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

[Phynaz]

Awesome as usual.

You need to talk to Anand. These articles should be on the main site, not buried here in the forum.

[Ferzerp]

Oh yeah, and for anyone looking for fine grit sandpaper... Try the auto parts store and not hardware stores. That one was a learning experience the last time I did this.

[akugami]

Last time I did this was many many years ago. I stopped at 1000 grit since anything after that is really just for cosmetics. However, the technique I used was a figure 8 motion. Roughly every 10 figure 8 motions, I would rotate the CPU 90 degrees. The reasoning behind the figure 8 is the same as rotating the CPU during the lapping process. A human hand applies uneven pressure which will naturally result in an uneven removal of materiall during the lapping process. Rotating the CPU will alleviate that but a figure 8 will also help in making sure material is removed evenly from all sides of the metal surface.

[lau808]

Originally Posted by Ferzerp

Oh yeah, and for anyone looking for fine grit sandpaper... Try the auto parts store and not hardware stores. That one was a learning experience the last time I did this.

i used to work at napa for 10 years and we only carried up to 2000 grit.

great job idc, another great thread. cant wait for the tests with the noctua and h100!!!

[Ferzerp]

Originally Posted by lau808

i used to work at napa for 10 years and we only carried up to 2000 grit.

You're lucky to find 1000 (usually 800 is all you can find) at Home Depot or the like though. I get mine at O'Reilly.

[Yuriman]

Walmart has sandpaper over 2000 - at least the one here in Daytona does.

[Triskain]

All AMD CPUs in the 125W class (and above) are soldered on, or at least that was the case with Phenom CPUs. So I would caution against delidding it, unless you want to brick your 200$ investment (again). By lapping it you have already realized most of the gains you could expect from delidding it. So there is no point to it really.

[Idontcare]

Originally Posted by Ferzerp

You're lucky to find 1000 (usually 800 is all you can find) at Home Depot or the like though. I get mine at O'Reilly.

Yeah home depot only carries grits that you would need for wood work, painting and the like.

To get the higher grits you have to go to an auto-parts store where they carry higher grits for truly polishing metal (chrome and the like).

I get my 800 and lower grits at home depot because they sell them in volume, and for cheap. Then the higher grits, more expensive per sheet, at Auto-Zone.

Originally Posted by Triskain

All AMD CPUs in the 125W class (and above) are soldered on, or at least that was the case with Phenom CPUs. So I would caution against delidding it, unless you want to brick your 200$ investment (again). By lapping it you have already realized most of the gains you could expect from delidding it. So there is no point to it really.

This is what I had heard as well. The only motivation I'd have for delidding is if there was non-solder TIM under the IHS, which I'd want to replace with some Liquid Ultra.

But if what you say is true, then there is absolutely nothing to be gained from delidding even if I could delid the soldered IHS without damaging the CPU. Soldered on IHS is the superior solution.

[Ferzerp]

Originally Posted by Yuriman

Walmart has sandpaper over 2000 - at least the one here in Daytona does.

I've looked at 2 and didn't find any high grit locally.

[AnonymouseUser]

Originally Posted by Ferzerp

I've looked at 2 and didn't find any high grit locally.

Are you looking in their Automotive section?

[GreenChile]

I would be really interested to see if there are any temp differences between a 220 grit lap and a 3000 grit lap. Has anyone done a comparison to see if lapping to a shinny finish is better?

I'm thinking the flatness of the surfaces may account for all the thermal transfer improvements and the polished surface does nothing but look nice. I would love to see some data on that though.

Excellent thread as always IDC!

[videogames101]

Originally Posted by Idontcare

I want to try my vapoLS on it first. I'm actually giddy to get that test underway

oh lawd

i love those exotic cooling solutions

[jvroig]

Originally Posted by GreenChile

I would be really interested to see if there are any temp differences between a 220 grit lap and a 3000 grit lap. Has anyone done a comparison to see if lapping to a shinny finish is better?

I'm thinking the flatness of the surfaces may account for all the thermal transfer improvements and the polished surface does nothing but look nice. I would love to see some data on that though.

That's pretty much the OP's own remarks (and I can think of nothing to dispute it as I personally also hold the same thoughts on the matter):

Originally Posted by Idontcare

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)

[KompuKare]

Another excellent article IDC! This really should be on the AT frontpage IMO.

Originally Posted by Idontcare

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



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)

So the lilac rows are where you've basically used your lower temps after lapping to hit a better vcc while remaining stable? Or least that's what I read. Pity it's just yourself doing this because I could think of a few more tests you could have run

For one thing I'd love to see would be how this affects underclocking: would your non-lapped vs lapped be able to run stable at say 2.0GHz with a lower voltage, and how much?

The second thing is which I'm sure you're going to get around too: overclocking past 4.4GHz. And unfortunately (unless you ran some figures and left them out) now that you've lapped it I can't see how you can get the figures for - say - 4.8GHz, 5.0GHz etc. before lapping. In other words now that the original is lapped I assume we will only be able to guess the results of 5.0GHz pre-lapped vs lapped. I guess it will may (should) be more than 10°C.

[Idontcare]

Originally Posted by GreenChile

I would be really interested to see if there are any temp differences between a 220 grit lap and a 3000 grit lap. Has anyone done a comparison to see if lapping to a shinny finish is better?

I'm thinking the flatness of the surfaces may account for all the thermal transfer improvements and the polished surface does nothing but look nice. I would love to see some data on that though.

Excellent thread as always IDC!

