/ Sign-up
Your question

The Graphics Card Cooling Guide

  • Graphics Cards
  • Cooling
  • Overclocking
Last response: in Overclocking
December 17, 2004 3:59:12 AM

Over the past two weeks, I have been testing, retesting and pushing my 9800XT to the max. In addition, I've also researched a bit about other cooling offerings. I’ve tested different types of cooling, from the silent to the high-performance, all with different airflow conditions as well. So here it is, my big, multi-solution, Graphics Card Cooling Review.

A few notes about the testing. The complete testing was performed in two locations: once in my dormitory and once at my house. My dorm is colder but without an accurate way to measure room temp, that is all I can say. My room in my house is kept at a constant 70F/21C. All testing was performed with the case side on and all sound gradings are subjective. Temperatures are measured by the ASUS SmartDoctor tool, which corroborates with ATiTool (which gets buggy at higher temps, in my experience). I intentionally only allowed one PCI slot of room to provide an “at worst” scenario with these cooling solutions.

Load temps were found by running rthdribl for 30 minutes. Idle temps were found thirty minutes later, letting the cooling get the components to as low of a temp as possible. Max OC was tested with ATiTray and then tested with 4 instances of 3DMark03. Never once, however, could the max OC withstand 75 minutes of gaming, so I tested to find the highest ‘gaming’ clocks (the highest possible without artifacts for 75 minutes on either FC, HL2 or UT2k4).

My system is a P4B 3.06GHz on an IWill P4R533-N with 1GB of PC1066 RDRAM. The heatsink on the CPU was a stock cooler of a 2.66GHz P4B (the one without the copper). The CPU is RIGHT on top of the AGP slot, not that should make too much of a difference. I will not list the idle temps of the max OCs since it my general policy to run at default or lower speeds in 2D.

Now onto the numbers and analysis:

First on the block was the stock cooling on my ASUS 9800XT. The 9800XT stock cooling, however, is not the typical 9800XT cooling, but rather a proprietary cooling that is also a single-slot solution. It utilizes two small fans and runs quietly. The block makes contact with the core and all the memory, but does not connect (for the purpose of cooling) the front and back of the card. So, effectively, the front BGAs are actively cooled by the fans while the rear BGAs only have their heat output passively spread. It has been said that the ASUS cooling solution is significantly better than other 9800XT cooling solutions.

Stock cooling, no fans on the side of the case blowing over it:
Stock clocks: 44/38 48/42
Max Clocks (432/774): 50/44
Actual Playing max: 425/760
Stock score on 3DMark03 (will be used for expected performance gain): 6517 (everything was set to high quality)

Stock cooling, 120mm 105CFM fan on the side of the case blowing over it. Set to low for idle (audible, but easily forgotten) and high (very loud and in your face) for load.
Stock clocks: 41/35 45/38.5
Max Clocks (436/789): 47/40
Actual Playing max: 426/780
Max 3DMark03 score possible with stock cooling (i.e., max OC w/o changing the cooler itself, still on high quality settings): 6857 or +5.2% performance

Next up is the Zalman HP80D, a very large and blue cooler by silence specialists, Zalman. It has very high surface area, utilizes two heatpipes, and can have fans of any size up to 120mm mounted onto it. It includes RAMsinks and thermal tape for adhesion, unfortunately, my RAMsinks refused to not fall off, so they were ignored. Installation is fairly complicated and uses a lot of thermal paste (I used AS5 throughout and not their generic crap). The part that actually makes contact with the GPU looks like it needs to be lapped, but I didn’t.

Zalman HP80D – passive
Stock clocks: inconsistent 48/45
Max Clocks (439/761): 51/46
Actual Playing max: 434/752
Interesting results here, although temps are higher, so is the OC on the core. The lack of memory cooling is really apparent as there isn’t even a breeze over them right now.

Zalman HP80D – 120mm fan on low (audible, but not a whole lot)
Stock clocks: 39/31 43/35
Max Clocks (441/770): 44/37
Actual Playing max: 438/760

Zalman HP80D – 120mm fan on high (obnoxiously loud—better wear headphones to play games)
Stock clocks: 36/30 40/33
Max Clocks (443/774): 41/34
Actual Playing max: 443/760
Max 3DMark03 Score using the ZM-HP80D: 6909 or +6.0% performance
Temps drop fairly considerably, yet clock don’t rise much. Looks like this puppy might need more voltage. The lack of any mem cooling is still rearing its ugly head.

HP80D w/ 80mm case fan set at full speed were equal to the 120mm low results while the 80mm fan on low (to emulate the fan Zalman wants you to buy with it) provided temps that were exactly 3C higher in every category than the 120mm low. Max clocks actually varied by where on the card the fan was positioned and will therefore be omitted.

HP80C (the much smaller and simpler younger brother of the HP80D):
Silent: no go under load and therefore a failure.

80mm high:
Stock clocks: 44/38 53/44
Max Clocks (422/754): 55/46
Actual Playing max: 417/750

80mm low:
Comparable to stock w/o the case fan.

ATI SILENCER 3 (not VGA Silencer 3!)

