Direct-Die water cooling

[timbob2469]

I have to agree with johnny6722. Everytime I think I'm going to pull the trigger and get a waterblock, along comes Thermalright with another great heatsink. The latest I just bought is the SP97 . I have it on my Epox 8RDA+ v1.0 running an xp1700+ JIUHB @ 211 FSB X 11.5=2427 MHz. I have been running F@H for 3 days straight, 24/7, and it stays at 41° C. I had a Zalman flower (pctoys cp6000) on there and it ran almost 10°C hotter with the same fan. Chipset voltage set to 1.9v (mod), and cpu is set to 1.85v. Room temperature is a comfortable 20°C (68° F).
I am running the previous model, the SLK900u on this my main rig NF7-S v2.0 xp2500+ @200FSB X 12=2400 MHz ,42.5°C .And it has been folding most of the day (@100%). I think the Zalman 92mm fans are great with these heatsinks, I run them full blast 1750 RPM and they are a lot quieter than my wifes rig with its xp2400+ and 5000 RPM fan. Just today she mentioned again that , "I really notice how loud my pc is because after I shut it off, the room goes quiet again". I think that is a hint to get yet another ThermalRight. I just can't justify spending either a lot of money or time for a little bit more speed. I usually put 1 80mm intake and 2 80mm exhaust fans in my cases, except for the NF7-s , I put 2 80mm fans in the front of the dual Raptors (which together cost probably as much as a "starter" h2o setup).
Hypothetical question: Where is the bottleneck or slowdown in a system?
Find the answer to that by widening everything.
Bob

[Daniel ~]

Thats what makes horse races! ":O}

I'd be interested to know how many of those who've tried water cooling with a decent rig, have gone back to air?

[Gizmo]

Originally Posted by johnny6722

While your fabrication skills are impressive to say the least, and not to throw a damper on your post I have to ask a question:

Is it worth the time, cost and risk for everything you have had to do for 2.5 mghz and borderline stability? I am running an air cooled xp1700 at 11 x 228 (2515) and my idle temps are under 40C and my full load never gets anywhere near 50C, and it will eat P95 all day long.

I have been seriously considering water cooling, but after all the research I have done and posts I have read, I see very little benefiet other than noise reduction, but alot more risk in general. Temps at at or even above where I am now with most water systems I have seen.

Ummm..............................no.

But that's not the point.

Despite the fact that the project has not yet yielded that kind of results I had hoped for, I am still hopefull. In addition, I have learned considerable about how my computer really functions, as well as aquiring new skills (such as working with plastic and learning how to braze) that may come in handy at some later point. And of course, it has been fun! In addition, if I DO happen to come up with some kind of breakthrough, it will all be documented right here for the rest of the community to take advantage of. Further, it might even end up yielding economic benefits, like being able to sell something. (OK, that last is a stretch, but you never know)

[Gizmo]

Originally Posted by johnny6722

My research has turned up some very dissappointing results with water. Even actual studies done where they compare the results with that of an SLK-900 on the same system, often its the same, or within a degree, maybe 2.

I was in the hopes that water would actually net a 3 to 7 degree (C) reduction, but I have yet to see any results, private or official, that show this.

Don't get me wrong, I certainly was not trying to rain on anyone's parade here, I was just noting the huge risk involved for little or no gain. But Gizmo obviously has exceptional knowledge and the skills to match. Fabricating your own water cooling system is one thing, but your own Direct-Die water system is incredible.

I am still considering water, if I can find a system that will net some kind of decent gain (at least 3C).

If I had the money to throw down on it, a Prometia would be the way to go, but at an price that equals that of a new system, its very hard to justify. Ahhh, but what I could'nt do with one!

As I mentioned elsewhere, this is actually stage one of my project. Eventually, this is all going to be cooled with a cobbled-together refrigeration system. I hope to be able to chill the CPU, chipset, GPU, and maybe RAM with it, and I also hope to get the CPU to -20C. But that is a bit in the future yet.

As far as the results of water vs. air, there is a LOT more to building a water cooling system than just getting a block, rad, pump, and some hose to stick it all together. If you make a poor choice in components, you can end up with dissappointing results. Even if you buy good components, you can end up in situations where they don't all work together well. This means that researching everything is extremely important.

IMHO, even though you can buy kits that have everything in them already, that isn't the way to go. Why? Because those kits are kind of like production cars out of Detroit: They are designed to work in ANY system, ALWAYS. That means that they make sacrifices in performance in order to work across a wide range of systems reliably.

If all you want is a quick and dirty way to get some good cooling, then air is definitely the way to go. However, water cooling can definitely provide better results than air (when properly implemented), based on what I have seen. As has also been pointed out elsewhere, a lot of people are doing water cooling, not so much for performance reasons as for noise reasons. The ability to increase the surface area used for cooling (and thus use large, low RPM fans) is very attractive from a noise management perspective.

