Okay but the fact that cracking 4ghz is equivalent to winning the silicon lottery for a single ryzen chip is not infinity fabric's fault.

There's also the fact that just having the cores split across multiple chips does not decrease the odds of winning the silicon lottery vs Intel's monolithic chip design. You're still looking at 32 cores whether it's on a single chip or 4, and any or several of those cores could drag down the others in terms of the all core all-core overclock ceiling. That's one of the several reason these monster cpu's usually aren't great at overclocking, whether they're from team red or blue.

AMD actually has an easier binning process, they can put 4 well-binned modules together, Intel has to take what they can get, they're lucky just to get more half a dozen mega-chips from a single wafer that function at all.

Here's a cool tool where you can simulate what I mean by the above line:
http://caly-technologies.com/en/die-yield-calculator/
It's actually night and day, Intel are pretty well fucked unless they adopt MCM in some form.

This post was edited by DCSS on Feb 22 2018 02:06pm

i'm not talking about binning their process of picking their clocks, I'm talking about silicone lottery for enthusiasts to overclock
the odds go againt you once you add another ball, mr powerball, that is what another chip does

How does the same number of cores across multiple die's instead of one reduce your chances? I still don't understand. You're looking at the same number of cores in either case winning the silicon lottery is having every single one of those cores being able to go well above their base clocks.

Also read my edit of that post

Binning is absolutely a factor, one core could be lagging behind slightly behind the rest on a monolithic intel chip, they might even to increase overall voltage slightly to compensate so that they don't have to trash the whole thing but that core might only be able to hit the target clock and go no higher. AMD can simply swap in a better die if that happens, Intel's only other alternative is to fuse that core off entirely (probably 2 tbh because Intel hates odd numbers) and sell it as a slightly lower end chip.

The monolithic design has no advantages over mcm in manufacturing,
Nor does mcm come with any silicon lottery disadvantages.

This post was edited by DCSS on Feb 22 2018 02:15pm

is all silicone equal? you get batch 123 on one chip
batch 126 on another

its similar to mixing memory throw 1 fast stick with 1 slow stick, and they will run at slowest speed right? you think intel amd and etc are going to go really indepth on binning because we see great random chips on every level all the time
they will look for their bottom line specs and move on

This post was edited by yupitsmeh on Feb 22 2018 02:21pm

Well if one ram stick's memory is binned lower than the other and is consequently set to run at a lower speed...what else do you expect to happen? It's not really equivalent. You aren't more likely to win the silicon lottery if the # of cores is the same. Every additional core is making it less likely, and due to the fact that Intel has to take basically any chip that doesn't have a killer defect.

Not going in depth? So do you think they really do just throw a cpu in the garbage if it doesn't work or perform up to expectations instead of discovering the problem and fusing it off if it's a core and/or selling it as a lower product? "I guess it's just fucked!" AMD does the same thing with their modules, if it isn't well binned they aren't going to use it for a high end chip like threadripper or epyc where heat/power consumption are very important to keep down on those, if a module has some shit going on but still works they will just fuse off the shit and sell it as a ryzen 3 or 5, and swap in a good module. Intel cannot do this, that in combination with the exponential affect of defect density on yields means very few functional monolithic chips at all for any given wafer... Intel takes what they can get when they can get it.

And also, Intel's 'golden wafer' is going to birth very few golden chips's compared to AMD's golden wafer, if you think it's tied to wafer quality. By a very large degree, refer to the calculator. Why are your odds of getting one of those chips from Intel higher than it is from AMD, especially when they are that much rarer?

AMD of course pairs high quality modules to produce less overall heat and draw less power; and i'll give you a hint.. these are the properties that make chips good overclockers and more likely to be silicon lottery winners.

There's more that goes into the likeliness of winning the silicon lottery ofc like process maturity and the difference between Intel and GF.. all I know is GF is rolling out 7nm and Intel are still stuck at 14nm, take that as you will but to me it looks like Intel have less capable fabs which is actually a point for AMD here. The heresay is that Zen 2 is going to be fabbed by TSMC instead of GF, I don't know how credible it is, but that is probably when the arch itself will crack well over 4ghz and we will see how it really stacks up.

This post was edited by DCSS on Feb 22 2018 02:48pm

silicone discrepancies
and comparing node size is ridiculous they all have different shit
ITRS Logic Device

Ground Rules
Intel Samsung / GF TSMC
Process name 16/14nm 14nm 14nm 16nm/12nm
Transistor Fin Pitch (nm) 42 42 49 45
Transistor Fin Width 8 8 8 N/A
Transistor Fin Height 42 42 ~38 37
Transistor Gate Pitch (nm) 70 70 78 88
Interconnect Pitch (nm) 56 52 67 70

AMD can still get like 20-30 epyc's worth of ryzen modules out of a golden wafer vs Intel's like 5 huge chips assuming defect density is the same. If you still think your odds are better of getting one of them you've been reading too many intel mud-flinging marketing slides.



This post was edited by DCSS on Feb 22 2018 02:56pm

This is all non-sequiturs anyway, most people don't overclock their servers. Comparing consumer chips between intel and amd, Ryzen is one die just like intel... how could the odds of winning the silicon lottery be any different even by your logic?

simple math
add another completely different variable you get completely different odds most cases

you act like i'm fanboying or antifanboying
multi chip cpus are going to be the future but I believe its going to kill overclocking somewhat unless you are going to take the time to do a per core overclock

This post was edited by yupitsmeh on Feb 22 2018 03:07pm

Alright, let's pretend people are overclocking their epyc/28+ core xeon servers, let's also pretend AMD just randomly glues their modules together ala the hilarious Intel slide which is bereft of actual arguments against their competitor... You win.