Martin,

 

I can't disagree with anything that you posted, but by the time that a consumer purchases a CPU through normal channels (not an engineering sample on FleaBay, etc.), the CPU is already part of a narrow quality bin defined by the reputation of the the retail vendor. Basically, with OCing potential, you pay for what you get.

 

Besides, often with Intel CPUs, the model number indicates an initial broad bin, where chips from the same fab, chip class (like Haswell etc.) and manufacturing batch are tested and binned by performance into specific models.

 

All this leads to a lot of uniformity in the "silicon lottery".

 

Jay

Well clearly not as uniform as you imply.

 

For example Skylake overclocks are known to range from 4.5 to 4.8 GHz on average. There are even chips around outside of that. The range isn't huge, it never has been. I have no problem with that, just the notion that the silicone lottety doesn't exist at all.

 

Rob seems to think that such a variance doesn't exist.

 

What will probably occur is that as the Skylake platform matures a tad more consistency will be manifest.

And those "flaws in the manufacturing process" result in it being a lottery for you and I when we attempt to overclock our CPU's!!!

 

First lets not get too emotionally charged over this "debate" ... it's really not that big a deal in the scheme of life.

 

Let me propose this scenario to you, when you buy a NEW car do refer to it as a "lottery"?  Maybe one engine produces 505 HP and the other produces 502 HP ... the manufacturer will "allow" for a certain level of variance, and even the test equipment requires regular calibration.  Just like my race cars, each dyno will produce variances in results (in some cases as much as 30%) ... that's why I always ignored the dyno results and used the same dyno for relative comparisons of performance progress.  Since I've hand built several of my race engines I understand the variances at play ... setting the ring gap appropriate to the type of induction, valve shims, oil supply, etc. etc.  If try to push the engines output to 700 HP ... is that a "lottery"?  No it isn't ... there will be real failure points and those failures can be predicted over time ... I'll accept those failure points so long as I get a specific amount of time out of the engine (30 hrs was my goal BTW).

 

CPU manufacturing problem is similar yet their challenges are different:

 

What I see:

 

1.  Intel can't or will not refine their manufacturing process to improve quality/consistency.  It's no secret Intel have consistently had problems getting FABs online ... as such, there is a point where they just accept the problem in order to ship product.

 

2.  Intel do have solutions to the problem, but those solutions would increase product cost, since there is no real competition for Intel their is no reason to implement better manufacturing solutions ... it's "Good enough".

 

3.  Quantum tunneling is not the issue, if that were the issue, then increasing voltage to obtain stable higher frequency should make the problem of barrier penetration worse, not better.  Besides, that's a solvable problem, you simply increase the barrier size (and as such the CPU size ... something Intel does NOT want to do).

 

4.  Structural variation comes from incorrect manufacturing ... there is a reason why Intel's process seems to produce "better" quality CPU's closer to the center ... this is a solvable problem (either increase rejects or use a better, more expensive, and more consistent process).

 

So no "It's not a lottery" it's a manufacturing problem that results in lower frequency ... the consumer purchases the CPU at the stated frequency - your suggesting a "lottery" when you want to go behind Intel's stated frequency ... I don't consider that a lottery, it's called no warranty

 

I can assure you, if Intel were to run into competition, this "lottery" would disappear quickly.  It's simply another word for "inconsistent quality above stated frequency".  Do manufacturing variances occur, yes sure they do, in just about anything that is manufactured by hand or by robot ... but I've never associate my purchases of any product as being a "lottery".

 

But for the most part one can overcome these variances in manufacturing in CPUs with the addition of voltage and timing adjustments or disable cores/HT etc. etc. ... there are far more potential variances in motherboard manufacturing than in CPU yet nobody seems to suggest a "motherboard lottery" exists or a chipset lottery or a capacitor lottery ... etc. etc.?

 

Cheers, Rob.

First lets not get too emotionally charged over this "debate" ... it's really not that big a deal in the scheme of life.


I'm not and never will. We are not fixing global warming or feeding the poor, it's just a debate between enthusiasts. However, I reserve the right to experience incredulity when faced with an opinion that is contrary to fact.


Let me propose this scenario to you, when you buy a car do refer to it as a "lottery"? Maybe one engine produces 505 HP and the other produces 502 HP ... the manufacturer will "allow" for a certain level of variance, and even the test equipment requires regular calibration.



YES strictly speaking it is is a lottery in terms of the variance in car engine performance. But the difference between 505 HP and 502 HP is minuscule so no one gives a damn. That's not the same as an enthusiast experiencing a max 4.5 GHz overclock and an enthusiast achieving a 4.9 GHz overclock. Such a CPU variance IS a big deal to an enthusiast. When a CPU is manufactured, there is a far bigger variance in CPU capability across the wafer than the equivalent of your minuscule car engine HP variable.



