MOTHERBOARD DUCTING: The Bonzai-Duck-t-ster Speaks

[BonzaiDuck]

EXPERIMENTAL DUCTING MODS FOR THE TR-ULTRA-120-EXTREME

As much as a couple months ago, AigoMorla and I had "an exchange" about the virtues of water-cooling versus air-cooling. I've been in process of building my "Chrome Lightning" E6600 system for a few months now. No -- that's an understatement. I spent hours weekly sitting on the patio last year looking at this Compaq ProLiant Server case (1994), pondering what I would do with it, and I starting cutting sheet-metal in February of this year -- just as the parts for the system arrived for installation in a junk-chassis for testing.

People privy to the heated discussion made me promise to post some pictures. I'm still working on this project, and I've compiled a million pictures on every aspect of it, so I'll wait until I'm finished to my satisfaction before I post them all in a big "project unveiling" here.

But also, in the course of our exchanges in the "Cases and Cooling" forum here at Anandtech, the topic of "motherboard ducting" has arisen. I'm a big proponent. So I thought I would post some "progress" shots.

THE "RIG"

The system now resides in the modded Compaq ProLiant:

Against Current Trends for small footprints -- My Chrome Lightning

E6600 @ 3.35 Ghz, VCORE = 1.46875V ("set" value); idle BIOS monitor value = 1.45V
2GB Crucial Ballistix PC2-8000 / DDR2-1000 @ 744 Mhz and 3,4,4,8,2T
BFG (nVidia) 8800 GTS
Striker Extreme motherboard BIOS version 1004
3Ware 9650SE PCI-E x4 RAID controller
4 Seagate 7200.10 320GB "perpendicular" drives in RAID5

Coolers: ThermalRight Ultra-120-Extreme CPU heatpipe cooler; ThermalRight HR-03-Plus VGA (8800) heatpipe cooler

I mentioned in threads created by other members that the size and orientation of heatpipe fins made it difficult even to anticipate an SLI configuration while preserving the use of PCI and even PCI-E slots.

The difficulty arises with a combination of the HR-03-Plus and TR-Ultra 120 that the latter, in its favored broad-face orientation to a rear exhaust fan, reduces the effectiveness of the HR-03-Plus when the VGA cooler is deployed with the heatpipes and fins between the VGA card and the CPU cooler. It was essential to use that orientation option to make it possible to use another graphics card cooperatively in SLI. Addition of a second graphics card would still preempt use of a PCI slot, but preserve access to the PCI-E x4 slot for the RAID controller.

Some may wonder why I went to the trouble and expenses of using the 3Ware controller when RAID 5 is already available on the Striker Extreme motherboard. Easy -- I wanted maximum performance without using CPU clock-cycles, so I chose to add a "hardware" controller to the system. And frankly, some reviews and evaluations I've read are not chock-full of stellar kudos for the nVidia Southbridge, other than "It's OK."

The sustained throughput for a single Seagate 7200.10 "perpendicular" drive has been variously quoted to be somewhere between 68 MB/sec and 78 MB/sec, as compared to the older WD Raptor drives spec'd at 72 MB/sec. Benchmarks for my 3Ware RAID5 array show sustained throughput between 177 MB/sec and 250 MB/sec, so I would suppose (for the trouble and expense -- as I said) -- that this is the sort of thing I was aiming for. It will not exactly reduce my electric bill, but for its nine-tenths of a terabyte, I can retire some other machines that I leave running 24/7.

THE ULTRA-120 ORIENTATION AS HANDICAP

I wanted to reduce the need for fans in my system. Currently, the rig is outfitted with two Aerocool 140x20mm fans in a case-bottom twin-fan intake duct, a Sharkoon 140x25mm intake at the lower case-front, a second Sharkoon 140x25mm hanging behind the RAID cage and blowing air at the ULTRA-120, and two Sanyo-Denki "DC San-ACE" 120x38mm exhaust fans rated at 12V, 0.52A, 104 CFM and a top-end of 2,500 RPM. In addition, ThermalRight recommends use of a 92x25mm fan on the HR-03-Plus, although I've found that an 80x15mm Zalman OP-1 option fan works quite well. I have Lexan plates for blocking fan-holes that I don't want to use, so that I can leave one or more fans installed but unplugged.

But for either choice of a VGA-cooler fan, the favored orientation of the ULTRA-120 just gets in the way.

We'd variously discussed improvements of custom-lapping the TR-ULTRA-120-Extreme, and additional drops in load temperatures with the use of nano-diamond or IC Diamond thermal paste. I'd posted statistics on the configuration above showing that I'd captured another 2 to 3C advantage for the use of the thermal paste, and I think I'd managed to get my ORTHOS load-temperature peak-value down to between 43C and 45C at a 72F room-ambient. But flipping the cooler around 90-degrees changed all that:

Degraded Performance Statistics Due to Cooler Orientation -- "Before Ducting"

Here, at about 73F room-ambient, the peak load temperature value was 53C, and even if the number of sampled observations at that temperature is a small blip at the right tail of the distribution, it is, ah, "tell-tale."

So I built a CPU-cooler duct to the rear exhaust fan using foam art-board. Here, the edge of the cooler is less than an inch from the exhaust-fan motor-hub, so it's important to try and force all the exhaust air through the CPU cooler fins, or there will be a "dead hot-spot" because of the proximity of the motor-hub. As I said, there is about an inch of clearance between the fan and the narrow edge of the heatpipe cooler, so the interior of the duct was designed like a tapered hour-glass -- with the narrowest aperture closest to dead-center of the cooler's broadest side:

An Unlikely Final Choice for Duct-Design on the TR Ultra 120 Extreme

The square box with the circular hole accepts the fan-exhaust from the 92mm fan sitting on the HR-03-Plus heatpipe cooler, and ducts it to the surface of the motherboard, where it is forced past the Striker Extreme's "heatpipe necklace" chipset and Mosfet cooler toward the upper exhaust fan -- with the air flowing between a flat motherboard duct-panel and the motherboard itself:

The Mobo Panel -- A Hidden Power Behind All Designs

Note here especially the perpendicular orientation of the TR ULTRA 120 Extreme, with a narrow side facing the lower exhaust fan.

The duct-cover for the ULTRA 120 has an open bottom facing the right-side of the case and motherboard-pan, but sits flush against the flat motherboard duct panel. This latter motherboard duct-panel has slots cut to fit around the two heatpipe clusters of the ULTRA 120, so that the heatsink base is completely covered by the panel:

Mating the Cooler Duct and the Mobo Duct-Panel

The only thing left to do now is to design a third component which mates the upper 120mm exhaust fan with the underside of the flat motherboard panel, pulling air from between the panel and motherboard as well as from underneath the motherboard. However effective this last feature might be, it eliminates the need to install a fan under the CPU by cutting a hole in the motherboard pan and adding more fan-noise to the system:

Ducting from Mobo to Upper-Exhaust Fan

More of the Upper Duct component

If any of these three components don't work as optimally as we would like, we'd hope that we can modify them without changing the other parts. I chose to work on these parts before extending the ducting over the VGA card and building a duct-box for the RAID drive-cage, because the CPU, motherboard and VGA heatpipe cooler have the highest priority in cooling, and any improvements with those components will be obvious.

