Lemme make sure i understand this

So I cannot find the article anymore, but I came across a statement that blew me away, and it makes sense, i just wanna make sure it is accurate, the specific proc in question is the Phenom II x4

The post/article implied this:
So AMD make a boat load of these, alllll on the same line, then has some form of benchmark that they all pass through for quality controls, at this point the proc is deemed stable at certain Ghz, anything under 3.4Ghz would be labeled a 955 for instance, above 3.4 it would be a 965, and the best of the best would get the 970 label, but they are in essence the SAME PART off of the SAME production line, its just how well they HAPPEN to come off the line that determines what rating they get locked at?!?!?

I guess I just didn't realize that the 955/966/970 or the 1055/1075/1090 were in effect IDENTICAL and it was effectively manufacturing tolerances that determined their performance rating? Did I mis read or is this an accurate description of how things work?

More or less yes, thats how it works.

There are X amount of CPU dies per wafer. Those towards the center tend to be the better of the bunch IIRC. But yeah, if it reaches x.xGhz with x.xxV keeping under their TDP thresholds and stable then its labled chip A and so on.

Yup that's how it works. Intel probably does the same thing I am guessing. And before the ones that had a bad 4th core get relabeled as a 720 X3. This is why every cpu is different at how high it can overclock and overclockers care about the batch number.

Right ^^. This process is not exclusive to AMD AFAIK.

bigben6 said:

So I cannot find the article anymore, but I came across a statement that blew me away, and it makes sense, i just wanna make sure it is accurate, the specific proc in question is the Phenom II x4

The post/article implied this:
So AMD make a boat load of these, alllll on the same line, then has some form of benchmark that they all pass through for quality controls, at this point the proc is deemed stable at certain Ghz, anything under 3.4Ghz would be labeled a 955 for instance, above 3.4 it would be a 965, and the best of the best would get the 970 label, but they are in essence the SAME PART off of the SAME production line, its just how well they HAPPEN to come off the line that determines what rating they get locked at?!?!?

I guess I just didn't realize that the 955/966/970 or the 1055/1075/1090 were in effect IDENTICAL and it was effectively manufacturing tolerances that determined their performance rating? Did I mis read or is this an accurate description of how things work?

Click to expand...

More or less, right on. That is the basics of overclocking. How it all got started if you will. See, when large companies...retail stores, Dell, HP etc., they generally consume more low end CPUs than high end CPUs. Simply meaning there is more demand for entry level speeds as opposed to the fastest of the fastest. For instance, if Dell calls AMD and puts in an order for a million 2.8ghz CPUs but AMD only has 500,000 sitting around...guess what? AMD will sometimes relabel 500,000 CPUs that could run 3.6ghz down to 2.8ghz to meet the demand. Then maybe New Egg calls up the next day and needs 100,000 2.8ghz CPUs...the re-branding of faster CPUs continues. You never know what speed your CPU could actually run at until you test it yourself. Thats why I laugh when people say they don't want to overclock something...you might not be overclocking it at all, lol.

It cost Intel/AMD no more or no less to make a given CPU of the same architecture regardless of speed. It cost them the same amount if it goes 2.8ghz or 3.8ghz. So they will sell whatever they have to meet the demand. This example is drastically simplified, but it gets the point across.

It should also be noted the the manufacturing process gets better and better the more chips are made. So (in theory) when AMD was selling the first x3 core CPUs those might have been broken x4 CPUs. Meaning one core wasn't up to snuff. But now that the process has matured I suspect most x3 CPUs that are sold are perfectly fine x4 CPUs...there is just demand for x3 CPUs still. For instance my 555 BE unlocks from x2 to x3 to x4...essentially making it a 955 BE with two cores turned off to keep up with the demand of entry level 555 BE CPUs.

This also helps explain why, as a percentage, entry level processors overclock higher then the high end ones...again as a percentage.

Quite interesting, I never really thought of it like that, that makes me happier knowing I bought a lower clocked quad in the series and saved some $$ lol.

rollincoaster said:

Quite interesting, I never really thought of it like that, that makes me happier knowing I bought a lower clocked quad in the series and saved some $$ lol.

Click to expand...

Exactly. Plus, you have to think about this. All CPUs are clocked slower (and running on less voltage) than they will actually run. When Intel makes a 930 they have no idea if it's going to end up being in 'Grandma Eskimo's' email machine in north Alaska or an Panasonic ultra durable Laptop commissioned for use by the Army in 130 degree Afghanistan. It just has to work no matter what. All CPUs have a little headroom.

Then I am glad i got the 955BE and not the 965 or 970.... it is intriguing, how does one find the batch number, and how does one know if it is a good batch or not?

^ Yea usually the chips of the same architecture overclock about the same. For the batches you just gotta find some forums that discuss it. Try xtremesystems.

Let us not forget that though all pass a basic QC all may not be equal. Binning still happens.

bigben6 said:

Then I am glad i got the 955BE and not the 965 or 970.... it is intriguing, how does one find the batch number, and how does one know if it is a good batch or not?

Click to expand...

Forums (mostly XS) start databases and keep track, but normally only on really popular CPUs.

There is nothing special (per se) about a given batch. Just some of the manufacturer's fabrication facilities have some better months than others. I doubt Intel/AMD sits around and says..."ok guys, this month we are really gonna make some good ones." It just happens that some "batches" are better than others. There are a ton of variable involved in this equation.

