50x Core i7 3770K OC test (5G air + LN2)

[Massman]

I see many people testing a shitload of CPUs, but so very little binning data is being shared in the forums. To me, it's a bit funny that someone who can test over 100 samples in a very methodological way would not share the information with the rest of the community ... but I guess that's how things go nowadays.

 

Anyway ...

 

So, alright, a second thread with results from the SF3D OC Gathering binning event. Why? Well, it's a different type of OC test and the sample size is a little higher. Also, the testing methodology was a bit less accurate than the IMC test, so ... well these are two different tests.

 

So, as you should already know by now, we had 50 retail 3770K CPUs at the event.

 

The batches were:

 

- L202B763 x2 (Malay)

- 3217B605 x18 (Costa Rica)

- L223B630 x10 (Malay)

- L220B232 x20 (Malay)

 

The following mainboards were used for testing:

 

- MSI Z77-GD65

- GIGABYTE Z77X-UD3H

- Maximus V Gene

- Maximus V Extreme (not actually used - one sample died, the other failed to boot)

- Maximus V Formula

- One AMD Llano setup :banana:

 

The binning part was split up into two parts:

 

- Lowest possible voltage for 5G 2C2T 32M

- LN2 testing

 

Disclaimer:

 

Please note that the even though the testing procedure was roughly the same for all event participants, given that the testing has been done on different platforms (board/cooler) and by different people, the results should be taken with a grain of salt. Due to time restrictions, we did not thoroughly test and re-test all the CPUs to verify that the initial reported lowest voltage is indeed the lowest possible or that the reported LN2 frequency is indeed the highest. Overall, we did see a correlation between the lowest voltage needed to run 32M at 5G on air and the maximum frequency on LN2, but it's not an exact science.

 

In fact, we also learned that unlike with Sandy Bridge, every Ivy Bridge needs to be tested under LN2 to know for sure that it's capable of a certain frequency. One of our most promising CPUs had spectacular cold issues and did not surpass 6.4G stable where we expected it to pass 6.7G easily. You can use the air binning method to filter out those that are probably really bad, so you can save LN2.

 

One last thing. I didn't see any correlation between my IMC test and CPU quality. So, a bad CPU can have a good IMC and a good CPU can have a bad IMC.

 

Anyway, I really hope that in the future it will be possible to repeat this kind of event and focus on more accurate testing methodology and reporting in order to extract more reliable information. We only had 2 days to bench and obviously most people were focused on getting good benchmark results rather than spend time to document the behavior of the shit CPUs . I think I was the only one brave enough to test one of the 10 really shit air CPUs Intel send us, only to find out it's also really shit on LN2!

 

The chart:

 

 

One more picture:

 

[[K]vlolv]

Wow! Such a big gap between Malay and Costarica. Btw where's Llano? LOL

[Hampti]

Wow the 6.8gig cpu is awesome. not only high cpu freq also an really good imc. nice work.

[[DRACO]]

Thanks massman and sf3d by this information,

 

But the Costa Rica up more? are best made​​?

 

better material? regards

[[K]vlolv]

;187769']Thanks massman and sf3d by this information' date='

 

But the Costa Rica up more? are best made​​?

 

better material? regards[/quote']

 

fact or rumor?

 

same question here

[knopflerbruce]

Finally!

 

I've been saying this since the SB charts were made, and people used to say "it's all about luck" when it clearly isn't:D C's were better than A's and B's for SB, and Costa Rica > Malay - on average. In your face, bitches:p

 

I just wish I had binned more in the SB days, found 56x chips left and right... could've made a fortune:p

[slaske6]

nice work

[I.M.O.G.]

Thanks for sharing!

 

In your face, bunnyes:p

 

pwned by swear filter, bunnyes! lol

[Massman]

Finally!

 

I've been saying this since the SB charts were made, and people used to say "it's all about luck" when it clearly isn't:D C's were better than A's and B's for SB, and Costa Rica > Malay - on average. In your face, bunnyes:p

 

I just wish I had binned more in the SB days, found 56x chips left and right... could've made a fortune:p

 

http://en.wikipedia.org/wiki/Law_of_large_numbers

 

It is all about luck. Even within good batches there are super crappy samples. Buy one sample and you'll need luck to get a great one!