Here's the "whole enchilada" for the corresponding duct-components -- fully installed:

Boxes in the box

When I first flipped on the power switch, booted to Windows, loaded CORETEMP and ORTHOS and began testing, I noticed that the aperture between the upper side of the ULTRA 120 and the duct cover was too wide, so that not enough air was being forced between the heatsink fins. At this point, ORTHOS was showing a CPU temperature for both cores of approximately 46C. The test had only run about 20 minutes, so I picked up a piece of scrap foam board, opened the case, and blocked that aperture:

On-the-fly Alteration to Plug a Hole

Looking back to the monitor immediately, I saw the temperature drop to 43C. While ORTHOS was running the same calculation battery that caused the 46C load-temperature, we could only speculate that blocking the hole caused the temperature drop. Otherwise, to prove it, we'd have to run ORTHOS for two hours with the duct in its original state, and another two hours with the scrap foam-board added. But we continued with this test -- ORTHOS fully stresses the CPU and RAM to their maximum in the Blend Test after it has been running for at least 20 minutes.

Here are the results:

Improving the Degraded Performance and Re-capturing All the Ground -- Test Result Statistics

Except for any variation in temperature departing from the ULTRA 120's optimal range, it's obvious that the ducting accounts for a 5C-degree drop in the peak ORTHOS load temperature. Comparing average temperatures between each two-hour test, the difference seems smaller -- or an "X-bar" value of about 45C for each processor core without the ducting, and 43C once the ducts were installed.

Since the cooler orientation degraded the effectiveness of the Ultra 120, we'd expect even more stunning results if I could flip the cooler around so it faces the exhaust-fan with its broad side, and that is what I shall do next. Here's why.

STUNNING VGA-COOLING RESULTS DUE TO THE SAME DUCTING MOD

I was concerned when I added the shallow duct-box to the ULTRA 120's duct-cover -- the one with the circular hole ducting the 92mm cooling fan on the HR-03-Plus -- that it would not be sufficient to allow enough air-flow for the fan's output. I could not have been more wrong about this.

I used nVidia Monitor to log temperatures of the GPU core on the 8800-GTS VGA card. I could show you a graph, but there is only a single temperature interval for both idle and load GPU temperatures -- and no statistical distribution!! Let me explain the history of my observations with the BFG (nVidia) 8800 GTS.

The original cooler on the 8800 GTS has a single copper heatpipe, and ducts (extremely) warm air through the card's vented PCI-E slot-cover. I could play Nadeo's TrackMania Nations for 15 or 20 minutes, and the peak GPU temperature would reach 70C degrees as shown in the nVidia Monitor logs. And the card would idle at around 50C degrees -- with these temperatures occuring at room ambient values between 72F and 75F degrees.

Supposedly, today's graphics cards and GPU's are made to take this sort of temperature stress, and magazine articles over the last year have pointed out that while Intel and AMD focused their efforts on reducing the thermal wattage of CPU's, nVidia and ATI have not done so with regard to their graphics cards. So the graphics card (or cards -- for SLI) are the hottest components in today's systems.

When I installed the HR-03-Plus heatpipe cooler, I was very pleased: idle temperature for the 8800 GTS was around 45C degrees depending on room ambient, but nVidia Monitor logs showed peak load temperatures around 51C degrees. Thus, by itself, the HR-03-Plus resulted in an idle-to-load spread of around 6C degrees.

With the ducting and 92mm fan, I ran two tests: 20 minutes of TrackMania, and Future-Mark's 3DMark06 graphics benchmark. For either test, the idle and load values were identical: 42C degrees for TrackMania, and 41C degrees for 3DMark06. nVidia Monitor logs showed that CPU usage at some times during these tests had even exceeded 90%.

Correct me if I'm wrong, but this result with the HR-03-Plus enhanced by my ducting mod seems to match results for some pretty good water-cooling. AigoMorla can provide me some input on that matter, and I might be wrong, but in any event, these are darned good VGA temperatures! No -- let me correct that statement -- this is a darned good VGA temperature (singlular -- not plural).

ThermalRight cautions that the HR-03-Plus should be used with a fan, and heeding these recommendations compelled me to choose the suboptimal orientation for the ULTRA 120 CPU cooler. I believe -- with improved ducting around the HR-03-Plus -- that I can dispense with the 92mm fan, and still achieve stellar temperature profiles for the VGA card.

The only way to find out: remove the fan and add a duct-extension one-inch deep to gauge the effectiveness of the case's upper 120mm exhaust fan for keeping the VGA card almost as cool and without the fan. Once I've satisfied myself on this matter, I'll simply flip the ULTRA 120 around to face the lower exhaust fan, and rebuild the duct-cover for it as a simpler integration of ducting for the CPU and VGA coolers.

FOOTNOTE ON THE NORTHBRIDGE AND SOUTHBRIDGE CHIPSETS AND MOSFET COOLING

My particular motherboard, as are several mid-range to top-end products by ASUS, is equipped with a copper "heatpipe necklace," which connects the Southbridge, the Northbridge, and all the Mosfets around the CPU socket. By covering the motherboard with even a partial duct-panel, and using the RAM modules as one edge of the ducting, pressurized air is forced through the narrow apertures of these copper components and the RAM modules, and should significantly improve cooling those components. I would need to install temperature sensors on them and run tests -- first without, and then with -- the ducting. Another way to find out, of course, is to see how much more I can overclock the E6600 and Striker board under the same voltages used to reach 3.35 Ghz. That may be another priority for the forthcoming days, but I may want to make the other changes to CPU cooler orientation and ducting before I proceed.

ORIGAMI, FOAM-ART-BOARD, LEXAN AND RUBBER FOAM

Art board is cheap stuff. That's how John Cinnamon's "OverClockers.com" article of 2003 gets its title -- "Cheap as Free." Even at the pricey Michael's Arts and Crafts, a 2'x3' panel of art board costs about $6, and I've found it at Target for less than $2.

All you need to cut it is a decent metal straight-edge and an Xacto knife. The trick, of course, is to build ducts out of single pieces, stripping away an eighth-inch-wide strip of the paper surface to make neat folds in the material.

In woodworking and cabinet-making, you learn to account for fractions of a millimeter in saw "kerf," so that the pieces fit together well in the joinery. It's too easy to overlook this with foam art-board: the tabbed design I've shown above still makes for a messy result. But by cutting flat-panels that are consistently smaller than the surfaces in your duct and gluing them to the interior of the visible panels to show a precise recessed edge, you can fit the duct component together as if cutting rabbet-edges in hardwood with a router.

Foam-board also has an advantage for a noise-deadening quality, although the paper surfaces are rigid. If the duct-components are large enough, you could line the interior of the pieces with Akasa Paxmate, Spire Foam or other noise-deadening products. This might also increase the weight of the ducts and create a need to anchor them to the case with standoffs or tape-on Velcro patches. Otherwise, the material is light as a feather and would make good box-kite material -- so it hardly matters whether one chooses to use double-ply panels.

For appearances, it lacks the advantage of transparency, but it can be painted with enamel or UV paint. And, as I said, it's cheap.

Working with the material is fairly easy with a sharp Xacto knife. By removing eight-inch strips of the outer paper layer, it makes neat folds. While one can cut each individual surface as a separate panel, it is challenging to think of ways to construct a duct from a single piece. It serves as a good refresher-course in high-school geometry:

High-school Geometry in 3D -- Is It a "P-I-T-A?"

For use as a prototyping material, it's better than cardboard, and you can either use the pieces or the leftovers to cut final ducts from Lexan:


After the Cookie-Cutter

Lexan is more rigid, easier to join more precisely, prettier and transparent. You may want to anchor the pieces to the case. It takes longer to cut, but the right glue actually welds the pieces together. With foam-board, you can use contact cement, but the solvents tend to dissolve the plastic foam. It's easier or more dependable to join the paper surfaces together.

