Lapped my FX-8350, 4.3GHz OC'ed temps lowered by 9°C

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
Lapping one's CPU and HSF is old-hat for many folks around here, it produces the kind of "oooh! shiny!" pictures and eye-candy appeal that makes it fun and rewarding to do (not too mention that it does happen to lower your operating temperatures as well, purely an unintended consequence of the lapping process ;)).

I was bitten by the gottalappit bug long ago and it was just a matter of time before I'd be putting my FX-8350 onto the sandpaper.

Now these preliminary results are from the 8350 and the stock HSF, but I have a handful of other coolers that are also lapped and will be tested (and results will be reported in this thread) in time as well.

First up though are the results with the AMD stock HSF...I have to say this little HSF is just amazing at how good of a job it does for something that came with the CPU. Puts every other stock air HSF to shame!

All right then, how about some pretty pics :D

First up - the sandpaper portfolio. I like to use sheet-sized (8.5x11) sandpaper when possible as it gives me lots of room to move around without over-loading the paper with metal. The sandpaper I use can be bought at any auto-parts dealer (A1 Auto, Autozone, etc).

I use six different grits ranging from 220 to 3000. Anything above 800 is really just for putting a mirror finish on the surface and isn't expected to improve temperatures. (remember lapping is done to make the surface flat, not necessary to make it polished smooth)

SandpaperPortfolio.jpg


I lay out one of each sheet and tape them down to the surface of a plate glass desktop.

LappingStation.jpg


To reduce the loading effect on the sandpaper, I wet-sand with 220 and 400 grits and dry-sand with the remaining grits (800-3000).

Sanding is done as straight strokes, out and back, 10 times and then rotate 90 degrees. This is done a minimum of 8 times per grit (resulting in 2 full rotations of the IHS and HSF during sanding).

Here are the unsuspecting victims, lazing about not realizing their world is about to get a little flatter :D

FX8350pre-lap.jpg


(I had a couple 3770k's I wanted to lap at the same time, batch-job style ;))

A few passes on 220 grit is all it takes to get a feel for how warped the IHS or HSF surfaces were initially. If you see copper in the middle while the perimeter is still covered in nickel then you know the IHS was originally convex (like a dome), whereas if you see copper along the perimeter (edges) and nickel remaining in the center of the IHS then you know the IHS was originally concave (like a bowl).

FX8350flatnesscheck220grit.jpg


Both 3770k's were concave, the one on the far right was oddly concave in that the bowl was most distinctly square and not a rounded by any means. The FX-8350 was the exact opposite, having a high center because it was convex.

FX8350post-220grit.jpg


Finishing up with the 220 grit, we have removed all the nickel and are looking at what basically appears to be "brushed copper". It took forever and a day to get through all the nickel on the FX-8350.

It was painfully slow enough that I contemplated going to 60 grit just to speed up the process, in the end I opted to just stick it out with the 220 grit. And an hour later you have the image above. :|
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
Once you get past the initial 220 efforts, which are lengthy because that is the stage in which you are basically doing 95% of the surface flattening that is ultimately accomplished with lapping, the remaining grits move along rather quickly.

By quickly I mean it takes me roughly 3-4 minutes per CPU per grit once I get to 400 grit and beyond. That 220 grit though, ugh, can be an hour sometimes (no joke!).

FX8350post-400grit.jpg


^ at 400 grit you will notice the sanding scratches are now oriented along the diagonal instead of straight up and down. This is intentional. You want to alternate with every other grit between sanding along the diagonal (400, 1000, and 3000) versus sanding parallel to the IHS edges (220, 800, and 2000) if possible.

The reason you want to do this is because it makes it very noticeable by the naked eye whether or not you have completely removed the scratches left in place from sanding with the prior grit size. You don't want to move on to the 800 grit from the 400 grit until you know for certain you've removed all the 220 grit scratches with the 400 grit paper. Rotating 45° at every other grit makes reaching this milestone very easy to discern by eye.

After 800 grit:

FX8350post-800grit.jpg


For all intents and purposes we could stop right here and be done with the lapping. Further polishing with higher grits will still marginally improve temperatures, but only slightly.

