Odd variation across same model CPU (silicon lottery?)

Sep 4, 2018
2
0
10
I've had the opportunity to test a bundle of CPUs of the same model (from Ivy Bridge E series), and have found a very odd variation in stock thermal performance when running Prime95: the average CPU is running about 55C maximum (most of them are around this), while one hot one was in the low 60s, and one extremely cool one was running mid 40s. I realise there is a silicon lottery, but this seems like a crazy variation: the hottest chip has almost double the deltaT of the coolest chip.

I thought the cool chip might be running crippled somehow, but it gets more or less exactly the score it should get on Cinebench.

Any ideas what's going on here?
 
Solution


I seem to recall that Intel says the on die temp sensors are +/- 5 C ... that could contribute. BTW, did you hold that chip back ... bet it overclocks like a champ.
The technical term is down binning. As I think you know, each CPU comes that comes off the line is test for it's potential speed. Each CPU has some small errors in it's lithography. It is those errors that limit the speed the of the processor (technically, spots of higher resistance cause heat. The more you try to pump through that spot the higher the heat. Too much heat and you get failure). As a CPU matures the number of errors go down and percentage of CPUs coming off the line that qualify for the highest speed increases. The issue is that a lot of people like to get the middle of the road CPU (hey, they are cheaper). So, even though the CPU can perform much better than what is stamped on the cover, the company sells it as a slower processor.

Now, with the Ivy Bridge E, it gets even more complicated because the CPU comes in 2 flavors ... 4 core and 6 core. On the wafer these two flavors are exactly the same, but sometimes you have errors concentrated in one core, so you disable that core (and one more) and call it a 4820. All cores are good, then you call it a 4930 or 4960. Sales of 4820s are brisk and lots of CPUs are binning higher, then you take a beautiful 4960, disable 2 cores and stamp 4820 on it.
 
Sep 4, 2018
2
0
10
I don't think it's cooler installation as that would require me to systematically do bad installs on all the other CPUs but the really cool one. Maybe it's a downbinned one, but even that explanation grates a bit because the difference in temps is so large.
 


I seem to recall that Intel says the on die temp sensors are +/- 5 C ... that could contribute. BTW, did you hold that chip back ... bet it overclocks like a champ.
 
Solution