WHY I BECAME A DUCTING FANATIC

It's simple. While I've cited other reasons to avoid water-cooling, my underlying personal psychology has been the true theme of my avoidance: "Everybody is doing it" gives me an excuse to search for some simple, low-tech computer-cooling method. But the thermal properties of water are undeniable. I like to cite examples in the automotive industry for the feasibility of air-cooling -- before 2001, "911" had a favorable connotation -- as a model-identifier for a popular Porsche sports-car. The Porsche was air-cooled.

This may or may not be the end of my Bonzai-Duck-ting pursuits, or it could be the continuation of them. I've ordered my Q6600 processor, and will initially use the ducts I've designed here to temperature-advantage.

But I suspect that multi-core processors exhibit a different "scaling" in thermal power. Even if the Q6600 has a stock TDP equivalent to a Pentium 4 Prescott, it's been shown that over-clocking it more than moderately makes water-cooling a near-imperative.

That doesn't preempt the use of ducting solutions to work with water-cooling, but with the advantage one would see in using top-end water-cooling components, it would be overlooked. Its advantages would appear to be marginal, and for maintenance of water-cooling components, it would easily be dismissed as a nuisance.

Even so, I think those looking to make quantum leaps in their current air-cooled rigs and anyone with a sense of practicality can see that even an initial 5C improvement in CPU temperatures -- such as demonstrated here -- is nothing to sneeze at. I'm still hoping to reduce my CPU temperatures another 5C degrees, so if I orient my cooler and fans so that I can get back to peak load temperatures around 45C, I might be able to hold over-clocked temperatures at 40C. At least -- 40C for my over-clocked E6600. What arises as a result with the Core-2-Quad Q6600 -- well -- that's anyone's guess.

What is most obvious here are the gains in my VGA temperatures, and as I said, I still need to measure the chipset-cooling improvements -- often cited by ducting aficionados as the biggest, most immediate reason for building ducts.

 

[scrawnypaleguy]

Has Anandtech hired you to do their cooler reivews yet? Because they should.

 

[aigomorla]

Originally posted by: scrawnypaleguy
Has Anandtech hired you to do their cooler reivews yet? Because they should.

AHAHAHAHAHAHAHHAHAHAHAHAHAHA

Im sooooooooooooo +1 on this statement.

 

[BonzaiDuck]

Gentlemen!! I'm flattered!!

Fact of the matter is -- on that question -- if they "pulled me back in to the world of work," it wouldn't be like work. So the question remains "Do they read these forums?" They say they do. But what fun it would be, to get manufacturer samples, and spend all one's time installing them and measuring "thermal resistance."

Playing around with my Xacto-knife these last few days, I began to feel like Tommy Lee Jones' character in "Blue Sky" (1994) with Jessica Lange (hot!) and Powers Booth. Jones' character Col. Hank Marshall and his wife (Lange) discover that the Air Force has been cavalier about protecting people and livestock during nuclear tests in the Nevada desert, and they eventually lock him in the psych ward under medication, where his only activity is to do leathercraft and make those western-style wallets (like the one Samuel L Jackson forces Tim Roth to hand back to him in "Pulp Fiction" (also 1994) -- you know -- the one that says "M_____ F_____" on it --hee, hee).

Well, I should go forward today and modify a leather-wallet -- I mean foam-board-duct . . . . I think I'll chill for a while, though, and see what's going on with "The View."

[And for that, I MUST be on some kind of . . . . . medication . . . . ]

 

[WoodButcher]

Bonzai,,,, excellent work, nicely documented. I've thought about ducting but haven't done much experimenting yet. Thanks for the motivation, nice temps!

 

[JEDIYoda]

As we have come to expect all good info!!!

 

[BonzaiDuck]

It can be a pain -- I-T-A.

If you look at OEM machines over the last decade, you can see that the manufacturers know that ducting improves their avoidance of warranty replacements. They've designed systems for purely stock settings and for mainstream users. My friend down the hill from here had a 1999 Dell Pentium 3 -- I think it was a processor rated at above 1 Ghz.

He gave me his cyber-junk when he replaced the Dell with a newer C2D model, and I used some of his sheet-metal in my "Chrome Lightning." The '99 Dell came with a Kelly-green molded-plastic CPU duct mated to the exhaust fan, and the duct and fan assembly was actually hinged to the case.

As I said last week on someone else's thread, OEMs can do this, since, like Outer Limits "controlling your horizontal and vertical," the OEMs pick their motherboards and they also pick their cases for mass-production.

For DIY'ers like us, neither the motherboard manufacturer nor the case-manufacturer will provide any kind of plastic ducting (other than with that Silverstone and its Lexan or Acrylic tube). The case manufacturers cannot anticipate the precise location of CPUs on fifty different motherboards, and the motherboard makers can't anticipate the myriad case-designs.

So it's purely a custom-fitting. If you evaluate your leisure time in dollars (and if that were precisely possible, but I'm not going to digress with citations from American Economic Review) -- you might decide to water-cool unless the tedious work of building ducts is a reasonable trade-off to bleeding the system every six months and replacing the coolant.

But, water-cooling being what it is, I'm still stunned by the results on my VGA card and the ThermalRight HR-03-Plus. There can be no mistake, because I have the time-series of nVidia Monitor temperature samples to prove it. But 42C at those room ambients beats out my old nVidia 5950 Ultra with a Zalman dual-heatpipe cooler on it. So I'm still asking myself "What!? Am I kidding? I don't know whether to hit the brandy-bottle to celebrate, or just steady my nerves!"

 

[SolMiester]

Hey Bonzai, nice work...I have always appreciated the ducting you find in servers, like why do they do all of that, desktops dont!?...LOL. Guess I'm lucky with my TJ06, already have the ducting there, though the 939 socket is currently out of line with the wind tunnel, the gigabyte 935 dq6 seems to be bang on.

 

[BonzaiDuck]

That's interesting -- that the motherboard would perfectly coincide with the case.

I tried to get Silverstone to sell me that Lexan ( or probably Acrylic) tubular duct. Never heard anything back. But I found a company in Ohio that sells acrylic tubing in various sizes and diameters. Minimum purchase is a single 5' length. You can get it in 4", 5" and 6" diameters.

For 5"-diameter, 5'-long, it's about $51. I'm flirting with the idea, but I'd only find use for about a fifth of it.

 

[Noubourne]

First of all, congratulations on the successful experiment. Many people are looking for cheap ways to cool their kit, and it doesn't get much cheaper than an exacto knife and some cardboard.

Maybe you talked about this, and maybe not, but why all the non-conductive ducting?

Surely metal would be more likely to absorb some of the heat being blown through it, and thus increase cooling potential, as more components in your case could act as a heatsink? Or do you think not? With good airflow, I would think that ducts made of metal would only increase the heat dissipation.

Metal would be harder to work with, but it may have the added benefit of not being so ugly, which is what I see as the main detractor from your work.

What point is the window, if you've covered up all your cool-looking components with cardboard or other bland-looking squares? One attraction to water for me has always been the style. I prefer the look of electronics to a piece of cardboard or plastic.

For looks, maybe clear curved acrylic material would allow the components to be seen. If it was curved appropriately, it might actually magnify some areas, making some use of the side window and those CCFLs that everyone is so fond of.

That's just my .02.

 

[Old Hippie]

Man you do good work!:thumbsup:

 

[cyburzaki]

excellent work, Ductmeister!

 

[BonzaiDuck]

Nourbourne --

On the matter of metal and case choices in general. True, Aluminum is better than mild SECC sheet-steel by a factor of three, and metal in general conducts heat. But -- forgive my tongue-in-cheek pronouncement -- I believe this is myth. Myth is often "half-truth."

If the components that are source of the heat are cooled immediately by air, and the air is blown immediately from the case without mixing with fresh intake, then the usefulness of a choice of metals declines to near nothing. That's why I can be comfortable with a 1994 recycled case.