1. Intel can't or will not refine their manufacturing process to improve quality/consistency. It's no secret Intel have consistently had problems getting FABs online ... as such, there is a point where they just accept the problem in order to ship product.



Nonsense! All CPU manufactures whether AMD Intel or anyone else experience the same inconsistency in CPU quality across the wafer. It's the nature of the process. You are using past, short term Intel issues in order to back up your argument. It's not about "accepting the problem to ship a product". That's not relevant at all to the natural variance that ALL CPU manufacturers face. This is not Star Trek, we have not achieved silicone chip perfection... yet!


2. Intel do have solutions to the problem, but those solutions would increase product cost, since there is no real competition for Intel their is no reason to implement better manufacturing solutions ... it's "Good enough".


Again nonsense! ALL CPU manufactures face the same issues. First you claim the silicone lottery "doesn't exist", and now you imply is does exist but there's a solution. Which is it? Whether there is "a solution" or not, the fact is the variance in CPU quality that results in the silicone lottery exists... you claimed it didn't!
 

 


3. Quantum tunneling is not the issue, if that were the issue, then increasing voltage to obtain stable higher frequency should make the problem of barrier penetration worse, not better. Besides, that's a solvable problem, you simply increase the barrier size (and as such the CPU size ... something Intel does NOT want to do).





Really... then maybe the BBC Horizon documentary I watched a few years ago was a figment of my imagination and the scientists in question imaginary. I didn't realise you were also an expert in quantum physics, as well as an expert overclocker who could counter the silicone lottery with your expert UEFI settings.
 


4. Structural variation comes from incorrect manufacturing ... there is a reason why Intel's process seems to produce "better" quality CPU's closer to the center ... this is a solvable problem (either increase rejects or use a better, more expensive, and more consistent process).



A statement that's contrary to everything we know about the CPU manufacturing process. Not the point any way because you stated that the "silicone lottery doesn't exist". But now seem to be saying it does exist but it's solvable. You seem to have reversed your opinion.


So no "It's not a lottery" it's a manufacturing problem that results in lower frequency ... the consumer purchases the CPU at the stated frequency - your suggesting a "lottery" when you want to go behind Intel's stated frequency ... I don't consider that a lottery, it's called no warranty



And that manufacturing problem RESULTS in a lottery fro the enthusiast overcloker. Of course it occurs when you go beyond Intel's stated frequency. Thus it's a lottery for the enthusiast overclocker. What on Earth did you think the debate was about. Something tells me you are now playing games.

Do manufacturing variances occur, yes sure they do, in just about anything that is manufactured by hand or by robot ... but I've never associate my purchases of any product as being a "lottery"



Well YOU might not associate your purchase as being a lottery, but you know damn well that in the context of this entire discussion, namely enthusiasts overcloking CPU's, that the variance in CPU manufacturing DOES result in it being a lottery.

I think Rob and Martin should open a new thread: "Silicon Lottery:  Does it Exist?"

 

lol

 

 

It's the nature of the process.

 

Yes, because the process is flawed ... there is nothing "natural" about it other than Intel haven't or will not expend the resources to come up with a better process ... be it technical limitation or a financial limitation or a little of both.  Intel have a process that produces an acceptable number of rejects at a defined frequency.

 

How do you know it's the CPU and not something else?  You don't, you don't have the equipment to deny or verify (swapping out one CPU and putting in another CPU that you can get higher frequency from isn't establishing a lottery) ... it does establish urban legend.  But ask yourself, if a CPU is "not as good" why would increasing voltage make it work?  It should have the opposite effect for some of the reasons you stated above.

 

If I slap a turbo charger on an engine not designed to be equipped with a turbo charge, sure I've just created my own lottery.  It's self created lottery ... if you increase the CPU frequency beyond manufacturer specifications you're the one accepting the risk and rolling the dice.

 

I agree with you there are manufacturing difference in EVERY product sold (not just CPUs) ... from cars to motherboard, to chipset, to capacitors ... so saying "it's a lottery" can be applied to anything.

 

Do you have any data at all that suggests a CPU running at 5.0Ghz with v1.3v is any better than the same model CPU running at 5.0Ghz @ v1.5v? (assuming other settings are equal)  I've been running CPUs at their max voltage (with adequate cooling) for many years now, not a single failure -- I just made them work with cooling, patients, and experimentation.

 

I've tried to get more data on Intel 5960X but so far it's not to any degree of accuracy, mostly web sites with "experts" tossing out numbers like 1.3, 1.4.1.5, 1.9 ... when pushed for verified Intel EE data, none is provided. 

