
His estimated pressure drop for the Cascade at 1gph of 1psi.
JoeC's measured pressure drop for the mcw50 of .26psi

Using that info, I'm guessing not more than 1C improvement with a stronger (greater max head but louder) mcp650, assuming that the caprice core has pressure drop similar to the procore. Is this reasonable?

The most significant weakness would therefore be the weak axial fans I'm using with such a thick core. If I could install a properly shrouded blower at the sound levels I want, it might help.

Does anyone see reason for me to reroute the tubing (from the most recent pic above) to pump-rad-cascade-gpu-res ?

[Ruiner]

I decided to upgrade to the muffled D4 from danger den after all, and changed the hose routing a bit, with a T instead of the res.



Still testing stability, but 2479MHz and 1.95Vcore (set in bios, MBM reports 1.9) seems ok in gaming, and at the same (reported by socket) temps that 2365 gave me before at 1.90Vcore.

The eheim is for sale if anyone wants it.