On the issue of appearances and case side-windows. True. It definitely detracts from the aesthetic factor to use foam-board, but the reason they call it "foam-ART-board" is that you can do much with it. For example:

Exotic Acrylic Enamel to "Pimp Your Rig" [and pricey, too . . . .

Even Rustoleum Metallics Specialty Silver or Gold Enamel would work wonders, but of course, you can't see the gleaming electronic components. You can also uses UV reactive spray-paint, as one might use with the components themselves (because once dry, it's non-conductive). In addition to that, and to avoid power usage, there are UV LEDs that can be purchased for about 80-cents each. (I have four of them, just to give me the option.)

But as you suggest, the better choice is Lexan. You can combine the Lexan with the foam-board when the foam-board serves to cover holes or duct in areas which cannot be seen anyway. If certain shapes cannot be had in Lexan, you can buy them in clear Acrylic, such as 5"-diameter clear tubing (1/8" thick). A 5-foot length of the Acrylic tubing, similar to the item found in the Silverstone TJ-06, is about $51.

You can shape Lexan with a heat gun. It requires practice, but I've had some success on simple shapes. One of our number here at the forums chose to use the packing material for his SI-120 cooler to make a form so that he could shape a duct to the cooler.

Even so, I'm debating -- for instance in regard to the upper duct that looks like a bellows -- to use flat, triangular pieces that are cut so precisely and glued so carefully that the result might look like some huge, blue South-African gem.

Lexan is UV reactive -- to some point. I have some other tricks up my sleeve for "FX." And frankly, that's why I picked up the UV LEDs. I want to explain, but I want to keep it a surprise. The problem is in the "art-work" of cutting the pieces. But for a "light-show," I have plenty to work with. The RAID5 array and 3Ware 9650SE controller have server-grade LED provisions. For one thing, there is a pin-out on the controller for an activity light related to each and every single hard-disk. There are pin-outs for "fault" or "error" LEDs, but you wouldn't expect that these would light up often. Finally, there is a general activity LED pin-out for the entire array.

At present, I've hooked these up in conventional fashion, although there is a sort of "Star-Spangled" suggestion in the choice of colors. I still have to upload the photos from my Olympus. The case-front-door requires clear acrylic prisms to transmit the LED flashes from the LEDs on the RAID drives and installed power-on and general-activity LEDs on the front of the chassis. I chose red for the individual drive lights, blue for the power-on, and white for the general RAID activity light. The case, however, came to me without a prism for the general activity light, and I had to cut a slot in the case-door for it. I was able to manufacture it from scrap acrylic pieces originally intended for the hard-disk "on", "activity" and "fault" LEDs. It was -- what I consider to be -- a great success, so stay tuned.

But for now, or short of installing the FX equivalent of a Tesla coil inside the case, I want to know if the ducting is worth the trouble with the Lexan, acrylic tubing, etc. It should be sufficient, and I can use the foam components as templates for cutting the Lexan pieces. Even so, I'm just about to demonstrate the advantage of going slowly with this sort of thing, in order to improve the ducting design and fan deployment. I've already switched around the Ultra-120-Extreme and modded the box-shaped duct for the cooler. It enables me to totally eliminate the flat piece covering the motherboard, insofar as the box-shaped duct simply absorbs that component.

I'm testing it at the moment. I'll post results soon on "mild" over-clock settings as well as those described in my initial post here. For now -- I can say this: It looks pretty good. I still want to test the VGA card, because I've totally excluded the 92mm fan on the HR-03-Plus cooler.

 

[JEDIYoda]

Thanks again for the information!!!
Very informative!!

Peace!!

 

[bigpow]

I concur, one should be compensated after doing such a thorough job!

 

[BonzaiDuck]

UPDATE

OK, folks. Before I summarize and present the results with the new CPU duct in combination with the removal of the 92mm VGA fan on the HR-03-Plus, let's recap with some observations.

As far as I know -- and I'm eager for any new information about just-introduced heatpipe coolers for air-cooling -- the THERMALRIGHT ULTRA 120 EXTREME is THE (capital-THE) most efficient heatpipe-cooler for socket 775 currently available.

Given that, I might have anticipated the results, which I'll present here in a minute.

Here are some important things to understand about this cooler. It has four sides of heatpipe fins exposed to the air. This implies certain things. First and foremost, even hanging a fan right on the cooler (and it was designed to offer that as an option), and assuming that you have an exhaust fan right behind the broadest side of the cooler, there will be some leakage of warmed air from the sides. On the other hand, a slightly pressurized case may actually compensate for that -- by forcing even more air from the sides.

When you evaluate fans, you discover that the specs show a number for an instantaneous exhaust air-pressure at the fan's exhaust side. For this, the most effective way to use a fan on the TR Ultra Extreme (and many other coolers) is to hang the fan on it -- on either side, but I'd recommend a "pusher" fan over a "puller."

My objective in this case-mod and Core-2-Duo build has been to provide a maximum potential for low-rpm, high-cfm intake fans -- for the largest number of fans to use with the case in a practical sense. Once all those fan-holes had been cut and fans procured, I also anticipated not using all of them, so I designed Lexan fan-covers to plug the holes for unused fans -- enabling me to keep those fans in the case while they were nevertheless unplugged.

What I showed with this thread was an attempt to eliminate a CPU fan hanging directly on the cooler. And currently, the cooler's orientation is broad-faced against the exhaust fan -- a choice that most people would logically prefer. That fan is now about an inch or so behind the broadest side of the cooler. The only "pusher" fan is a 140mm Sharkoon hanging on the back of the RAID cage to do double-duty as a hard-drive intake cooler, pushing air onto the TR Ultra 120 Extreme. I did not seal this fan's output to force it onto the TR cooler, because my experience with (for example) side-panel intake ducted directly to a "face-down" cooler suggests that you lose the advantage of a slightly pressurized case that way.

But with the "puller" exhaust fan (120x38mm Sanyo-Denki DC "San Ace") and the Sharkoon working together, it may be worth sealing the duct I'm about to reveal here to increase the pressure at the intake side of the cooler.

Here's the new duct. Incidentally, since I had to flip the Ultra 120 EX around, some would think that I needed to remove the motherboard from the case to do so. Not true. If your case has just enough room, you can insert some non-conductive, flat object (like a plastic ruler) under the motherboard to shim the X-bracket so that its bolt-hole extrusions stay in place in the motherboard holes. Then, just unscrew the spring-tensioned screws from the Ultra's retention-plate, flip the cooler around after servicing the the thermal paste, and re-insert the screws. Nothing to it.

Here's the "revised" duct for the Ultra 120-Extreme, showing near-completed construction. The bottom fin of the cooler just fits the hole in the bottom of the duct covering the heatsink base, so it actually serves as part of the duct separating the CPU-cooler air from the motherboard air. And I took the three-sided box design and closed it, so there's no need for a motherboard flat-panel. Because the fan is off-center from the cooler, I had to cut a hole in the side of the duct to fit the cooler's narrow edge:

Partially constructed duct

Then, I added a double-ply plate to cover that side and seal it, with the outer ply of the plate serving to block off air to the narrow side of the cooler:

Completed duct

I took greater care with the construction of this duct, so that baffles which prevent air from going over the top of the cooler also serve to increase rigidity. There is no obstruction to the full face of the fan, and no obstruction to either broad side of the cooler. In fact, more of the cooler's broad sides are exposed to airflow than would be the case with hanging a fan on the cooler with the wire-clips provided by TR.

Also remember that I wanted to remove the 92mm fan from the HR-03-Plus VGA cooler to see how ducting the motherboard would pull air from its fins. Recall that my rock-solid load temperature for the VGA card with the 92mm fan was a steady 42C.