 

But ask yourself this, if Intel believed in the silicon lottery don't you think Intel would work the same angle as memory manufacturers that sell "verified" high frequency RAM at 2X and 3X the cost?  Something like Intel verified 4.6Ghz CPU, for $1200 vs. Intel verified 5.0Ghz for $2000.  Intel do NOT "cash in" on this so called "lottery" -- they don't cash in for a reason (and that isn't from the goodness of their heart).

 

In summary, I'm not disagreeing with you that there are manufacturing variances, but those variance are certainly NOT significant enough to be associated with a "lottery".  

 

 

 

I think Rob and Martin should open a new thread: "Silicon Lottery:  Does it Exist?"

 

No, it'll be to close to a religious debate and violate ToS.  

 

Cheers, Rob.

I would just like to throw in I got a 6700K off the shelf at MicroCenter 2 months after release (That was the only mission, not wanting to buy some bundle pack) and it runs great. Easily got to 4.80Ghz, even got to 5.00 but volts were too high for my 24/7 liking. Extremely happy with my 6700K.

30 hours for a race car engine. I hope you had a sponsor to help defray some of the cost. Puts addon cost into perspective.

When a PC enthusiast buy a CPU, he has maybe a 1 in 500(?) chance of getting a "golden chip". Not very good odds, but not that bad compared to other lotteries.  What more needs to be said about it being a lottery?

Lotteria or not . You have also the problems with bad cores more cores more problems.

Iam sure you now witch cores ar worse in your 5960x?

As you say 5ghz 1.3 V or 1.5v probaly they perform same, but the heat be a bigger problem.

If you go all out at 2.0v the difference can be up to 400-500 mhz and a diff inside bad to best core 100-200mhz.

 

Here you se a app that is good for detrimite core capabilites.

You can test each core on the fly.

 

Agree with Martin , have problem with eng not my native Lang , cant catch up with Robs fansy words.

But my tip to Intel hire him

 

https://www.dropbox.com/s/du7tnddyt9r1m2w/X99-pcratio.png?dl=0

Yes, because the process is flawed ... there is nothing "natural" about it other than Intel haven't or will not expend the resources to come up with a better process ... be it technical limitation or a financial limitation or a little of both.  Intel have a process that produces an acceptable number of rejects at a defined frequency.

 

 

Again... nonsense! It's irrelevant if the process is flawed, because, and I'm repeating myself again, you said... "the silicone lottery doesn't exist". You didn't say it was small, you didn't say it was minor, your original statement that I and others objected to was that "it didn't exist at all"! And given that the silicone lottery, [experienced by overclockers] is born from the variation in the CPU manufacturing process [that you now admit exists] it's clear that your original statement is diametrically opposed to reality and that you now contradict yourself.

 

And once again, in your quote above, your logic baffles me. You say "Intel will not expend the resources to come up with a better solution" as if they actually could, but can't be arsed... but then contradict yourself by pointing out that it's probably a technical or financial limitation. If there are technical and financial limitations, then obviously they CAN'T "come up with a better process"! 

 

And I repeat myself again... why the emphasis on Intel? Intel, AMD, Samsung, VIA, and in the past, Transmeta, IBM, etc, etc, etc... have all experienced an inconsistent yield across the wafer [that leads to what the enthusiast overclocker refers to as the silicone lottery] 

 

How do you know it's the CPU and not something else?

 

 

Because we know, as you yourself now admit, that the yield across the wafer is inconsistant, leading to what overclockers refer to as the silicone lottery. And because, just like you proclaim to be, there are a multitude of very talented overclockers out there that have the experience to determine such a thing.

 

Not least of which is the motherboard manufacturers themselves. Lets not forget, that manufacturers like Asus, get the latest CPU's long before we do. In fact Asus and the like test hundreds of CPU's before we get them. and not forgetting of course, that the individuals that work for companies like Asus, EVGA, MSI etc, are highly qualified engineers, who have qualifications in the field that you don't. If they tell us that in their testing, a Skylake CPU,on average, with cooling variables the same, has an overclocking range between 4.5 and 4.8 GHz on average... I wouldn't be arrogant enough to dismiss it as nonsense!

 

Rather than accepting that there is a vast wealth of experience out there in the community, amassed over decades, both by amateurs and professionals, that tells us that the silicone lottery is real and significant... you arrogantly brush aside that huge wealth of experience and proclaim something diametrically opposed.

 

 

In summary, I'm not disagreeing with you that there are manufacturing variances, but those variance are certainly NOT significant enough to be associated with a "lottery".

 

Any variance is significant enough to be associated with a lottery, according to the dictionary definition. Thus, it's a lottery and the lottery exists. 