Regardless the questionable technical value, I do like to go for that mirror-polish look simply for the aesthetic appeal, naturally.

FX8350post-1000grit.jpg


^ at 1000 grit we are starting to see the first indications of reflections which bear some resemblance to the images being reflected.

In these images I have positioned the retail CPU boxes nearby and captured their reflections from the surface of the IHS's respectively.

FX8350post-2000grit.jpg


At 2000 grit the reflections are really coming into focus now, but a fair amount of hazing is still observable.

FX8350post-3000grit.jpg


3000 grit makes things look nearly mirror-perfect. They aren't flawless though, by eye you can still see remnants of the very subtle scratches which 3000 grit makes in the soft copper surface. But the camera does a good job of unintentionally glossing over these surface flaws, behold the eye-candy :D (now we know what Cosmo did for those cover-shoots before photoshop existed, 3000 grit sandpaper of course D:, kidding)

One comment though, the reflections do look blurry but that is purely my poor camera skills, they are nowhere near as blurry when viewed by the eye. I couldn't get the camera to bring the reflections into focus without defocusing the CPU PCB itself (which made things look even weirder) so I chose the lesser of the two eye-soreness options and let the reflections go out of focus.
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
A side-comment on another big difference between lapping an AMD chip versus an Intel chip - CPU pins :eek:

Now most people, from what I gather, use a styrofoam block to cover the pins before lapping their AMD chips. I intentionally avoided this because I was quite keen to avoid risking an ESD (electrostatic discharge) event and insulating dielectrics such as styrofoam are notorious for accumulating surface charges.

Instead I just assumed/accepted in advance that I'd be bending a few pins around each corner of the CPU and that I'd simply bend them back when all was said and done. And the CPU did not disappoint, it obliged me in meeting my expectations ;)

FX8350BentPins.jpg


To straighten the pins I used a bit of wire insulator, the stuff you strip off a wire with wire strippers. (I stripped the wire and used the hollow tube to bend the pins as needed)

FX8350BentPinsTube.jpg


FX8350StraightenedPins.jpg


^ not perfect but it was good enough to sit fully into the AM3+ socket without issue. (and booted without issue)

Having finalized the lapping of the FX-8350, I then turned my attention to that wonder of a freely-bundled CPU cooler that is the stock HSF.

FX8350HSFpre-lapgrid.jpg


To perform a flatness check I drew the star-grid on the HSF surface with a blue sharpie pen.

FX8350HSFflatnesscheck220grit.jpg


^ a few swipes on the 220 grit sandpaper and we see the HSF is noticeably convex (domes in the center), the same as the CPU's IHS.

This observation immediately suggests we are going to see dividends in terms of lowered operating temperatures from lapping. Anytime you have two convex or two concave surfaces being mated there is going to be thermal conductivity issues because of the resulting gap.

Finishing up the 220 grit leaves us with the following:

FX8350HSFpost-220grit.jpg


Notice you can see a slight image reflection in the surface...this is because the 220grit sandpaper became so heavily loaded with copper filings from lapping the HSF that the grit number effectively became larger while I was sanding.

Judging by the reflectivity I'd say the 220grit paper must have been functioning like ~800 grit paper by the end of the lapping. Time to toss the sheet of 220 grit sandpaper and pull out a fresh one :awe:
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
Here's what it looked like after 400 grit (no notable reflectivity), but note the angle of sanding was rotated 45° from that used with the 220 grit.

FX8350HSFpost-400grit.jpg


800 grit yielded:

FX8350HSFpost-800grit.jpg


(slight reflectivity is back)

After 1000 grit:

FX8350HSFpost-1000grit.jpg


FX8350HSFpost-2000grit.jpg


^ 2000 grit, getting shiny!

FX8350HSFpost-3000grit.jpg


And looks like a mirror after 3000 grit.

And the moment captured right before mounting the stock HSF to the CPU with NT-H1 TIM:

FX8350post-lappre-mountNT-H1.jpg


Look at those reflections, I love it :)

And what about the resulting operating temperatures? Well those improved quite nicely too.