While I thank you for your complements regarding my skill and knowledge, I should point out that I am not doing anything terribly unique. People have been doing Direct-Die water cooling all over the web for quite some time now. The work of pioneers such as Spode and SurlyJoe helped give me the incentive (and guidance) to make my own attempt.

[whatever]

Originally Posted by gizmo

As far as the results of water vs. air, there is a LOT more to building a water cooling system than just getting a block, rad, pump, and some hose to stick it all together. If you make a poor choice in components, you can end up with dissappointing results. Even if you buy good components, you can end up in situations where they don't all work together well. This means that researching everything is extremely important.

agreed...

IMHO, even though you can buy kits that have everything in them already, that isn't the way to go. Why? Because those kits are kind of like production cars out of Detroit: They are designed to work in ANY system, ALWAYS. That means that they make sacrifices in performance in order to work across a wide range of systems reliably.

nice one....

you said it so well. i like it..

[Gizmo]

Well, I was gonna post some more pics and stuff, but right while I was typing in the new message, one of my water lines came loose (I didn't have everything bolted down real well because I was running an experimental setup) and gave my system a thorough drenching. Now my system won't boot. I'm hoping that letting it dry out overnight will solve the problem.

A few things I did find out:

1) There appears to be some critical distance from the core that is necessary to get good water flow. This critical distance seems to be a function of the water pressure and flow rate of the jet. Once you are within this critical distance, changing the distance doesn't have a significant effect on the temp until you get so close to the core that you reduce water flow significantly.

2) Playing with the jet configuration enabled me to get to 2586 Mhz, but I had to crank the VCore up to 2.1v to get there. This cpu appears to be a real pig to OC.

3) At the default speed setting for the CPU, I have a temp rise of 10C over my water temp (water temp based on thermometer in water, CPU temp based on socket thermistor). At the OC setting of 2584, my temp rise is 30C over water. Doing the math on the difference in watts vs the difference in temp rise yields a C/W of .2 for this configuration. However, in doing some underclocking measurements, I was able to get figures ranging from .1 to .2 C/W. In other words, my heat dissipation is non-linear. I'm picking up heat from somewhere else in my system. Maybe heat is being conducted from the switching transistors from the VCore supply to the CPU thermistor? Dunno. Need more info, and a better way to check the CPU temp. Is anyone aware of a mod that can be done so that the on-die temp diode can be read by the Winbond chip?

Edit: After I posted this, I came up with one, which you can find here.

I'll let you know tomorrow how the system has faired.

[Staz]

lesson #4) always use hose clamps.

I have heard that putting your board in a 160 degree oven for a couple of hours helps.

BTW at those voltages your motherboard voltage regulators may be getting mighty hot. Here and here are some articles on how to cool them off, or may just want to put som heatsinks on them.

[Gizmo]

Originally Posted by Staz

lesson #4) always use hose clamps.

I have heard that putting your board in a 160 degree oven for a couple of hours helps.

BTW at those voltages your motherboard voltage regulators may be getting mighty hot. Here and here are some articles on how to cool them off, or may just want to put som heatsinks on them.

Errr......yeah. I just hope learning that lesson didn't fry my mobo. I tried it this morning, with the same results. With no RAM in the system, I get one long beep. This is to be expected when there is no RAM, but it at least lets me know that the CPU and some basic BIOS stuff is coming up. However, when I plug the RAM in, I get nothing. Zip. Zilch. Nada. I was gonna bring the RAM to the office and test it in a system here to see if it would work, but forgot to. I hope all I've done is toast the RAM.

As for the VReg, yeah, I know they are getting hot. My infra-red thermometer shows 80C on those switching FETs, which is right toasty. I've seen the two articles that you mention above, and was considering something along those lines, although I thought I might try some basic heat sinks first. However, for the heat sinks to really be effective, I'd have to desolder the surface mount transistors, rework the mobo to take the heat sinks, and then figure out how to remount the transistors.

[Gizmo]

Well, CRAP!!!!!

#$%^%@!%$@!!!

My mobo is officially umm........................toast. Gonna have to get another one.

[Daniel ~]

Another martyr to our cause! Sorry gizmo R.I.P!

[Staz]

Originally Posted by gizmo

My mobo is officially umm........................toast.

You could always make a clock out of it.

[Gizmo]

We are gatherd here today to remember our honored dead.......it went before its time, and it will be missed.

I just ordered an Abit AN-7. I also went ahead and popped for two sticks of OCZ Enhanced Latency PC3200 matched DDR (512 MB total).

Hopefully, by the time the wife finds out about it, I'll be able to beg for forgiveness.....

A carefull post mortem on the other board revealed some very fine copper filings, probably from when I built the new water block and didn't get the copper tubing completely cleaned out. I imagine this is what killed the mobo, not the water itself. This is only like the 7th or 8th time I've dunked this mobo. (BTW, it's AMAZING how fast a 700 GPH pump can empty two quarts). I'd post pics of the new water block that caused all the problems, but the pics are on the drives attached to the machine that doesn't work. Anyway, I hope to have the parts in time to play this weekend.