 

And of course those variances are significant. The vast wealth of experience, obtained by the overclocking community over many decades demonstrates that conclusively. As does the experience over many decades of the engineers that work for motherboard manufacturers. 

 

You may not wish to accept this, but the engineers that work for the motherboard manufacturers are actually qualified in this filed, and are more talented and capable than you!

 

When you make definitive statements like, Intel could do X if they wanted to, but won't because of Y and Z... you come across as very arrogant.

 

That arrogance is compounded when we consider that not once in your life have you set foot in a CPU manufacturing plant, and that not once have you been granted access to propriety information that relates to advanced CPU manufacture. 

 

This was your original statement...

 

But I'll maintain the lottery is NOT real.  What is real is the "guess work" that goes into OC

 

Quite clearly here, you are maintaining that there is no lottery at all, just a variance in the required UEFI settings. What this implies of course is that ANY CPU can be overclioked just as high as any other, with the aid of your special skill and knowledge of UEFI settings. There will be a multitude of experienced overclockers out there that will be raising an eyebrow at that statement!

 

And in your last post you said...

 

In summary, I'm not disagreeing with you that there are manufacturing variances, but those variance are certainly NOT significant

 

So now you DO admit that there are manufacturing variances [that are defined as a lottery by overclockers] but the effect is small.

 

You remind me of a global warming denier... first it's not happening at all, and now it is happening but the effect is small. 

 

 

That arrogance is compounded when we consider that not once in your life have you set foot in a CPU manufacturing plant

 

This is definitely getting way too emotionally charged, no need to call some one arrogant.  You've avoid key points and picked up "parts" of my other statements and then inferred/added content that was NOT said or suggested ... so you've successfully mangled the debate.

 

At this point in time, lets just say we disagree.  

 

Cheers, Rob.

Interesting (entertaining!) read for sure, I strongly disagree with Rob hardware wise on this topic, which is disappointing.  Martin, there is absolutely a "lottery" in the sense that of course some chips are better than others, have been for as long as I've been assembling PC's from the 80's.  All I can gather at this point is Rob is simply defending his initial no lottery line, but we all know some chips are simply better performers than others, regardless of other components.  This is nothing like hand building a 700HP engine, this is simply what variation of wafer quality will Newegg be sending me....

 

Disappointed how this played out with such a well respected member of our community, but life goes on.

 

But, we are all here for fun afterall!!!

Have the same feeling, i higly respect him .

To get deper in a easy way without fansy words , bad eng not my native Lang.

The " lotteria " you can have. Differences in a chip core wise from 0-300mhz.

Its not easy for a Non hardware nerd to find the right tool to detrimite that.

For years AMD have a confedenial tool for that that nowdays is out in public PS-Check.

My best AMD chip a FX8320 have a diff in core of 300mhz , i Mange that to 8.46ghz 1.95v ln2 on pair 4-5.

For intel you have X99- pcratio that is out for public .

As more cores you have more difference but not always if lucky you find straight chip, ex one of my good 6700k have no difference good for bench all cores.

I can give a good example Haswell 4670k start bin a couple the worse one 4.2ghz and the best one 5.0ghz on the same MoBo , cooling, bios settings , Thermal paste, and remounted tre Times' with the same difference.

 

If one of you get the 4.2ghz chip and your friend the 5.0ghz chip Its fair to say my friend was lucky on the lotteria.

 

Its the same with memory modules, you not need a lab as Rob say , only a CPU with a really strong memcontroller thats all.

Then if those ar worth the astronomic price tag , is another thing.

 

Hope that you understand what iam try to explain on a easy way with bad eng.

 

Regards/ Hasse

This is definitely getting way too emotionally charged, no need to call some one arrogant.  You've avoid key points and picked up "parts" of my other statements and then inferred/added content that was NOT said or suggested ... so you've successfully mangled the debate.

 

At this point in time, lets just say we disagree.  

 

Cheers, Rob.

 

 

 

Not emotionally charged for me really. Just experiencing incredulity.

 

I'm not sure what these key points are I've missed but would be glad to address them if you point them out. Same for anything "not said or suggested" would be glad to address it.

 

The point here is that it's not up to me, Westman, Allen or anyone else, to "prove our point". We are not challenging conventional wisdom, we are not disagreeing with decades of experience by the PC enthusiast community, or indeed, disagreeing with qualified engineers at Intel, Asus, MSI etc. 

 

When one puts forward an opinion that is controversial, opposed to conventional wisdom, then the onus is on that individual to provide the evidence.

 

As Marcello Truzzi said... "extraordinary claims require extraordinary evidence. As so far we have seen none.