If the chances of getting a particular 8700k look like this:
I wasn’t really expecting anything less from Intel. But, I would really like to see silicone lottery start binning AMD.
Someone in the comments crunched the numbers further:
I am sure that even person with no understanding of statistics would reach the same conclusion.
Edit: Sorry bit of a rant.
I have bachelor’s degree in finance and I love math (especially calculus) but I fucking hate statistics. Mainly because nothing it true. In stats you reach no conclusions and its only use is a tool to convince people who don’t understand stats of ridiculous things to sell em more shit. Like standing desks for example. Its not that sitting down is detrimental to your health it just that people who sit a lot tend to be physically inactive therefore less healthy.
If Intel was doing this t-test it would just use significance test factor of 10 and say its all good. And only people like me understand that the test is just shit. Even though the test would be technically correct.
TLDR: The point is critical thinking is infinitely better than stats.
There is no point to bin AMD… they would be just buying chips and tossing out a fair many of them. What makes Silicon Lottery so good for Intel is the delidding is done by them, you just have to slap it in your system and put a cooler on it.
I agree, buying AMD just to bin em doesn’t seem like a profitable business model. I still would like to see it just for the clock variance. I don’t even care if AMD is also sending binned CPUs to reviewers. Its the numbers that turn me on…
For the first gen Ryzen chips the variance was 3.8GHz to 4.2GHz, mostly depending on how long you want it to live. 
Most tech outlets have pushed their chips to 4GHz and called it there, because it is a number that makes a good headline. Some of those overclocks were done with questionable voltages though. So I would not expect the AMD samples to be binned.
But of course I have no real data so take this as my opinion more than anything else.
I had my wife look at it and after about half an hour of talking about it, the spreadsheet is not very good. Right track, but kinda crappy.
She’s on her way to a PhD in Stat FWIW, but then again, you’ve mentioned you think that’s worthless so…
I hear that a lot from anyone who asks what my wife going to school for 
Edit: actually now she’s curious and stuck on it so might come up with better results. I’m no statistician but I’ll post if she is satisfied with them.
Not really much you can do. One thing though the clocks are inclusive so the sample just has three data points:
Clock Number %
5.1GHz (20) 100%
5.2GHz (17) 85%
5.3GHz (9) 45%
You do have to make assumptions about the silicone lottery, all your given is a distribution and a sample of 1000+. So if 99% do 4.9GHz does that mean 100% do 4.8GHz? and so your data looks like:
4.8GHz 1000
4.9GHz 990
5.0GHz 880
5.1GHz 540
…
gl
There is little to be surprised about with companies putting their best foot forward… supplied expensive tech.
May not represtent the average, tho.
Here’s my wife’s analysis of the ConLake data, but keep in mind she hasn’t watched the video. She was more interested in deconstructing the spreadsheet than the tech stuff. I think I can call her pretty unbiased, she doesn’t care about Intel/AMD whatsoever.
Not sure if there’s a way to embed HTML on Discourse so I printed a PDF, which I have to link to via Google Drive since I can’t upload it
https://drive.google.com/file/d/1a0kSLaN2brh2B8W2SB2mooRaaQTaVQAv/view?usp=sharing
Anecdotal but I don’t have any press samples from Intel. No delidding.
The voltage is more of a factor for non delidded processors. Delidding mainly affects how much higher you can push voltage and still keep temperatures under control. For a sample size of 3 I’ve been able to hit 5ghz on all cores with 280mm or better aio cooling. Negative avx of one or two. Do we know the test type and durations that silicon lottery uses to determine? If I dont do exotic workloads or prime 95 I can manage somewhat higher clocks. Negative avx is not a perfect countermeasure to an overloaded CPU…
It’s nigh impossible to keep CPU temps under control past 1.35/7 volts but this varies from board to board because load line calibration.
I’d say a voltage bump plus managing temps but without also bumping the clocks improves stability
I screwed up and completely blanked on the t-test spreadsheet missing the voltage numbers.
She says that the results are still subject to the same limitations of the test though, and we do not know what Silicon Lottery’s process involves, much less how random the reviewers’ methodology is as well.