[Daniel ~]

Sail on! ":O}

[Gizmo]

I've been poking around some. Maybe I'm being a bit harsh on my CPU's OC potential. Bearing in mind that the 2500 barton and up CPUs are all basically the same core and same line, is 2.586 Ghz at 2.1V reasonable? What have the rest of you experienced with this? BTW, this is a week 13 chip.

[Daniel ~]

My unlocked 2500 Barton (EPoX 8RDA+ 1.1) hits the wall at 2.3 gigs. don't matter if you come at it with x11.5 or x9.5 after 2.3 instabilities start showing up. I'm at 1.95 VCore and 2.90 Vdimm

[Áedán]

Originally Posted by gizmo

Need more info, and a better way to check the CPU temp. Is anyone aware of a mod that can be done so that the on-die temp diode can be read by the Winbond chip?

Only way I know of is to cut the traces from the attansic thermal protection, and bring them over to the winbond chip. You might need to put a replacement diode across the attansic chip to keep it happy though.

[Gizmo]

Well, got my AN-7 yesterday. I gotta say, I really LIKE this board!

No socket thermistor, reads temps right of the CPU diode. Unfortunately, neither MBM nor 8rdavcore can handle the board, so two of my favorite utilities are useless. On the other hand, it does come with a utility called uGuru that replaces some of the functionality of those two utilities.

So, it monitors CPU, Motherboard, and PWM temps. The PWM is a nice touch, 'cause it lets you see if you are in danger of suffering a nuclear meltdown on your VReg for your VCore. It also has fan controls and speed monitors built in so that you can adjust the speed of the fans on the CPU, NorthBridge, and System Fans.

All in all, a nice upgrade from the NF7-S. And it only cost me 1$ more at NewEgg.

BTW, with this mobo, I'm running stable at 2600, with a VCore of 2.15, using my water cooler. That's 134 Watts at full load. And my temp never gets over 45C. My VReg does hit about 75C though. I'm definitely gonna have to do something about that.

[Gizmo]

Ok, so here's the pics I was gonna post when last I skragged my mobo.

First pic is of my CPU when I pulled it out of my previous setup (the fixed position water jet). Notice the discolored area on the right side. I'm not quite sure what this is, but it seems to be baked on. I actually had to sand part of it off, although most of it simply scraped off. Careful inspection of the CPU does not appear to be the result of corrosion. That's just a guess, as I'm no metalurgist. However, it suggests that I'm not getting enough cooling on this part of the chip. Since my existing jet is symmetric (covering all of the core equally), this suggests that I need to make the jet a bit asymmetric to divert more water over this part of the core. That actually makes sense, since it seems like about half of this core is devoted to the cache, IIRC.

Since I was building a new water block for testing purposes, I thought I'd try to document some of what I was doing. Unfortunately, the pics didn't come out very well. This pic is of the bottom piece of the cube, which mounts directly over the CPU core. Notice the rectangular cutout for the CPU core. Notice also, that I have beveled the edges of the cutout. This helps water coming off the core to get back up into the block efficiently and be removed.

Here we have pics of two different jet arrangements that I wanted to try. The left one is basically a rectangular nozzle coming from 1/2" (12 mm) copper tubing and narrowing down to a size and shape similar to the core. This design sucks . The one on the right is my 8 mini-jet rig. Notice the ring around both nozzles. This ring allows me to attach an O-Ring to the nozzle, so that I can slide it in and out of my block without leaking water. This is how I was able to play with nozzle height placement. It also allows me a moderate amount of side-to-side and up-down movement as well, thus allowing me to place the nozzle at any point within a fairly limited 3D space. After looking things over, I took 4 of the jets off one side of the the 8 jet arrangement, which allowed me to focus all of my water onto one half of the CPU core (the half that gets hottest). BTW, please don't make fun of my brazing skills.

Finally, this is a pic of the two units as seen from the top.

[Gizmo]

BTW, getting the groove for the O-Ring into the copper tubing was a real PITA. I don't have any tools for doing that, so what I did was take my square-shank screw driver and roll it across the tubing using LOTS of pressure until it created a groove. The trick is to use enough pressure to groove the tubing without actually distorting the tubing. Since I was using soft copper, I actually had to heat the copper and put a slight temper into it to help it retain its shape.

The adjustable rig was made using a hose bushing for a kitchen sink water hose (you know those spray hoses they have next to the faucet) that was cut down. The 1/2" tubing fit inside this with about 1/16" (about 1.5 mm) to spare all around, so then I got a 7/16" (about 11 mm) O-Ring, and grooved the tubing until it would fit tightly into the bushing when fitted with the O-Ring.

[Gizmo]

BTW, pulling .1 degree C/W with this rig. As far as I know, there ain't a water block out that will do that. I'd like to build one up and send it off to Overclockers and have them bench it to see how it fairs against the others on the market, but I have no idea who to talk to.