Question for the Gurus of Cool(ing)

[BorgBob]

I want to use a Swiftech MCW-50T pelted water block to cool my ATI 9700pro. Problem is, I want it to cool the RAM as well. Why shouldn't the RAM get as much benefit as the core? BGA heatsinks simply won't do.

My thought was on whether some sort of heat spreader couldn't be fashioned. Initially, I was considering buying a Tyan Tachyon board, ripping the fan off, and attaching the waterblock to the RAM + core heat spreader that's already there. However, I read about how that spreader doesn't make that great a contact with the RAM and decided to blow it off.

Here is the question. Is there some way to take sheet copper, bend it to contact RAM and core, mount the water block so that core and block sandwich this sheet, and have it work out?

I'm thinking of copper that is flexible, and off clamping it while the glue on the RAMs dries as a way of ensuring good RAM contact.

But will the copper carry the heat well enough over that much distance, or will ramsinks work just as well and be less trouble?

Is there some similar technology that could be made to work? For example, would soldering a copper pipe with water help move the heat?

Any ideas from you cooling scientists out there?

[jaydee]

From all the stuff I read cooling the ram on a vid card does little to no good for overclocking. Probably just wasteing your time.

[nOv1c3]

You plan on using pelts on your CPU ??

I wouldnt mix and match ... If you go with pelts you need to do your whole system ..If you just use a pelts on your GPU your CPU temps are going to go sky high

[gmat]

Originally posted by BorgBob
will ramsinks work just as well and be less trouble?

Yes.

If you insist on uber-cooling your GFX RAM, just for the sake of it, then do a search here on posts by morphling1 (in this forum), he made a special baseplate that both touch the GPU and the RAM around it. Or look at BladeRunner's creations, on his site or here (in this forum as well).

[BorgBob]

Originally posted by nOv1c3
You plan on using pelts on your CPU ??

I wouldnt mix and match ... If you go with pelts you need to do your whole system ..If you just use a pelts on your GPU your CPU temps are going to go sky high

This is not an issue. In fact, I've seen a couple of reviews where the pelted GPU was paired with an ordinary water block on the CPU and while CPU temps increased a couple degrees, they did not go sky high.

I'm waiting on a CPU pelt for a couple of things. First, I'm not sure there's room for a Meanwell or equivalent power supply in my case--it's pretty full! Second, I just finished a major project to build a block from scratch that I'm very happy with:



Yup, it's a Swiftie air-cooler encased in acrylic. It seems to be fairly insensitive to upping the voltage on the CPU--generally temps are about 33C idle, 35C load (i.e. Folding + RTCW for a few hours). It's only got one 120mm fan on the core, and not a high CFM at that. I figure I can always up the fan action if I need to lose more heat.

My plan is to wait for the Hammers, buy the Swiftie air + pelt unit, and encase one of those when I upgrade to Hammer. I think we're too late in the current chip's life cycle to be worth my engaging in that right now.

As to the efficacy of further RAM cooling, I love bladerunner and morphling's work, but don't want to undertake that much effort if I can get a lot of bang for less bucks. Again, who knows how long this silly ATI card will be current? I may get the Swiftie block relocated to a new GPU, but I doubt a custom made block for the ram will be transplantable.

It sounds like this is one of those deals where I would just have to try it and see. I think the jury is out on the benefits. It's true, most of the OC's on RAM have been poor, but most of them I've seen didn't voltmod either.

[psychofunk]

Nice, I have the same sink and right now it's got a delta on it, the pump is on the way, but I really like that idea. If you turn it up how low do the temps go? How did you do it? Any suggestion on how I might do that?

[nOv1c3]

Originally posted by BorgBob
This is not an issue. In fact, I've seen a couple of reviews where the pelted GPU was paired with an ordinary water block on the CPU and while CPU temps increased a couple degrees, they did not go sky high.

I would think it would increased more than just a few degrees ..You are adding 200+ more watts of heat to your water cooling loop .........But i could be wrong



Oh btw great job on the WB and rad

[Cathar]

I have personally found that it helps to not have any ramsinks at all when using a strong peltier on the GPU core.

Why? Because the ram chips are located very close to the GPU core, and are connected by tons of metallic traces. These conduct enough each to chill the memory down quite substantially.

For example, I did an experiment on a GF3Ti200 with a water-cooled 80W peltier. The card at best would do 240/540 stable using air-cooling alone. With the peltier on top, the card was hitting 280/590 core/mem, and the ram chips, which had no heatsinks on them, were cool to the touch, even under full load. Sticking heatsinks on the ram chips is the wrong way to do things. You're now giving extra surface area for the ambient air to heat the ram chips back up again. If anything, you want to be insulating them from the air, and let the peltier do its job by drawing the heat away via conduction through the card PCB.

The extra heat that the peltier adds to the water system is just something that needs to be dissipated by the radiator. An 80W peltier at 12v adds something like 120W of extra heat to the system, which for a largish radiator like I use, means maybe 2-3C higher water temps (and therefore CPU temps) depending on how hard I run the radiators fans.

[BorgBob]

Cathar speaks!

And with fascinating insight too, I might add. The idea that the traces mean you shouldn't even have a RAM sink had (obviously) not occured to me.

With this thought in mind, my life is simplified, I'll just get the block and rock on!

psycho, if you really want the whole sordid story on the block, you can find it: here

I have to say, it was a LOT of work. I love it, but mostly because it looks very cool. I don't think my temps are particularly spectacular, but I'm happy with them.

Enjoy!

[gmat]

Originally posted by BorgBob
I don't think my temps are particularly spectacular, but I'm happy with them.

Way to go !

[jtroutma]

I have been reading this thread for the content of adding a GPU pelt to the system and yet having only regular water cooling for the CPU.

I am planning on attaching an 85W Pelt to my Radeon 9700Pro and going to town but became worried when nOv1c3 mentioned that it might seriously hinder the CPU temp.

Also the talk about insulating the RAM chips and letting the pelt cool them as well is very interesting. Did cross my mind but figured that fiberglass is not a very good thermal medium.

Well, what I can do is test both situations and see what the results are. I can attach the pelt to my radeon w/o any RAM sinks and see if the RAM is cooled effectively AND see how much higher my CPU temps go on my new Tbred "B" @ 2.0+ Ghz.

Anyone else have more suggestions that I can try?

[airspirit]

I don't know that I really like the theory of the traces wicking heat from the RAM. Though I can't put my finger on it, there seems to be something fundamentally wrong with it (though it could just be my paranoia and skepticism). My leading cause of doubt is that such fine traces can wick that much heat over that much distance. Was the pelt that was used in that example insulated properly? Were the RAM chips building condensation (they would if they were that much colder than ambient and there was any humidity in the air at all)? If this is a real phenomena, it could be a big help for us all.

[bigben2k]

I'm not too happy with it either (i.e. not convinced?).

Heat will radiate in a number of ways: rising hot air, and radiation (infrared). The first can be fixed with a strategically placed fan.

But I have no idea about the second, I wouldn't even know if it's something relatively important.

[gmat]

Originally posted by airspirit
Were the RAM chips building condensation (they would if they were that much colder than ambient and there was any humidity in the air at all)?

Not obviously. Copper traces *do* conduct a good amount of heat, and since the GPU has a zillion of traces going to memory (128-bit or more bus) it's a good connection.
There's an issue of Electronic Cooling about secondary heat losses in IC packages, i may get my hands on it (not now though i'm overflowed with work)

BB2K: usually radiation accounts for between 5% and 20% of the total heat transfer, of course depending on the conditions (reflectivity / emmitivity of surfaces mainly). But otherwise the same goes for radiation: heat goes from hot spots to cold spots.

[Cathar]

Guys, it's also important here to keep in mind that RAM chips consume anything from 0.2 to 0.5W each. ie. next to nix.

It doesn't take much of a copper contact to conduct that small amount of heat away.

If the ram chips were emanating something like 2 or 3W each, then it'd be a different story.

[BorgBob]

Originally posted by Cathar
Guys, it's also important here to keep in mind that RAM chips consume anything from 0.2 to 0.5W each. ie. next to nix.

It doesn't take much of a copper contact to conduct that small amount of heat away.

If the ram chips were emanating something like 2 or 3W each, then it'd be a different story.

I have to say, I've zinged my RAM pretty hard with a voltmod and it never seems to get warm. OTOH, my vid RAM is very warm. I almost wonder if the core isn't cooling well (that damn Radeon thingey that keeps heatsinks from making good core contact) and is passing the heat the other way on the traces!