This is just a gem of a cooler. It exhausts hot air from the case, cools memory and has better cooling on the core than either of the other 3 solutions thus far. Not only that, but it runs very quietly (I CANNOT hear it in my case with just a few other fans!) and installs just as easily as stock cooling. For testing I used AS5 and not the included crap again. I put a very thin layer on the core and applied a liberal amount to the RAM. I then squeezed out the extra from the RAM to make sure there wasn’t too much there. When all was said and done, I was quite impressed.

ATi Silencer 3, on high (since I can’t hear it even then, I won’t try low)
Stock clocks: 38/35 40/38
Max Clocks (452/793): 41/38
Actual Playing max: 447/782
Max 3DMark03 score: 7090 or +8.8%
Interestingly, the temps were higher than those of the HP80D, but the OC was also higher. I’m guessing it has something to do with the almost zero difference between stock and max OC temps (and very low difference between idle and load temps as well). This only shows that the card really does need more voltage.

Note: any case fans blowing in its direction actually worsened all results—turbulence is my best guess. So now I get shockingly higher OCs while also reducing noise (the 120mm fan was quite loud).

So what's my conclusion? Get a freakin Arctic Cooling Silencer for your card! Granted, my ATi3 does have more core fins than the coolers for the lower cards, but there will still be improvement.

So what's next for my 9800XT? Well, first I'm going to try to do a voltmod, I've tried before, but nothing changed, which frusrtated me, but I'm gonna try again. Then after that, my card might see some watercooling before it gets replaced with my next system. To note: watercooling will probably only provide an extra 5% clockspeed bump over the ATi3 as what is really limiting the card is the voltage.

So how does this apply to other cards? Well, it simply shows that even when you try to make the other available cooling solutions as good as possible (with loud, obnoxious fans), the Arctic Cooling Silencer still prevailed without making a peep or necessitating a complicated install. I guarantee that the right cooler for your card will decrease running temps as well as increase performance. I know I was on the edge of bumping a setting or two in some of my games and now I can do it, so this isn't just a way to get higher benchmarks but can really be used to get a higher image quality.

Oh, btw, sorry for the formatting, adding spaces to space things out doesn't work (who woulda thunked that?). Oh, and can I get this stickied?

Maxtor disgraces the six letters that make Matrox.

More about : graphics card cooling guide

December 19, 2004 2:09:13 AM

bump / sticky request

Maxtor disgraces the six letters that make Matrox.
December 21, 2004 11:27:06 AM

A summary table would be nice and a LOT easier to digest, before digging into the detail.

But interesting all the same

MIS645Ultra/P42.4@2.8Gig/2*Corsair 256MBit CL2-2-2-7XMSLLPT PC3200 DDR RAM/450W Enermax PSU/1*40Gig 7200rpm Wes-Dig/1*20Gig 7200 Hitachi Disk/GeF4 Ti4200 XFX Turbo 670/250 Ram/Core MHz.
Related resources
December 24, 2004 2:40:36 PM

Good idea, I'll start working on one. IIRC, the forum disregards multiple spaces and tabs and all that useful stuff...w/e, I'll figure something out. Thanks for the idea!

Maxtor disgraces the six letters that make Matrox.
December 30, 2004 10:08:23 PM

Alright, still can't figure out how to actually enter in a clean looking table (or three, actually)...but I've got some more data!

As I promised earlier, I would be working on a voltmod in addition to the modified cooling, just to show you what aircooling really can bring to any of your cards! (well, performance based cards at least)

Alright, using the pencil technique on my 9800XT and the ATi Silencer 3, I upped my Vcore to 1.94V (from 1.76V--which is low for an XT) and my Vmem up to 2.69 (from 2.63--VERY low for an XT, should have been 2.8 from stock--would be higher but no matter how much graphite I load on, it just wouldn't budge). With that, my load temps at stock speeds rose to 51/50. I tested for a new load manually this time (4+ runs of 3DMark03 sans artifacts or crashing) and got an AMAZING 480/400. Load temps were at 53/52 and boy was the hot air coming out of the back of my Silencer!

I might have pulled my vcore up too high (it's recommended to never go past 1.9V with just aircooling, but as I said earlier, I've got a replacement in the wings just in case), and it might be causing me to, rather than artifact at higher clocks, to simply crash. Fortunately, the crashes are only to the desktop, so it takes less time to recover.

Anyway, to sum it up, this was a very worthwhile modification, IMO. Furthermore, I feel that I could use even better cooling to really get more out of my card.

Numbers summary:
STOCK Voltages: <b>1.76/2.63</b>; max OC: <b>452/793</b>. Max extended playing clocks: <b>447/783</b> (any higher: artifacts). 3DMark03 performance gain: <b>8%</b>.
MODDED Voltages: <b>1.94/2.69</b>; max OC: <b>480/800</b>. Max extended playing clocks: <b>480/800</b> (any higher: crashes). 3DMark03 performance gain: <b>12%</b>.

btw, tell me you if like ^^^^^^^^^^^that format better, I think it's usable for the above section.

Maxtor disgraces the six letters that make Matrox.