[TaPaKaH]

fyi, law of large numbers does not work for CPU binning as you're looking for the best sample rather than the average.

Either way, since we (=mortals) are working with a small sample size, no decent statistical analysis can ever be possible.

[_mat_]

I've binned 4 retail cpus so far, all Malay.

 

Week 08 => 6,5 GHz 1M

Week 20 => 6,6 GHz 1M

Week 20 => 6,8 GHz 1M

Week 20 => crap but good memory controller (3000 MHz+)

 

Not too bad at all.

[sin0822]

batch consistency? Yea same as sandy bridge i would say. Some CPUs in same batch can't even do 5G, not all of them get high frequency?

 

my theory is that out of every batch you can get a few good CPUs. The 6.8G is very good.

 

Why do all malaysia suck?

 

Thanks for all the hard work massman, we appreciate you sharing your results.

Edited July 25, 2012 by sin0822

[ZoLKoRn]

i don't care for batch anymore now but now very interesting for Costa Rica vs Malaysia ! seem Ivy from Malaysia so bad !

and some user on my forums (not ksin LoL) his have good one 3770K from Costa Rica for IMC, he can boot by 2933 strap

and can LinX stress for 2950MHz(1:14) all air.

 

btw. but i just got one 3770K about 6hrs ago and it Made in Malaysia L221B386 LOL hope it will good one hehehe

 

Thanks for sharing Massman

 

more btw. so have some good chip for air cooling for sell ? like 5GHz and around 1.3-1.4V hehehe

[knopflerbruce]

fyi, law of large numbers does not work for CPU binning as you're looking for the best sample rather than the average.

Either way, since we (=mortals) are working with a small sample size, no decent statistical analysis can ever be possible.

 

The sample size we had for SB was large enough, we had more than 100 B batches at least. The difference between CR and Malay was roughly 100 MHz when I did my calculations, and from what I recall, C batches were roughly 100MHz better than A, too and A again about 70-80 better than B. That's a fairly huge variation, and with as many as 100 results I'm sure that the chances of this being just a coincidence is minimal. I could refresh my knowledge about statistics and prove it, but hell - if you guys don't believe it it just makes it easier for me to find good chips if i decide to start binning these things:p

[knopflerbruce]

http://en.wikipedia.org/wiki/Law_of_large_numbers

 

It is all about luck. Even within good batches there are super crappy samples. Buy one sample and you'll need luck to get a great one!

 

Well, if you increase your chances by picking the right batches. 10% is better than 5%.

[Massman]

fyi, law of large numbers does not work for CPU binning as you're looking for the best sample rather than the average.

Either way, since we (=mortals) are working with a small sample size, no decent statistical analysis can ever be possible.

 

Essentially the theorem says that the larger the CPU binning sample size, the closer you get to the average overclocking capability of that one specific batch. If there's such a thing as better/worse batches, that average would be different. I think we can all assume there are better and worse batches.

 

As you point out, though, most of us are working with small sample size. For that case, the theorem basically points out that there's no guarantee for the sample to come close to the expected (="average") overclocking capability. Essentially ... luck.

 

Well, if you increase your chances by picking the right batches. 10% is better than 5%.

 

The fact that you use the word 'chances' indicates it's still about luck .

[K404]

Never mind 225-229 will be here soon!

[Splave]

IMO try to get 5-10 of the same exact batch number

[Viss]

15 pcs 3218C017, 4 of them need >1.35v to boot 50x100 4/8 @ air.

[zzolio]

the one 3218C017 i have will 1,295V 4/4 and 1,33V 4/8

 

and 6150mhz on cascade 32M 2/2

[dumo]

IMO try to get 5-10 of the same exact batch number

The best and easiest way to bin:)

[ZoLKoRn]

My new one for L221B386 Malaysia is unbox about hours ago

 

5GHz need minimum for boot 1.45V and Spi32 need 1.47V but seem it strong for IMC it easy for 2800C11 1.65Vdram, 1.25Vimc

 

if eight thread hyperPi need for 1.52V !

 

btw. testing on water cooling and room temp around 33C