I've loaded the new retail version of TrackMania United formula racing game. Without the fan, there is no variation in load temperature -- it is about 54C at a 73F room-ambient. And that graphics card has to be working pretty hard with this new version of the game. CPU usage during game-play actually went above 90%, and fluctuated between 50% and that extreme. But there was no fluctuation in the load temperature. Recall that with the freeware version of TrackMania and the stock BFG 8800 GTS cooler, load VGA temperatures fluctuated and peaked at 70C.

Let's repost the earlier results with the E6600@3.35Ghz and the TR Ultra 120 in the "face-up" position, where the duct forced air through the narrow sides of the cooler:

The previous CPU-cooler duct box

Now, here's the result with the cooler in the favored "broad-face" orientation using the new duct:

E6600 @ 3.0 Ghz, FSB 1334/DDR2 667 and Vcore 1.40625V

So far, so good -- these are darned good temperatures, but remember we started this exercise with an overclock to 3.35Ghz and the FSB at 1488 [DDR2 = 744] using a VCORE or 1.46875V:

73F room ambient, E6600 @ 3.35 Ghz and Vcore 1.46875V

Now -- I was really crestfallen by this, knowing that I promised to post these results here. I expected much better. But now that I look at the "before and after" graphs with the two cooler orientations and ducts, I find this very, very interesting. The "unconventional, narrow-side-to-fan" duct WORKS BETTER!! This absolutely has to be the case, since the "Before" statistics comprise two hours of data at the same sampling interval, yet there are only about 25 observations reaching the peak temperature of 47C. The "After" data is only one-hour's worth, but there are some 70 observations at that temperature. Well, we better check our thermometers, but there's only a half-C-degree difference in the room ambients, and I just don't think that this accounts for these peak-temperature results.

I didn't show the average and standard error values for the distribution in the graph for the first (narrow-orientation) ducting -- it was 41.99C and about 3.5C respectively. So there's hardly any difference in the central tendency and dispersion measure between the first, 2 hour test, and the second one-hour test.

There was something "better" about the first duct.

I think this also shows that we've pushed the ducting for the TR Ultra 120 to the limit, and it's more difficult to capture any gains in its already stellar efficiency. There are a couple things I'd like to do next, just to convince myself. First, I want to seal off the case more thoroughly, and install the duct cover to the RAID cage -- so the fan behind it isn't sucking air from the cage holes admiting other case intake to the mix. This might serve to pressurize the case to a marginally greater degree.

Second, I may attempt to seal that fan's output to the input side of the CPU duct.

But I'm not as confident about reaching a load temperature of 40C (at 73F room) through the ducting process.

Even so, look at the two option results for the VGA cooler. And there WILL be no doubt about it -- the motherboard components are much cooler for the ducting -- we just haven't measured that difference. So I need to try some higher over-clocks, and maybe even see what happens when I drop the VCORE on either of these configurations while holding the same 3.35Ghz speed settings.

There could be more surprises.

Also -- a note about measuring room temperatures. I should call this "Radio Shack Gets the Theta-101 Treatment." I probably should simply have put a thermal sensor at the case-front exterior and hooked it up to one of the Striker board's two-pin thermal sensor pinouts. Instead, I wanted a "digital thermometer," and went to Radio Shack to get one. Well, I'm unhappy. For reference, I have a backpacker's analog thermometer and the thermometer on the house thermostat, which is about ten feet away. And the Radio Shack thermometer seems to be in error, posting temperatures that exceed the other two devices by 5F degrees. Using the Radio Shack thermometer would make my results look better than they are. Using the other two devices makes me just a tad less certain about accuracy.

But I don't see how a half-degree Fahrenheit difference in test conditions will give me 25 observations at a full 47C for the two-hour test, and some 70 observations at that value for a one-hour test. The only loose-joint in that situation is the accuracy of the devices and the readings.

With that, I may just decide that a 42C load temperature for the BFG card is perferable to a 54C value, and reinstall a modified version of the first duct design.

And that still leaves the matter of chipset and other component temperatures as a certain, yet not precisely known result of all this. I'll have to see whether I can OC better under these circumstances, and I'll post results later to resolve that question, too.

 

[BonzaiDuck]

AFTERTHOUGHT:

I want to tentatively revise my conclusions about the two duct designs, because you can also see that the largest frequency for the newest version is at 43C, while the largest frequency for the previous duct is 46C. So while the averages and standard errors are about the same, and the peak temperatures are identical, the CPU spends more time at 43C with the newer design. But temper that fact with the number of observations shown at the peak value of 47C . . . . .

. . . . Maybe I WILL get to a 40C peak -- or maybe at least a 43C peak -- with some more . . . . "refinements" . . . . .

The Q6600 processor arrived today . . . . I see a whole new regime of tests ahead, but these results should drive the ducting approach with the new processor.

Tom Cruise, "Top Gun" -- "Do you feel . . . . the NEED . . . . for SPEED?!!"

[And that new TrackMania retail -- GEEZ!! Kick-in-the-A__!!!! The only difference between that and getting real dust and sand between my teeth -- I can't taste it!!]

 

[BonzaiDuck]

"EEEEEEEEE!----- HAAAAAAAHHHHH!!! EEEEEEEEE! ---- HAAAAAAAHHHH!" [Slim Pickins, riding "the bomb" in "How To Stop Worrying And Love The Bomb," [196-ty something, with Peter Sellers

POSSIBLE TITLES FOR THIS POST:

"IT SUCKS, AND IT SUCKS REAL GOOD"
or
"ROCKY AND BULLWINKLE MEET THE ABDOMINAL SNOWMAN -- TO TAKE THE SNOWBOARD TO THE MOTHERBOARD"
or
"HOW TO STOP WORRYING, DROP YOUR VCORE AND LOVE YOUR BOMB"
or
"WATER-BOYS, TAKE HEED"
or
"DON' WORRY ABOUT THE CPU -- IT'S THE MUTHAH-BOARD, PINDAHO!! THE MOTHAH-BOARD!!"

"Introduction"

On or about noon on Sunday, July 1, Bonzai Duck got in his Duck-Mobile and went to the house of his brother, Dan-zoDuck. But before he left home, he wanted to certify an over-clock setting for his "Chrome Lightning" at a VCORE of 1.4625V -- 3.35 Ghz, 1,488 FSB and DDR2 744.

Upon his return, Bonzai was so very, very dismayed to find ORTHOS "STOPPED" with the big, red warning sign, after only two hours running at that setting in an air-conditioned room.

"Darn!" He quacked. "The only way to get there is to bump up the VCORE to 4.6875V -- maybe higher!"

And so it was that Bonzai made sure to include an OC profile on his motherboard for an FSB 1,334 setting and lower VCORE, after he certified the high-voltage setting for 12 hours.

IT'S NOT JUST THE CPU, PINDAHO!! IT'S THE MOTHERBOARD AND CHIPSETS!! THAT'S WHY THEY CALL IT "MOTHERBOARD DUC(K)_TING!!!"

This is like some . . . . Michelson-Morley experiment, where you get some measurement indirectly. It speaks to the wisdom of installing thermal sensors on North and Southbridges as well as memory. [And so would we have done so, if it weren't for those nefarious villains at Sunbeam-Tech and their silly, useless Theta-101 fan controller.]

Per the settings operative from the beginning of this thread, I dropped my VCORE back a notch "just to see."

I call it "partially certified" at 4 hours, 5 minutes, still running, but interrupted so I can post these results. AND -- AAAA-NNNDDD!! I've got some more hot ideas -- I mean uber-kuhl ideas -- about modding the ducting mod.

