lapping my Ultra-120 Extreme (pics and temp data)

[graysky]

After about $20 worth of sandpaper and 4 hours later, I have a somewhat shiny Ultra-120 Extreme that really sucks at cooling -- this whole experience was a complete and total debacle. I have a Q6600 @ 9x333 and my pre-lap temps (load) were as per TAT:

60
61
55
55

After I lapped it the first two cores are >80C. I really blew it! I think must have taken too much off. The material came out of the center first suggesting the thing is convex. It took about 2-1/2 sheets of 220 grit to get all the nickel color off the thing. Then I went up to 400, then 800. It's not a mirror finish, but it is pretty shiny with a few fine scratches.

I didn't apply enough AS5 to the lapped HS originally. I used the same amount I did for the "stock" HS which clearly isn't enough. I put on a FAT line this time through, and my temps are the best yet:

The data as measured in speedfan.exe for a ~1 h x264 encode (uses all 4 cores with a CPU load of >99 %). I made speedfan log the temps (which it does every 3 seconds) and I averaged the whole data set per core for the 2nd pass of the 2-pass encode (the 2nd pass is the most CPU intensive). Room temp for both experiments was ~23 C.

Before lapping:

Core 0: 51.9
Core 1: 51.4
Core 2: 45.6
Core 3: 45.6

After lapping:

Core 0: 49.9
Core 1: 49.4
Core 2: 44.0
Core 3: 44.4

Delta:

Core 0: 2.0
Core 1: 2.0
Core 2: 1.6
Core 3: 1.2

 

[Henny]

Sorry to hear but you used 220 grit???

I start at about 400 grit and go to about 1500 or so.

What kind of surface did you use? You really need a granite surface plate to do it right and you need to make sure you're holding the HSF totally flat while lapping. It's easy to take too much material off the outer edges.

Is your CPU totally flat?

If you want to scrap it out send me a PM. If the price is right I might like to try salvaging it.

 

[graysky]

I used a pane of glass; I believe it is flat, if I put a drop of water on the glass and then place the heatsink on the water drop, it looks very uniformly distributed from the bottom, looking up. I also was very careful not to apply pressure on the thing while lapping it. I dunno if the CPU is totally flat; I won't be messing with it

I dunno, once I calm down I might try using to make a mirror out of it or something. I'll post some pics of the process I used.

 

[graysky]

Here are some pics of the process (they're all in one 1.2 meg file - sorry for your analog modem folks). Anyone tell me what I did wrong?

Pics

 

[Henny]

I think you're just not making good contact with the mount or you used too much thermal compound. Here's what I'd do:

1. Make sure you're using thick plate glass. (single pane glass isn't flat enough and deflects slightly).
2. Get a black marker and completely cover the HSF base. Put some fine sandpaper (ie 800+) on the glass. Now gently scrub the HSF on the sandpaper using very small movements. Now turn over the heatsink and look to see that the black ink is being removed uniformly. If so then it's flat.
3. Now remove any compound from the CPU and HSF using IPA.
4. Apply just a tiny ball of compound in the middle of the CPU. (about the size of 1/2 a grain of rice)
5. Now apply the HFS and rub it onto the CPU. Tighten the HSF mounts by alternating from opposite corners and tighten it just a little bit from each side. You want it to be tightened down uniformly.

Good luck

 

[Smonkey]

Are you sure you just didn't get a bad mount or too much compound?

edit: too slow on the post : )

 

[graysky]

@ Henny: I don't think flatness is the problem. When I took it off it seems as though there are some marks on the HS now that correspond to the CPU mounting thing implying that the HS is too high. I never use too much AS5, when i inspected it after removing the HS, it had little peaks on it sort of like the top of a Hersey's kiss also consistent with poor contact e.g. heat sink is now too high.

If you look at my pics, I think you can convince yourself that I didn't take off an excessive amount of material. Opinions?

 

[Zardnok]

You say that the Nickel came off the center of the HS first which means the center of the HS was convex. Most C2D processors have a problem being concave, which means the outside edges are higher than the center area. Perhaps your HS matched up well with the convex HS fitting into the concave CPU heatspreader. Now that you have lapped your Ultra-120 extreme, it no longer fits down into the center concave area on your CPU. Perhaps you need to lap the CPU as well to insure perfect contact again.

Edit: If you decide to scrap the Ultra-120 extreme that is lapped, let me know and I may purchase it from you. I have been wanting to compare one to my Titan Amanda TEC cooler.

 

[eternitykh]

hm... that's some interesting results, but ya doesn't look like you've taken off excessive material, perhaps your cpu is concaved, and your heatsink WAS convexed, and now that it's flat, it's not contacting the center of the cpu properly (so in other word, it's gone from bad contact to worse contact). i would try lapping the cpu also, but it sounds like you've had a bad experience with lapping and probably dont want to lap anything else again, so that might not be an option for you.

have you tried reseating the heatsink? same results?

you should try printing out the poor man's flatness test grid, and just to double check if it's flat. if it IS flat, and you have properly applied thermal compound then i can say it's probably the cpu being concaved.

@henry, 220grit is fine, i started with 200~220 myself, as long as the 400 is done properly after that. which looks to me light graysky did.

 

[graysky]

Originally posted by: Henny
2. Get a black marker and completely cover the HSF base. Put some fine sandpaper (ie 800+) on the glass. Now gently scrub the HSF on the sandpaper using very small movements. Now turn over the heatsink and look to see that the black ink is being removed uniformly. If so then it's flat.

Yeah, I got bored and tried this market test. After about 5 little circles on my 800 grit paper, all the marker was removed so it is indeed flat.

Check out this pic of the HS after it was removed. Notice the hard marks in the rough shape of the CPU mounting bracket. It seems to me like it's hitting the bracket which is preventing it from making proper contact w/ the CPU. This wasn't the case before I lapped it. I'll bet they're designed to be convex due to this.

 

[Henny]

OK, I'd suggest you check the CPU for flatness. Someone suggested using a razor blade edge which sounds like a good idea.

 

[RallyMaster]

Should've just left it as it is. Don't fix what ain't broke mmkoiiii.

 

[aigomorla]

If your base is flat. And im pretty sure it is from all that work you did:

Then lap your CPU.

But start at 400 and not 220. and go slowly.

BTW: Autostore FTW!! thats the only place you'll find 1000+ grit sp.

 

[RallyMaster]

my local ACE Hardware sells 1500 grit. I think I'm gonna get some of that soon :beer:

(currently, my F64P was done I think on 400 grit?)

 

[graysky]

Everything is good, see the first post of the thread which I edited. Thanks to all who replied.

 

[eternitykh]

=) good to hear everything worked out!! hope u can still cancel that new HS order in time!

 

[thecoolnessrune]

But you dont want to perfectly mirror your CPU I heard.. I personally wouldn't use over 800.

 

[Zardnok]

Autozone and Oreily's both have a $3 multi-pack of sandpaper with a sheet of 220, a sheet of 400, and 2 sheets of 800. That should be more than enough to lap a CPU with.

 

[aigomorla]

Originally posted by: Zardnok
Autozone and Oreily's both have a $3 multi-pack of sandpaper with a sheet of 220, a sheet of 400, and 2 sheets of 800. That should be more than enough to lap a CPU with.

i like to go up to 1500 grit. it leaves a nice mirror finish on it.

 

[graysky]

Originally posted by: eternitykh
=) good to hear everything worked out!! hope u can still cancel that new HS order in time!

I did

 

[JAG87]

now lap that chip!

 

[TankGuys]

It's bizzare that using MORE AS5 helped your temps. That suggests to me that it wasn't mounted very well in the first place, which is why it looks better now. The whole point of lapping is to create better contact between the IHS / Heatsink, thus requiring *less* thermal paste. So something doesn't add up

In any case, glad it's working well for you now, and that you don't have to buy a new one!

 

[graysky]

@TG: Me too, thanks!

 

[graysky]

Something I just discovered is that the AS5 got cooler overnight. I just finished lapping the CPU and had to remount the HS. My load temps for the hottest core were hanging around 48 which had be bummed out since they were better before. I figured I'd like to run x264.exe overnight working on files and that I'd pull the HS and remount in the morning. The room temp didn't change that much; 1C maybe, but the load temps dropped from 48 to 42-43 which is below where they were before. Bottom line is something happened with the paste that helped the temps. I'll keep an eye on it over the next few weeks to see if there's an additional drop. People claim 2-5C after break-in which can be 200 h.

 

[Ambress]

Hello all,

This thread caught my attention after having just purchased and installed a Thermalright Ultra-120 Extreme in my system, to replace a less efficient Zalman CNPS9500. Although the Zalman has worked fine and allowed me to stably overclock my Core 2 Extreme to 3.6GHz, that also required the fan to run at max speed all the time, thus making for an acceptable but nonetheless louder noise level. Once I learned of the Intel Thermal Anaylsis Tool and started looking at core loading temps more, I found the Zalman cooler couldn't keep the CPU below Tjunction until I reduced the CPU clock from 360 to 350Mhz and dropped Vcore from 1.4v to 1.375v. With those settings, the Zalman maintained a pretty stable 77-80°C temp on each core.

Now, what I've done is more than a cooler swap...I also removed the stock Gigabyte 1000rpm "silent fan" which had so low an airflow that I couldn't feel much air movement outside the case grill, and installed a SilenX Ixtrema Pro IXP-76-18 (manuf. claims 90cfm/18dBA) in its place. On the cooler, I'm using an SilenX IXP-74-14 (72cmf/14dBA). This greatly improved the airflow and cooling, yet also improved the noise level...no surprise, as the Silenx fan on the cooler is only running at 1250-1300rpm rather than the smaller Zalman fan's 2500rpm. I had a very quiet Enlobal fan mounted on a side screen panel, and moved the Gigabyte fan to there, as the Enlobal was becoming prone to a rattle or chatter.

Since installing the cooler required pulling the mobo, an ASUS P5W DH Deluxe, I chose to remove, clean, and prep the NB/SB heatsinks with Artic Silver 5. Glad I did...I don't know how hot the SB gets, but it had only a line of thermal material along two edges, so contact wasn't as good as a full surface treatment with AS5.

I'm getting too long-winded here, but moving on to the Ultra-120 Extreme, I didn't really notice any concavity of the base. Admittedly, I didn't examine it all that closely and am not sure exactly what folks mean about the concave base, but I was expecting to see two opposing edges that were higher than the other two, such that one could see an air gap under the shorter edges. I didn't see that. I did find it intersting that a NewEgg customer review quoted a reply about the concavity they'd gotten from Thermalright, saying it was "by design", as only the core area of the processor needed firm contact with the heatsink base, and not the complete surface of the CPU. Sounds like a flaky reply to me, but maybe there's truth in it.

In any case. the result I'm seeing is a significant improvement. Running my X6800 at 3.6GHz (360x10) and a Vcore of 1.4v, and loading both cores in TAT at 100% for 20 minutes, the core voltages stabilized at 75-76°C, 76/71 being a common result. The CPU temp held stable at 57-58°C and system temperture at 31°C (these values per SpeedFan). Curiously, while SpeedFan and CoreTemp both generally showed the cores at temps within 1-2°C of each other, TAT often showed as much as a 5°C difference and both core temps in TAT were 1-3°C higher than SpeedFan or CoreTemp. Perhaps just a variation in the sampling rates? During routine use, I'm seeing my CPU and System temperatures are commonly about 10°C below what they were prior to this mod, so I'm happy. Meanwhile, 20 minutes of 100% load testing exceeds what I'd ever anticipate tasking this PC with, even if rendering video or doing photo editing, so I think it will quite happily live at this level of overclocking without any problems. I did try pushing it higher but couldn't get a successful boot over 3.7GHz without some driver failure. At 3.65GHz, I got a successful boot but failed the 100% load testing with a BSOD after about 7 minutes. So, 3.6GHz it is...not much gained in speed over my original setup, but nice thermal/noise improvements. And, I can't complain about a 23% air-cooled overclock.

Daryl