Not to be confused with new steppings...for instance Intels 920s going from C0 to D0...that is not a "good luck batch" thing, stepping is an acknoledged improvement in the manufacturing process.

A good example is in the days of the PentiumII, where the demand for 300mhz cpu's were so high that Intel remarked some 450's as 300's and clocked them at 66mhzfsb instead of 100mhz with a 4.5 multi, and the only thing most people had to do is raise the fsb to 100...they didn't even have to raise the core volts as the cpu's were the same. I believe the remarked PII was SL2W8 or SL2WM, I can't remember which. I have one of the two (I think mine is an SL2W8)

The very first pentiums were the best example i think.
There were two speeds, 66mhz and 60mhz.
Intel originally intended to make only 66mhz parts, but a good chunk of them failed to run at 66mhz but would run at 60mhz just fine. Hence, intel created the 60mhz product line from scratch to fit their defective products.

Before that there were the 486SX cpus that had defective FPUs and had said FPUs cut out.

Archer0915 said:

Let us not forget that though all pass a basic QC all may not be equal. Binning still happens.

Click to expand...

Binning? Elaborate could you please?

Oh wow... I had no idea this how they they made the different processors

Makes sense tho... Why have 3 different machines to make 955, 965 & 970 chips or one machine to do all 3

Im glad i joined this site cause i have a huge thirst for knowledge

bigben6 said:

Binning? Elaborate could you please?

Click to expand...

I'll +1 to that cuz i dont know either

bigben6 said:

Binning? Elaborate could you please?

Click to expand...

Finding the cream of the crop chips and designating them as such. For instance, most people would probably agree that Extreme Edition/Black Edition CPUs are normally above standard CPUs...people buying them are expecting such. (Many times server chips, Opeterons/Zeons are the result of high binning.) Binning is simply a term for finding the CPUs that perform above the average and treating/labeling them as much.

Don't get that misconstrued, not all special chips are binned...but all binned chips are special, lol. One cannot make a blanket statement about EE or BE CPUs being binned well. A lot of factors come into play when we start down this road. You will notice that Intel has only a couple of available Extreme Edition CPUs at this time...while AMD has quite few. This is marketing and the fight for market share, not that AMD is making consistently better CPUs. Remember, a CPU of whatever architecture cost what it cost to make, regardless of speed, regardless of having an unlocked multiplier.


I should also point out...many overclockers (the top dogs with money or connections) bin CPUs/Video cards as well. Every CPU is different, you have heard that a million times, but it's true. Some overclockers will acquire (through whatever means) many of the same exact processor. Then bench them to find out which ones go faster. For instance, if you benched ten 955 BE's there would be one or two of them that would be faster than the others, this is also binning. You might have read on the forums that someone is selling or using a "golden" CPU...this simply means it's fast as compared to other CPUs of the same moniker.

This is kinda cheesy and old school...but a neat watch for some of you guys new to this kinda stuff.

xoke said:

Finding the cream of the crop chips and designating them as such. For instance, most people would probably agree that Extreme Edition/Black Edition CPUs are normally above standard CPUs...people buying them are expecting such. (Many times server chips, Opeterons/Zeons are the result of high binning.) Binning is simply a term for finding the CPUs that perform above the average and treating/labeling them as much.

Don't get that misconstrued, not all special chips are binned...but all binned chips are special, lol. One cannot make a blanket statement about EE or BE CPUs being binned well. A lot of factors come into play when we start down this road. You will notice that Intel has only a couple of available Extreme Edition CPUs at this time...while AMD has quite few. This is marketing and the fight for market share, not that AMD is making consistently better CPUs. Remember, a CPU of whatever architecture cost what it cost to make, regardless of speed, regardless of having an unlocked multiplier.


I should also point out...many overclockers (the top dogs with money or connections) bin CPUs/Video cards as well. Every CPU is different, you have heard that a million times, but it's true. Some overclockers will acquire (through whatever means) many of the same exact processor. Then bench them to find out which ones go faster. For instance, if you benched ten 955 BE's there would be one or two of them that would be faster than the others, this is also binning. You might have read on the forums that someone is selling or using a "golden" CPU...this simply means it's fast as compared to other CPUs of the same moniker.

Click to expand...

That pretty much sums it up.

So in effect Intel and AMD bin all of their procs in a general sense, maybe not to an extreem but enough to classify them as 955/965/970 for instance, then some people will go through a dozen 970's looking for the ONE that can hit 4.2ghz or something crazy... I always thought there was ALOT more consistency in this stuff...

At work, so i cannot watch that link, but ill watch it tonight if i can, thanks!

bigben6 said:

So in effect Intel and AMD bin all of their procs in a general sense, maybe not to an extreem but enough to classify them as 955/965/970 for instance, then some people will go through a dozen 970's looking for the ONE that can hit 4.2ghz or something crazy... I always thought there was ALOT more consistency in this stuff...

At work, so i cannot watch that link, but ill watch it tonight if i can, thanks!

Click to expand...

There is more than one type of binning. Silicon may be binned for production needs, the 955 is selling good so they will get the 955 multi. Crap one in that batch had a core fail so we will bin them all as X3 though only one on the wafer had a bad core. There is a lot to it but in essence the inital thoughte were correct with an outlier here and there.