Check this out. If I go "water-cooling," it may be a while, because I don't think there's a hurry. This is short of my expectations, but maybe not so much -- and I may try and trim the voltages again -- with possibly lower temperatures for those that I can "see." And since water-coolers may overlook getting water-blocks for their chipsets, it offers another, cheaper option:

1st hour, ambient 72 to 73F, 1.4625 VCore

2nd hour, ambient 73 to 75F, 1.4625Vcore

3rd hour, ambient 75F to 76F, 1.4625 Vcore

4th hour, ambient 76F to 78F, 1.4625 VCore

For the CPU-cooler: someone suggested that even more CFM than a reasonably quiet, midrange fan could deliver, would capture even more gains from the TR Ultra 120 Extreme. Suppose you squeeze the motherboard duct's size down? It will pull air with greater velocity, cooling the mobo components even more. And suppose, knowing that, you open up a channel between the CPU duct and the motherboard duct, on the exhaust side of the cooler and motherboard, but on the intake side of the exhaust fans?

And -- it's also time to work on getting that VGA temperature back to 42C at peak gaming load. I can do it. I know . . . . that I can do it . . . . . I'm gonna do it!!

"Ya see? She's changing . . . . . ya see?" [the Red Dragon in "Red Dragon," a "Hannibal Lector" sequel.]

I think I'll drop the VCORE again before I get ready to do the swap-over to the Q6600.

That's all for now, folks!! You thirry-thavadg-eth!!

Bonzai Duck

 

[BonzaiDuck]

Same over-clock settings, but with VCORE at 1.4563V.

2 hours, 15 minutes, terminated deliberately -- no errors:

Another drop in VCORE to 1.4563V and still stable . . .

Here, I can see how people might push up their VCORE voltage when it only compensates for the heat generated by chipset and other motherboard components -- Certainly a needless risk.

Who cannot see now the prudence of motherboard ducting?

I really want to swap out the E6600 for the Q6600 sitting here, but I cannot resist continuing to search for the point where VCORE is insufficient for these clock settings. Note that the range of temperatures is now shrinking. If, for instance, the room temperature had been 70F, one would expect the range to be:

{ LowerBound = 34C - (5F / 1.8) = 31.2C, UpperBound = 45C - 2.8C = 42.2C}

I'll be eager to post additional results and pictures for the Q6600 and finished Lexan ducts. For those who've expressed an interest, stay tuned, and may the Duct be with you!

I'm going to drop it another notch, now . . . .

 

[bigpow]

Duck Dude, how many cups of coffee did u have?
I'd like to invite you to 'optimize' my rig..

 

[BonzaiDuck]

I actually read once, in a "wellness" newsletter before I retired, that caffeine is a brain-vitamin. I've been known to go through 12 shots of espresso in less than a day.

* * *

THE BONZAI-DUCK[T]-MEISTER VERSUS THE C2Q Q6600-MONSTER


REVIEW AND PROGRESS TO DATE

I left off at a point where I'd modded the main CPU-cooler duct and re-set the cooler in its more conventional, broadfaced orientation to the lower exhaust fan. This pre-empted (temporarily) the integration of a small duct which directed air from a 92mm fan on my VGA cooler (the TR HR-03-Plus) to the motherboard and out the upper exhaust fan. It also eliminated use of the 92mm exhaust fan. Use of the fan brought VGA temperatures on the 8800 GTS down to 42C, where they stayed rock-solid during game-play and 3dMark06 benchmarks.

The graphics card now remains rock-solid at 51C -- unwavering under load from gaming or other stress. I've built a duct extension over the top of the HR-03-Plus heatpipes drawing air from the component-side of the VGA card:

Extension to cover the HR-03-Plus to the component-side of the VGA

I cut several pieces of black foam board using the packing material for the cooler as a template-- no reason in the choice of the color except to make use of some scrap. Bonding them together forms a chamber that just fits over the top of the cooler's heatpipes and fins without touching, so there is a narrow, 1/4" gap around the edges for air to pass. I glued the box extension to the existing CPU-cooler duct, and added a panel which restricts airflow a bit around the CCFL light and intake-fan secured to the back of the HD cage:

Duct Extension Installed

I'm still a bit leery for restricting CPU-cooler air-input to the intake from this fan. The Seagate drives run at about 40C according to SMART, and I've yet to seal off the leakage through the holes in the hard-drive cage. But sealing off the cage may add to cool-air pressure inside the case, and force more air into the CPU-cooler duct where it's "in the face" of the CCFL intake fan, so we'll wait and see.

If this only caused VGA temperatures to drop a single Celsius degree, it offers hope in that a further duct extension and placement of a smaller fan between my 3Ware RAID controller and the VGA card can force air again past the VGA cooler and drop those temperatures back nearer to the 42C record, and also enhance cooling of the RAID card in the process. But at least the VGA card doesn't show any changes in temperature during intensive game-play or graphics benchmarks. It just hangs rock-solid at 51C. So here I'll say that there are circumstances where one can safely ignore ThermalRight's caution to put a fan on the cooler.

There has been a significant change to my configuration -- in the replacement of the E6600 Core-2-Duo processor with a Q6600 Core-2-Quad. The C2D had a stock thermal design power (TDP) of 65W while the C2Q's TDP at stock is 105W. Further, since there are four cores instead of two, the scaling of thermal power may not be linear with VCORE and speed.

Many thanks to AdairUSMC for providing information about stock load temperatures with his Q6600 in a ThermalTake Armor case with no modifications. Adair's load temperatures -- if I'm not mistaken, acquired in the mid-70's Fahrenheit room ambient -- ranged from 59C to 63C. There is no doubt that his case is well-ventilated. I forgot to ask him what his room-ambient was, although his remarks showed various tests at the same approximate ambient. But from his idle temperatures in comparison to mine, his room-ambient appears to be in the low to mid'70s. It could actually be lower, because he notes initial idle CPU values of of 32 to 34C, and mine are closer to 36 to 38C. These comparisons, of course, are for the stock 2.4 Ghz speed of the C2Q, and a VCORE that is either "Auto" or below 1.30V.

The Q6600 B3 stepping retail processor is given a recommended "maximum" on the box label by Intel of 1.35V. Other posts here on over-clocking strategies show a range of operation between 1.2+V and 1.325V. I'll return to this matter as I begin to explore my over-clocking strategy with the motherboard ducting enhancements.

THERMOMETER CALIBRATION FOR MEASURING ROOM-AMBIENTS

There is probably more than one way to measure the reference room-ambient. One could easily run a thermal sensor from the motherboard or other PC device to the front of the case at an intake vent, or simply use an analog or digital thermometer placed in reasonable proximity to the computer of interest. But the measured room-ambient is a reference value, and so differences between temperatures measured at "carpet-level" versus waist-level may not be important as long as the measurement is consistent. I'd have preferred measuring room temperature using a thermal sensor attached to a device inside the computer, but was unable to do that at least for now for certain reasons -- although it doesn't require a major effort.

The problems with the thermometer of my household thermostat are: first, the thermostat is in an adjacent room, the "CHrome Lightning" generates enough heat to raise temperatures in its own room, and the gradations of measurement on the thermostat are in 5F-degree intervals -- a problem of accuracy. I have another analog backpacker's thermostat that is tiny, and the gradations of measurement coincide with those of the household thermostat. So the only advantage of referencing both of those devices lies in their mutual confirmations of reasonable accuracy -- given the 5F-degree markings.

I decided to purchase a Radio-Shack digital thermometer with "remote" -- thinking that its accuracy to a tenth of a degree Fahrenheit would be of great advantage.