Years ago, back in 2007, I purchased five Q6600 CPUs and built five identical quad-core rigs. I lapped all of them, and the Tuniq120 towers I bought to go with them, and comfortably OC'ed all of them to 3.3GHz.

In the process of lapping those chips (I was interested in the statistics of OC'ing at the time) I systematically checked the temperature reduction gained after every grit and found the bulk of the benefit came from the grits up to 800, but nothing was really gained after that.

I wasn't into documenting or publishing my results at the time so there are no threads or links for me to point you to, and I doubt I even kept any of my notes from so long ago.

You could call it quits after just 220 grit and still have around 80% of the benefits. It really is just a matter of getting two warped/uneven surfaces to be flat, the surface roughness itself (the scratches) is not as big of a deal as you might otherwise think.

Originally Posted by KompuKare

Another excellent article IDC! This really should be on the AT frontpage IMO.

So the lilac rows are where you've basically used your lower temps after lapping to hit a better vcc while remaining stable? Or least that's what I read. Pity it's just yourself doing this because I could think of a few more tests you could have run

For one thing I'd love to see would be how this affects underclocking: would your non-lapped vs lapped be able to run stable at say 2.0GHz with a lower voltage, and how much?

The second thing is which I'm sure you're going to get around too: overclocking past 4.4GHz. And unfortunately (unless you ran some figures and left them out) now that you've lapped it I can't see how you can get the figures for - say - 4.8GHz, 5.0GHz etc. before lapping. In other words now that the original is lapped I assume we will only be able to guess the results of 5.0GHz pre-lapped vs lapped. I guess it will may (should) be more than 10°C.

Thanks for the kind words

It is true, I did not collect >4.3GHz data for the CPU prior to lapping...but the reason I didn't is because my NH-D14 and H100 are already lapped. It makes little sense to lap just one surface versus lapping both. That is why I lapped both the IHS and the stock HSF before testing to compare to the non-lapped stock case.

Everything I own that could get the 8350 to 5GHz was already lapped, the only thing that wasn't lapped was the CPU itself and the stock HSF. Now that everything is apples-to-apples it is time to get that chip to 5GHz. (but I'm intentionally waiting until tomorrow before I remove the stock HSF because I need to get some sound measurements with the stock HSF per sandorski's request)

[Abwx]

Originally Posted by Phynaz

These articles should be on the main site, not buried here in the forum.

Fully agree with the proposition but it might be that it will
be deemed as "too technical" for a general public....

[Idontcare]

Originally Posted by sandorski

How loud does the fan get?

OK, everybody went to bed so I was able to run the sound pressure tests without having a compromised ambient reading.



The green data are for the AMD stock HSF that comes with the FX-8350.

The blue data are for an NH-D14 that has had its stock fans removed and replaced with three quieter Noiseblocker fans (@$25 each).

So the answer to your question is: it can get as loud as ~48 dB at 18inches. Noticeably louder than an NH-D14 but not crazy loud either.

Now hopefully I can find my AM3 mounting bracket for the NH-D14, next tests will need it.

[Gikaseixas]

Like everyone's been saying, amazing thread IDC
Anand should let you use his lab and turn up the heat

[Idontcare]

Well I put on my NH-D14 and was once again reminded that it is a beast

At stock, 4GHz (turbo-core disabled) the NH-D14 is turning in peak temperatures that are a full 20°C cooler than those turned in by the lapped stock HSF!

Even when OC'ed at 4.7GHz @ 1.487V the temperatures are cooler (52.3°C) than the stock HSF temps at 4GHz (56.6°C).

The question I have is how high can I really push the volts on this 8350 before I start getting into stupid territory? The chip is still very cool, thermally I have at least another 10, if not 20, degrees to play with but I'm already pushing close to 1.5V.

[inf64]

I'd be looking at reported temps and would not go above 65C when you push on the Vcore and clock.

[Idontcare]

Originally Posted by inf64

I'd be looking at reported temps and would not go above 65C when you push on the Vcore and clock.

How high in Vcore do people go with piledriver before they call it quits?

Is 1.5V already too high?

[Rvenger]

I have seen 1.6 and even 1.7v. I am not saying that it's safe though. But likely your CPU will survive if not run 24/7 at 1.6v.

[jvroig]

Originally Posted by Idontcare

Is 1.5V already too high?

We'll never find a real answer to this because AMD just plain old keeps mum about it.

We've been down this road before, but to wit, the only close-to-official AMD recommendation we have is from the Deneb/Thuban (45nm chips) era where they advised a max of 1.5v on high-end air (courtesy of their Dragon Platfrom Performance Tuning Guide). 1.65V is already on "Extreme" cooling, which they quantified as liquid LN2, not just water cooling.

I have no reason to believe >1.5v is in anyway safe with the shrink to 32nm. But I have no data to say it isn't.

You decide, that's about the only data point we have that actually came from AMD. Anything else is just people guessing wildly and doing really crazy things like what Rvenger mentioned above (1.6-1.7v). Since Bulldozer, AMD hasn't published voltage ranges anymore, so we can't even try to "extrapolate" for 8350 using 8150.

[inf64]

1.55V should be max safe Vcore you would want to run your chip on. Make sure temps don't go over 65C and it should be OK.

[jvroig]

Originally Posted by inf64

1.55V should be max safe Vcore you would want to run your chip on

? [citation required]

Where is that information coming from?

How do we even start to verify the veracity of that information?

Are there published docs regarding the characteristics of 32nm AMD products?