LappingthestockHSF.png


Comparing test 1a to 1b, with the voltage optimized for the unlapped setup (1.274V, measured by voltmeter) and held there for the test condition 1b, we see a mild temperature reduction of ~2°C. (all reported temperatures have been normalized to a 25°C ambient)

Test 1c is the situation where I have re-optimized the operating voltage by taking advantage of the reduced operating temperature. In these tests I required the voltage to be sufficient to ensure stable operation of Prime95 LargeFFT for ~1hr.

Comparing 1a to 1c, at 4GHz we see that lapping has enabled us to reduce the operating voltage by a smidgen as well as reducing the operating temperature by ~3°C. Not bad but not stellar.

However, the real dividends from lapping are to be found when overclocking as tests 2a through 3c show us.

Condition 2a was the max stable configuration I could achieve without going over 80°C operating temperatures before lapping. Lapping reduced the temperature by ~5°C right off the bat (2b) and further re-optimization of the voltage enabled an additional 4°C reduction for a net reduction of ~9°C at 4.3GHz (and a slight voltage reduction, always a nice thing).

But the temperature reduction gave me a new opportunity to up the OC clockspeeds one more notch without going over 80°C...that is what case 3c captures.

4.4GHz with 1.407V and a peak operating temperature of ~76°C. Sweet. :thumbsup:

Not bad for basically maximizing the performance of what came in the box for $190. I'm really quite impressed with this stock HSF.

Next up I'll be replacing the stock HSF and looking at performance with some much beefier coolers.
 

Haserath

Senior member
Sep 12, 2010
793
1
81
If Intel supplied a stock HSF that good, most overclockers wouldn't need an aftermarket heatsink.:hmm:

(Can nice thread just be implied from now on?:p)
 

LagunaX

Senior member
Jan 7, 2010
717
0
76
I see you have a 3770k there.

Be careful, you might get the delid/liquid pro bug.

Nice mirror finish BTW.
 

Lepton87

Platinum Member
Jul 28, 2009
2,544
9
81
Why lap Ivy if you can just ride bare? Much better that way, protection is always a hindrance ;)
 

FlanK3r

Senior member
Sep 15, 2009
312
37
91
very nice thread man! It could be interesting with better cooler, how it change temps in load. Example with some Noctua+1.475-1.5V at CPU. Before and after. Maybe one day Il try lapping too.
 

Abwx

Lifer
Apr 2, 2011
10,842
3,295
136
9°C is a lot given that the same gear is used but neverless , i found
that it s odd that thoses lapped components are so grossly manufactured
given all the industrial precision machinery that are available....
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
Nice work again IDC :D

Fun to see that its completely opposite of Intel. High convex vs low concave.

But as said, it is abit sad that both companies are so far from perfection in terms of "simple" metalworking.
 

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
You went all out :)
I don't think I've ever lapped anything past 600 grit myself.
150 400 600 is what I usually do.
 

inf64

Diamond Member
Mar 11, 2011
3,680
3,943
136
Very nice results and great topic IDC!
~9 degrees C lower operating temperature is massive gain when OCed, it payed off very well. I suppose even better cooler than the already good stock one would bring -few degrees too.
 

zebrax2

Senior member
Nov 18, 2007
971
61
91
Ohh shiny

Pretty decent gains specially on stock cooler. Any plans on deliding it after testing other coolers on it?

Nice work
 

daveybrat

Elite Member
Super Moderator
Jan 31, 2000
5,723
942
126
Very informative article. While i would never have the patience for doing what you did, it was nice to live through it via your eyes. ;)
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
Is your warranty gone?

Warranty was gone the moment I overclocked it, lapped or not.

How loud does the fan get?

I have a sound meter so I can answer this but it'll have to wait until late this evening when kids are asleep and not making noise ;)

I see you have a 3770k there.

Be careful, you might get the delid/liquid pro bug.

Nice mirror finish BTW.
Oh yeah, been there with my first 3770k :p

Delidded my i7-3770K, loaded temperatures drop by 20°C at 4.7GHz

Why lap Ivy if you can just ride bare? Much better that way, protection is always a hindrance ;)
True, lapping ivy is of little to no value if you don't plan to delid.

Lapped my i7-3770K and there was zero improvement in operating temperatures!?

But if you do delid with the intent to simply replace the CPU TIM and put the lid back on then lapping will help after-the-fact.

And ultimately bare-die mounting provides the best results but they really are not all that superior to putting the IHS back in place.

Bare-die testing: A delidded 3770k, an H100, and 9 different TIMs

Having done the bare-die thing with the first IB, I doubt I'll do it again. But I don't mind replacing the CPU TIM, biggest bang for the risk taken is to be had from just replacing the CPU TIM (reducing the gap really) and putting the IHS back on top of the CPU.

very nice thread man! It could be interesting with better cooler, how it change temps in load. Example with some Noctua+1.475-1.5V at CPU. Before and after. Maybe one day Il try lapping too.

I've got both the NH-D14 and the H100 to try out (lapped to 3000grit as well). I also bought one of those Spire TherMax Eclipse (TME) III specifically for this AMD chip based on FrostyTech's review results from the TME II model (they rate it as #1 for AMD rigs, better than the #7 NH-D14 and #10 H100).

Lapping really is easy to do once you get through it the first time. There is a bit of a learning curve, but I think the biggest barrier is just the psychological one - people fret over the prospects of ruining their chip. Understandable, I know I worried about it.

9°C is a lot given that the same gear is used but neverless , i found
that it s odd that thoses lapped components are so grossly manufactured
given all the industrial precision machinery that are available....

It is odd but I don't see it as any more of an oddity than what Intel has done in going with non-solder TIM for the CPU-to-IHS. Meaning regardless the opportunity for performance improvement that we enthusiasts are able to tease out of either our AMD or Intel rigs - be it from lapping or delidding - the stock solutions provided by either company are clearly ample and sufficient in getting the intended job done.

At stock, without OC'ing or over-volting, the AMD surface flatness (as non-ideal as it is) is still good enough to keep the processor plenty cool. The same could be said of Intel's choice in CPU TIM with IB.

The cost-benefits analysis must be considered. In a hobby like this I value my personal time as close to zero, but it did take me roughly 2 hours in labor to lap the CPU and HSF, plus supplies. I just don't see the cost-benefit for AMD to take on that effort themselves.

Ohh shiny

Pretty decent gains specially on stock cooler. Any plans on deliding it after testing other coolers on it?

Nice work

Can an FX-8350 be delidded? I asked before and the responses I got back ranged from "absolutely not, it is soldered onto the chip" to "no one has tried it, but FX-8350 OC'ing is voltage limited, not temperature limited, anyways so why worry?".

However I see Soulkeeper's thread on delidding his APUs and I'm left thinking if GloFo packages the 32nm APU's using CPU TIM instead of solder then it is not likely that they switch packaging styles just for the FX chips. So now I am thinking there is a chance the FX-8350 IHS is not soldered on after all...so yes, I am definitely tempted to try and see what happens.
 
Dec 30, 2004
12,554
2
76
I'd probably pitch in $25 to see if IDC can de-lid one. I'd love to see some benchmarks at 5.whatever ghz he can get it up to.
 

SickBeast

Lifer
Jul 21, 2000
14,377
19
81
OP you're a beast. This thread and the one about de-lidding Ivy Bridge are both incredible and are a huge addition to the forums. You're making me want to lap my next CPU.

I actually don't mind the 8350 as a CPU and might buy one in my next rig depending on what I do.
 

USER8000

Golden Member
Jun 23, 2012
1,542
780
136
An FX4300 or FX6300 use the same chip,so it would be a cheaper way to find out!! :p
 

Ferzerp

Diamond Member
Oct 12, 1999
6,438
107
106
be it from lapping or delidding - the stock solutions provided by either company are clearly ample and sufficient in getting the intended job done.

And providing a better solution would increase cost for *everyone* because things like creating that smooth of a finish are not trivial tasks for machinery.