Radio Shack Digital Thermometer with remote sensor

I was right on that count, but wrong to assume that the device would be properly calibrated -- it was not. What seriously worries me is how long I fretted about this, when one might have skipped high-school physics and chemistry yet still been able to arrive speedily at a simple calibration solution. [I think I'm getting old.]

So to calibrate the Radio Shack thermometer after needlessly wasting energy like Chicken-Little, I went to the refrigerator ice-box, dumped as much ice as I could find into a tupperware bowl, filled the bowl with cool tap-water, sealed the Radio-Shack's remote component in a zip-loc bag, immersed it in the ice and water, sat down to watch TV with the Radio Shack receiver unit at my side, and waited. I knew the thermometer was off by approximately 5F degrees in the "high" direction, but I wanted to calibrate it to a tenth of an Fahrenheit.

The Radio Shack thermometer has a button on the receiver that cycles through the maximum and minimum recorded temperatures and the current temperature. The minimum was shown as 35.1F, so the correction factor would be -3.1, since water at freezing is 32F.


NORMALIZATION OF DATA AT DIFFERENT ROOM AMBIENTS

Various reviews, for instance comparison reviews of thermal pastes, normalize data sets taken at different room ambients for comparison purposes, and those reviews match my approach. If two sets of core temperatures are taken at the ambients of 73F and 76F respectively, then the second data set can be compared to the first by adjusting each and every observation by:

- 3F / 1.8 ~= -1.7C degrees

-- something that can be done in a few keystrokes using a spreadsheet program.

DIFFERENCES IN MEASURING SOFTWARE

Adair noted that he couldn't run CoreTemp in Vista, so his temperatures were taken using Everest Ultimate Edition. Since the use of different software has raised an issue about consistency, I ran both programs for at least a half hour to satisfy myself that they recorded identical core temperatures -- without exception.

CHANGES IN ROOM AMBIENT

Under laboratory test conditions, the computer would be placed in a chamber for which temperature control would be easy and automatic corrective adjustments would be swift, but none of us have an interest in that sort of accuracy at the moment.

Under "home" conditions, air-conditioners cut on and off automatically, but the adjustments are inexact. In my case, setting the target temperature to 75F may cause the unit to remain on until the thermostat's thermometer temperature reads 71F, and even the temperature in the computer room may drop as low as 72F. This requires either constantly resetting the thermostat, or monitoring the computer-room temperature and making note of significant changes. The only other alternative would be to check start and end temperatures for each elapsed hour and report the range of variation for each hour instead of showing a single value, or averaging the range upper and lower extremes. This also complicates any normalization of data sets for comparison.

THE TEST-BED

I show the essential components, without wasting time on expansion cards or matters peripherial to motherboard, CPU and VGA temperatures. The hard-disk subsystem uses a 3Ware 9650SE "hardware" RAID5 controller outfitted with four Seagate 7200.10 "perpendicular" 320GB drives and almost 9/10ths of a terabyte available storage. The drives sit in a massive cage with a 140mm intake fan pulling air through them. SMART shows the drives heating up to around 40C, so they add something currently to CPU temperatures only for the air directed from this fan toward the CPU-cooler duct. [And such factors are correctable.]

Intel Core-2-Quad Q6600
Striker Extreme motherboard, nVidia 680i chipset, BIOS revision 1004
Crucial Ballistics DDR2-1000 dual-channel kit SPD 5,5,5,18,1T (recommended maximum voltage 2.2V)
BFG-nVidia 8800 GTS

BASELINE DATA -- THE FIRST PASS AFTER INSTALLATION

For the stock settings, I took note of the BIOS monitor's report of a VCORE voltage under "Auto" settings, and then proceeded to fix the VCORE at about 1.28 to 1.29V. All other settings were left on "Auto" -- including the motherboard's default application of DDR2-800 speeds to the memory. The memory itself defaulted to about 1.85V

Here are the results (posted before in different threads) to provide baseline comparison with processor thermal behavior under stock settings.

Baseline 2xORTHOS stress-test -- small FFTs -- for C2Q Q6600 @ stock 2.4 Ghz, 1066 FSB

These are results showing extremely discrete results for each core, even though each core varies in temperature just slightly over two or three Celsius degrees. Core 0 and Core 1 seem consistently to run the warmest, and it is likely that this results from irregularities in the heatspreader cap, but one wouldn't notice these variations in the stock values. It turns out that the variations between cores are enhanced with over-clocking -- something we'd anticipate with increases in thermal power given the CPU-cooler's limited ability to deal with the increases, in addition to heatspreader irregularities or thermal paste application.

Note that the peak temperature for a Q6600 processor with my case and ducting mods is about 52C degrees at stock settings given a room-ambient of 73F. These initial room ambients were taken through comparison of two analog thermometers, and are accurate to +/- 1 degree Fahrenheit.

According to sources here at the forums, the cores are arranged as follows:

0 1

2 3

You can consult Intel for the Q6600 B3-stepping's construction, but I understand that it is essentially two E6600 processors incorporated under a single heatspreader.

VARIOUS COMPARISON STRATEGIES

Now that we have four cores, the bar-charts are just a tad more complex, and require four sets of average temperatures over the test-time-span instead of two.

Since there is consistent variation between the cores' individual temperatures, we could also average the averages to arrive at an "Average per core" value -- and these can be compared between different tests. Of course, the minimum and maximum range values are of special interest, especially because of the variation among cores. If, for instance, a test is going to provoke a failure with ORTHOS reporting a hardware or rounding error, we'd expect to see it happen more often -- or even exclusively -- on the hottest core. So our cooling strategy and concerns might be especially concerned with the peak load temperature for the hottest core.

FIRST OVER-CLOCKING TO THE MOTHERBOARD SPEC

The Striker Extreme is ASUS' flagship "gaming" motherboard for the nVidia 680i chipset. It has a few cousins, like the P5N32-SLI, and across ASUS' market-segment categories. I once saw a customer review criticize another buyer, saying "If you want a SERVER board, get a SERVER board. This is a GAMING motherboard!!"

But if it walks like a duck and talks like a duck, it can be any duck you want it to be. I wanted it to be a board capable of SLI with the 49 PCI_E lanes, with concurrent x16 PCI-E graphics, and a slot that could accommodate a hardware RAID controller in either x8 or x4 flavors. This board also features a PCI-E x1 slot, which might be used for a TV-capture-card, for example, and there are two PCI slots. Despite "getting what I wanted," I had to go through some hair-pulling per the orientation of the TR HR-03-Plus VGA cooler. As I said earlier, in one installation option, it hangs over the PCI-E x8 slot. If I install another cooperative graphics card, putting another VGA cooler into this very large case will still choke off one of the two PCI slots.

The motherboard specification offers 800, 1066 and 1333 front-side-bus speeds. Since the memory is rated another 466 Mhz above the DDR2 rate matching a 1:1 correspondence with this motherboard spec, there should be no trouble running it at that speed, either. So the only question remains, "Can this toaster-oven processor get to 3.001 Ghz?" We know that it can. But the thermal load temperatures come so close to the throttling point, one wonders if it's worth it.

Let's see what, if any, headroom the ducting mod provides:

Load temperatures over 12 hours, 73F to 78F, Q6600 @ 3.001 Ghz, 1,334 FSB, DDR2 667 and 3,3,3,6,2T Timings

I didn't stay up all night to note the temperature deltas every hour, on the hour, but this double-ORTHOS-run started with the room temperature pushing close to 80F degrees. So for those values, you'd normalize them to a 73F equivalent by deducting 5.8F / 1.8 ~= 3.2C from the peak values. So at 73F, the peak value for the hottest core would be close to 54.8C -- not more than 2C-degrees higher than the peak value under the stock settings for the same double-ORTHOS stress-testing. And of course, at 80F, you'd barely see an increase to 59C for the peak value. So, given the higher temperature peaks in good cases with minimum VCORE settings, you'd feel pretty safe with this over-clock -- even in midsummer with the household air-conditioning turned off.

