• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

TEC solution for non-removable stock nVidia HSF?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

HardwareFreak

Registered
Joined
Apr 28, 2004
Location
St. Louis, MO
I have an eVGA nVidia reference design FX 5900 AGP (one of the originals--circa 2002/2003), the one with the 'huge' cast aluminum black painted heatsink and tiny variable speed squirrel cage fan. I've attempted removal of the heatsink, but prying to the point of "fear of damage" leads me to believe it was attached with a thermal adhesive, not just paste (much like Intel used to do on socket 5/7 Overdrive CPUs). I'd like to TEC cool the unit to get a little more performance out of it before I drop ~$300 for a 7800 GS AGP, or spend any money on other hardware upgrades. I'll buy one more card for this system before I'll upgrade mobo/cpu/etc, but I'd like to get another year out of it, and this GPU.

My XP Barton 2500+ @ 2.2GHz (stock voltage, stock AMD heatsink but better fan, on NF2 dual channel DDR 400) is just fine for all the games I currently play, or will soon be playing. With a good copper air cooler it can go substantially higher if need be. Since most games, even the uber DX9 PS 3.0 games, are more GPU bound than CPU bound, I'd like to get more from my 5900. I must say it has been a great performer since I built this system in Aug 2003. The only "current" game I play that pushes it is UT2004. I run 1280x960 75Hz on a Sony CRT, and with all settings highest in UT. I have global driver settings at 2xQ AA and 2x AF, and the games look fantastic, with little image quality gain with higher settings. I only get minor lag on some CTF maps with lots of bots in close quarters. Disabling AA and AF cures this. Other than that, it's liquid smooth. I dont play BF, HL2, Doom3/Q4, or Fear yet I plan to soon.

With the standard nVidia ref driver and CoolBits I am getting stable 480MHz core, and 900MHz memory. Stock is 400/850. This is with the standard nVidia cooler setup. 500MHz core locks the system tight within a minute or two of UT2004 gameplay. Once the core gets much over 60C, it hits a wall. My temp numbers are from the driver panel, with ambient ~30C. I'd like to get the core to 550-600MHz, if possible. Since the stock cooler gets me to 480, I don't think this goal is unreasonable with a TEC. Attempting to replace the stock ramsinks with copper or attempting to integrate the cooling of them with the TEC would be far too complex. Besides, with 28.8GB/s of bandwidth already, I think this should be sufficient even with a dramatically increased core speed. A standard 6800 with 12 pipelines at 375/700 only has 22.4GB/s and is faster in most games than my 5900. Thus, memory bandwidth doesn't seem to be a huge factor once we're over ~20GB/s with the 59xx and 68xx series, and "sane" resolutions (1280x960x32) for a 17" CRT.

So, my idea...

Since removal of the stock heatsink would destroy my GPU/card, I'd like to mod it and attach a TEC. I plan to remove the fan, and seal up the heatsink with sheet metal and epoxy, and integrating a filler spout and air bleeder valve. I will then fill it with regular ethylene glycol (car) antifreeze. I will now have a sealed (no-pump/reservoir/ circulation) low cost low complexity liquid cooling solution, in essence a "sealed water block"--a rather large one. It will obviously have pretty terrible thermal coefficient compared to a purpose built copper waterblock or direct attach cold plate, but I can't avoid this. I'll attach the TEC to the top "plate" of the modded heatsink, which will be a "thin" aluminum (cost cost cost) plate, say 5mm. I have access to this scrap aluminum, but not copper. This should give maximum contact area and maximum use of the TEC's cold plate. I've got the structural bracing of the card/case worked out already to support all the added weight, including up to a 5 lbs heatsink for the hot side of the TEC, ground isolation bushings, etc. I have a supply of massive aluminum heatsink material at my disposal that I can cut to size. I have a 6 inch AC impeller than can flow 500 CFM over the TEC heatsink, which I think should be sufficient (although not quiet, but that's of no concern). I have (or can build or acquire) the necessary 12v PSU to supply enough power to the TEC. I'll coat the board front/back with spray on polyurethane, paying special attention to the hard to reach area under the heatsink. I'll carefully and strategically install closed cell foam to prevent possible condensation and icing effects.

I'd like to get the GPU core temp down to around 5-15C under load. I should think this will give me at least some room for additional frequency, hopefully 20-25% (more would be great).

Given the terrible thermal coefficient this setup will most likely have due to the compromise in design, can I even get close to that? If so, what size TEC will be required? I know that's a hard question to answer without knowing the coefficient of the stock heatsink + antifreeze, but I'm not looking for the "perfect" answer. I'm looking for advice on a "starting point" size of TEC. Or, is this idea even feasable, considering the abysmal coefficient of the stock thermal adhesive eVGA used to glue the heatsink to the GPU?
 
before you do anything try this:
a)put your card in a plastic bag and put it in the freezer. Then wait a half hour and then try to take the heatsink off
b) run the card under intense 3d load for a half hour and then turn everything off and quickly try and take the heatsink off

A tec slapped to the heatsink won't do any good whatsoever. If your card IS in fact epoxied on to the cooler (other way around I mean, just too lazy to find the backspace key) then you might be sol.
 
pwnt by pat said:
A tec slapped to the heatsink won't do any good whatsoever.

I find this hard to believe. Having a chilled liquid on the current heatsink (which is my idea) should be better than blowing case temp air through it.

Are you just saying this because no one has done it (or posted results of it)? Or do you have data to show it won't work?

Did you read all of my post?
 
Chances are the base of the heatsink is not flat, so you're losing there. Second, if it's one of those large heatsinks. Second, water is not the best medium for transmitting heat in this case, although it's the only way. With watercooling, it's the continous flow of water that does the cooling (in a nut shell). A sealed "waterblock" first has no water movement, and second, probably wouldn't be remotely near designed well enough to allow natural convection to work it's magic. If you want to talk heat pipes, those use refridgerant as opposed to water. Also, they only seem to work really well when the heat can travel up.

How do you plan on supporting the "waterblock" through the provided holes on the video card while keeping the block sealed?

From the sounds of it, you're looking to add a cold plate between the tec and the waterblock? What the waterblock should essentially be is the cold plate and adding yet another piece of metal will only complicate things even more.


It could work, however it's obviously not the best solution. With all the efford you're putting in to this project, you'd probably have a cost advantage of selling the card and buying a new card (time = money, remember).

I really do belive that you'll be able to get the heatsink off by either freezing it or heating it up. Then you can play with the tec to it's fullness.

If you really want to play with this idea, I say go for one of the 50x50mm 320w tecs on ebay. You can always supply the tec with less power. IF this turns out to be a remotely effective means of cooling the card, then I'll apologize. Until then, I'm going to stick with my ideas.
 
Do any card makers use thermal epoxy instead of paste? Has anyone posted that they weren't able to remove a HSF from their eVGA brand card? (I searched but didn't find any).

I only assumed so in my case since the HS wouldn't come off for me. If it's just thermal paste (not glue), is it common for it to become hard and like an adhesive after a few years of heavy GPU use (high temp over time), making it extremely difficult to remove?

If it's definitely not an adhesive, I'll try your suggestion, as I'd much rather install the TEC the "correct" way. I just don't want to try any HSF removal methods that may ruin the card.

Oh, and I guess at this point I don't consider my time = $$ ;)
 
HardwareFreak said:
Do any card makers use thermal epoxy instead of paste? Has anyone posted that they weren't able to remove a HSF from their eVGA brand card? (I searched but didn't find any).

Ahh, I found this, which is very helpful:
http://www.ninjalane.com/display.aspx?docname=ultra1600xp&page=2
I would assume since my card is a reference design, it's manufactured in exactly the same manner. Which means I should be able to pull that sucker off. Yay! I'll give it a shot (not right now though, as I have to go get some thermal paste so I can reinstall it until I'm ready to buy/install the TEC).
 
Back