[Cathar]

Originally posted by BorgBob
I almost wonder if the core isn't cooling well (that damn Radeon thingey that keeps heatsinks from making good core contact) and is passing the heat the other way on the traces!

It is. The GPU core pushes a lot of heat out through the traces into the memory. This is exactly why when on my Ti4600, I can higher memory overclocks, despite the need to remove the ram heatsinks, when I water-cool the GPU core, as opposed to the default configuration.

A lot of the heat from the GPU core makes its way out to the ram chips. The reverse is therefore true if the GPU core is chilled with a peltier, and this directly explains why the memory can be clocked dramatically higher when the GPU core is cooled better.

[UnloadeD]

Well guys, I gotta believe Cathar, specially since he seems to be talking from experience rather than "theory" and I've never know him to BS yet. Also I recently got a Zalman cooler for my 9500 and was wondering how it doesn't stifle and cook the ram chips, Cathar's comments shed some light on this too.

Speaking of my 9500, I'm sad to say I think I screwed it up while installing the zalman. While I had the hsf off, I tried to remover the spacer on the gpu. It looked like it was epoxied on or something as I could see a slight bead of it around the inner edge of the spacer. Without thinking, I ran my razor knife along that edge trying to get the thing loose. I never did get it off, but when I put everything back together, I get strange artifacts in 2D and really strange 3D. I'm pretty sure I knicked a trace 8( I'm not 100% sure I borked it, it's possible the solvent I used before putting on my thermal compound washed away a bit of my defogger mod and I now have a conflicting set of bios/harware, but I have a bad feeling about it. The main reason I mention this is to caution bob when he goes to install his cooling on the 9700. Bob if I were you I'd try and get the bottom of cold plate milled so it can sit down inside of the spacer and contact the gpu fully. Any mucking around trying to get that thing off might result in a $300 oops compared to my $150 one. Good luck and be sure and post results.

peace.
unloaded

[Since87]

Having designed circuit boards where some components have substantial power consumptions, I can also say from experience that the circuit board itself can spread heat fairly effectively.

Nearly all high speed digital PCB's have at least one copper 'plane' which spreads heat throughout the board fairly well. The heat flowing through, even a thin trace, only has to travel through a thin layer of fiberglass to get to the plane, and be spread effectively.

Using the PCB as a heatsink is very common practice, and many component datasheets discuss the amount of copper surface area that should be provided to adequately heatsink the component.

[BorgBob]

Originally posted by UnloadeD
The main reason I mention this is to caution bob when he goes to install his cooling on the 9700. Bob if I were you I'd try and get the bottom of cold plate milled so it can sit down inside of the spacer and contact the gpu fully. Any mucking around trying to get that thing off might result in a $300 oops compared to my $150 one. Good luck and be sure and post results.

peace.
unloaded

Dude! That's a bummer.

I see an interesting thing though. It looks like the Swiftie MCW-50T has a milled out edge from their pictures. Can it be that the block will work fine even if the dreaded spacer ring is left in place? Can the Swiftie Techs be this good?

Here's hoping!

[nOv1c3]

Originally posted by BorgBob
Dude! That's a bummer.

I see an interesting thing though. It looks like the Swiftie MCW-50T has a milled out edge from their pictures. Can it be that the block will work fine even if the dreaded spacer ring is left in place? Can the Swiftie Techs be this good?

Here's hoping!

Yep I do beleive they are that good

[UnloadeD]

I see this info along with the pic on thier site:

Base plate:
0.225" thick CNC machined copper base plate
Outer face is lapped flat to 3/10 of 1/1000", and mirror polished. Center section is raised to provide clearance for ATI® Radeon™ stabilizer frame assemblies.

Only thing that concerns me is they call it the "base plate" instead of "cold plate" also the pic of bottom shown for the TEC version is the same as the one for just the waterblock, they don't show the bottom of the whole TEC assembly. Hopefully they got it figured out right. To me it seems optimal configuration would be (for TEC model) to have the base plate completely flat and the bottom side of cold plate milled to fit inside the spacer.

peace.
unloaded

[MadDogMe]

From my small amount of experience cooling GPU's I've found that cool GPU's allow for a higher mem clock regardless of TEC~wicking. It seems to me the added headroom gained seems to allow for a higher mem clock. Just like cooling the NB on my IWill KA266R allowed me to run the Crucial2100 @ Ultra instead of Fast, the Crucials temp was'nt 'crucial' to the increase...