I'm currently certifying another over-clock with the multiplier dropped to 8, and the load temperature values look pretty much the same. I'll post those results when I have them, but it's been running some six or seven hours so far. In the interim, here's the sample observations for a test at lower VCORE of 1.300V -- which was terminated before it had a chance to fail. The room ambients for this 2xORTHOS run ranged from 73.5F to 76.8F:

Preliminary test run at multipler = 8, Q6600 @ 2.88 Ghz, FSB 1,440, DDR2 720 and 3,4,4,8,2T

Minor increases in VCORE settings under the 1.35V maximum seem to have negligible impact on the monitored values, although they could impact the load voltage readings and reduce the droop -- I'm just speculating here. But at the 2.88Ghz (multiplier=8) and FSB at 1,440 Ghz (DDR2=720 and 3,4,4,8,2T) -- I've recaptured much of the memory speed I had with the E6600 at a 9 multiplier, and only lost about 100 Mhz from the earlier CPU over-clock.

Trying to push to 2.92 nearly gave me a coronary today -- I had to reset my CMOS to get the machine to post again. There may be trouble OC'ing a B3-stepping Q6600 to above 3 Ghz at a multiplier of 9, but I will cautiously continue to tweak it. Even so, since these over-clock temperatures I've shown so far wouldn't make the processor throttle, it may not be temperatures which make it troublesome.

One more thing on the temperatures. The ducting around the CPU-cooler was done as much to provide a duct-cover for the motherboard as to separate the CPU-cooler from the VGA cooler. It also makes it possible to avoid hanging a fan on the cooler itself. And I've shown that there is a positive impact on over-clocking due -- not just to CPU cooling -- but especially to cooling the chipset and the motherboard.

Now -- for all this trouble -- one could insinuate (a) that the B3 stepping is a "bad investment," but I could wonder what the opening price on the Penryn processor will be; and (b) -- this is a lot of trouble to take with air-cooling, when I could have a water-cooled rig capable of higher temperatures.

But this opens up a lot of options. When the Penryn eventually finds its way to my rig (some while after it's released and I can get it for a lot less), I should be able to post some frosty temperatures with it. If I decide to water-cool, I can enhance it by ducting the motherboard (adding a few holes or slots to my existing ducts, or modifying some Lexan panels that I may install in the interim.) It's a lot of trouble, but it works.

And I may already have said it, but the overclock peak temperature values are only about 2C above those for the baseline stress-test at stock Q6600 settings.

 

[Amart]

STICKY!

Very interesting thread.
Thank you. ^_^

 

[BonzaiDuck]

Amart -- Thank you -- that is very flattering. I wouldn't be embarrassed if it got into a sticky, but wish I hadn't been so uncontrolled in my exuberance per the first results on the E6600 processor.

I haven't added anything to this thread since the first results on adding the Q6600. Since then, I've had a chance to look more closely at two things: A C2D Temperature Guide at THG, and several reports by others on their CoreTemp load Celsius values.

C2D Temperature Guide

There may be aspects of the quad-core that would be exceptions to some few statements in the article, but the thermal guidelines for the processor are published at Intel's web-site, so I think the basics very much apply.

Also, I think the baseline load temperatures for the stock settings are around 52C at these room-ambients (with the ducting), and would probably be closer to what I'm showing now without the ducting.

As far as I can tell, the ducting got me a 10C improvement in the temperatures. I'm currently stress-testing with 2xORTHOS at these settings:

MULTIPLIER 9
CPU_FSB = 362 Mhz
FSB = 1,448 Mhz
DDR2 = 724 Mhz, with timings 3,4,4,8,2T [very likely rock-solid at 1T]
VCORE 1.4318V
1.2VHT = 1.35V
CPU_VTT (FSB) = 1.40V
NB_Core = 1.40V
SB_Core = 1.50V

Monitored values:
VCORE = 1.38V and 2.22% over retail-box maximum (1.35V) under 2xORTHOS load
= 1.43V and 6% over retail-box maximum at idle
1.2VHT = 1.42V
CPU_VTT = 1.47V
NB_Core = 1.42V
SB_Core = 1.52V

HERE ARE THE TEMPERATURES AT A CURRENT ROOM AMBIENT OF 73.4F (23C):

CORE 0 = 62C
CORE 1 = 57C
CORE 2 = 61C
CORE 3 = 58C

There is no longer much of a statistical distribution on these as there was with the E6600, but they of course scale in a predictable way as the room ambient increases. Thus, at 80F room ambient, CORE 0 would be between 67C and 68C -- lower than temperatures I've seen reported for mid-70's room-ambients with other forum-members' systems running at 3.0 and 3.1 Ghz.

However, one must prove the system stable at those settings, even though the actual (and real) temperatures are solidly predictable to within a degree Celsius, but I should have that done today. It was my goal to at least get to 1,440 Ghz FSB at multiplier 9, so this is going to do it for me, I think. After I've finished confirming stability, I'll drop back from 362 Mhz CPU_FSB to 360 Mhz.

There may be some minor -- even noticeable -- improvement in voltage required at some speed setting with the cooler temperatures, but even with water-cooling, we will not super-cool the circuits to eliminate resistance to the point where these voltage advantages would be overwhelmingly significant. They may be significant enough for some higher over-clocks, but one should be wary about voltage settings too far out of spec. Just to begin with, the thought of pushing molasses through an eye-dropper is an appropriate analog to voltage and 65nm processors -- if you put the eye-dropper under too much pressure, the glass may break!

I believe the advantage lies elsewhere -- in the predictable impact of heat on the processor. Another THG article in their recent "over-clocking" series, points to two things. Increases in temperature require increases in voltage, but we've explained the simple aspects of that issue. Second, increases in temperature at some level (65C for the TCase temperature, with Tjunction (those shown above) being 15C greater than TCase) -- may cause separation or degradation of the contacts between the IC silicon die circuitry and the metal components.

Further, for the silicon-die parts of the processor, some increases in temperature may improve conductivity and performance, while increases in temperature to the metal components will decrease conductivity and performance due to increased resistance.

For this, that is how Intel chooses for the C2D processor a TCase threshold in the low 60's Celsius. Here -- and I've yet to measure it but I think I'm pretty accurate per the 15C-guideline and previous experience -- the TCase for my rig (currently under stress), is probably around 47 to 48C degrees.

I am very conservative about over-volting processors. Here, I had been reticent to simply exceed the retail-box maximum, knowing that the most significant factor is the load voltage, which here, is only .03V in excess of what's written on the box. But these percentages, while they might void a warranty, should not be risky from a statistical or probabilisitic standpoint, or at least, that's my opinion.

I just wish I'd got the G0 stepping, but this is STILL gonna be schweeet!!

 

[tylerdustin2008]

Very nice work. But no way in hell would i ever type that much. You have to much free time...

 

[BonzaiDuck]

Well, I shouldn't need to explain myself. Some newer acquaintances and political associates assume that I spend all my time writing this stuff, or that it's some sort of . . . . need for attention.

But I just come back to the keyboard several times a day, jam out my thoughts, and go back to the vegetable garden or whatever I was doing.

It's not a lot of effort. I feel regret and sadness for the burdens I put on my Chinese students in the '90s. I'd go into the lecture hall on a Sunday morning, pumped up with a quadruple Espresso in my travel-mug, and start talking.

You've got to imagine such a situation as much worse, because the listener could not just take the time to do a leisurely read . . . .

As for me . . . . imagine the baggage that comes with this territory . . . . :laugh: