Why unlock 4th core on Phenom X3?

[blackrain]

I have been looking to get a 710 or 720. However, I just read that the reason why the 4th core is disabled is because it is defective. The manufacturer can still make money off the chip by simply disabling the 4th core and selling as a triple core. I don't understand why you would want to unlock that 4th core? Has anyone done tests on the 4th core when unlocked?

 

[Denithor]

There are at least two extensive threads on this topic in the forum already. Go read!

Cliffs:
-Fourth core is disabled.
-Some are defective, some aren't.
-Grab a 790GX or 790FX board with SB750 to be able to activate that extra core.
-If you can boot and are stable with it active, congratulations, you win!
-If you cannot, you got exactly what you paid for - a triple core cpu.

 

[Idontcare]

Originally posted by: Denithor
Cliffs:
-Fourth core is disabled.
-Some are defective, some aren't.
-Grab a 790GX or 790FX board with SB750 to be able to activate that extra core.
-If you can boot and are stable with it active, congratulations, you win!
-If you cannot, you got exactly what you paid for - a triple core cpu.

Nicely summarized. Should be a sticky. +1

 

[shaolin95]

Agreed.
Mine is perfectly stable at 3.7GHZ with the 4th core and 1.44vcore. But there are others that are not so lucky so its if you buy one, make sure your expectations are to get a great 3 core CPU so that you are not dissapointed if your 4th core is bad.

 

[WaitingForNehalem]

So if the 4th core isn't defective why would they sell as an X3? They're losing money that way. Maybe the core doesn't pass their standards but is good enough for desktop use.

 

[jones377]

Well, I imagine that what AMD considers a defective core differs from what the people here consider defective

 

[Idontcare]

Originally posted by: WaitingForNehalem
So if the 4th core isn't defective why would they sell as an X3? They're losing money that way. Maybe the core doesn't pass their standards but is good enough for desktop use.

Supply and demand, I think you might be assuming AMD can sell every yieldable X4 at the ASP they currently price the various X4 SKU's.

The same thing (down-binning) happens with Intel chips which is why you can buy i7 920's that overclock just as high as i7 965's. Not all of them of course, but to be sure there are some perfectly capable 965's getting fused multipliers and sold as 920 as there are more than enough 965's rolling out of the fab to supply the 1k ASP demand curve.

Your next question will naturally be "why not slightly reduce price so demand increases to match the supply, thereby maximizing the profit area under the curve?"...all I can say to that is that the proof of the pudding is in the eating and the fact that neither AMD nor Intel do this (given their legions of smart and intelligent MBA's on the payroll) would suggest my/our simpleminded view of supply/demand way oversimplifies the realities of the complexities of their modern inventory management systems.

(i.e. they don't do it because it actually isn't the path that leads to the highest gross margins as we forum CEO's would like to believe)

 

[shaolin95]

There are several theories. One of them I believe is the possibility that the batch that seems to have the best sucess rate 0904, could be part of the upcoming quad processors and its was either cheaper for AMD to produce lots of quads and disable a core on some of those to meet demand for the 720 x3..who knows.
As for what AMD and users consider reliable or stable. I dont see how a processor Prime and OCCT stable at 800 to 900MHz like many users (at leat with the previously mentioned batch) are getting, could be consider unreliable by AMD when they sell them stock at 2.8

 

[dainthomas]

The price difference between the X4 and X3 is so minimal why bother? Especially given the fact that even if the fourth core unlocks, it's probably not gonna OC for crap. The 810 will run at 3ghz no sweat.

I guess that gamble isn't worth $40-50 for me.

 

[Flipped Gazelle]

Originally posted by: dainthomas
The price difference between the X4 and X3 is so minimal why bother? Especially given the fact that even if the fourth core unlocks, it's probably not gonna OC for crap. The 810 will run at 3ghz no sweat.

I guess that gamble isn't worth $40-50 for me.

In some cases - particularly gaming - the X3 matches or even outperforms the X4 810, chiefly due to the fact that X3 has more L3 cache/core than the X4 810 does.

Look at it this way: the X4 810 costs ~40% more than the X3 710.

If you've taken a look at a couple of X3 threads here on AT, you'll find a group of us with unlocked & OC'd X3's. My unlocked X3 710 has gone from 2.6 -> 3.5 Ghz.

 

[Fox5]

Just because a large percentage of the cores in a batch are defective doesn't mean all are. They only test a few, if too high a percent of those are defective, the entire batch gets chucked, even if the 20% in the middle of the wafer were fine. Or something like that.
It's also possible to be just barely defective, like some random instruction that's unlikely to come up. If x87 instructions were defective, it wouldn't matter for x86-64, it drops those anyway, and I think they've largely been abandoned in 32-bit code as well.

BTW, I think triple cores are a real sweet spot right now. Fast enough single-threaded performance, while still having the extra core for multitasking or things that can make use of it. Of course, if Intel ever launches dual core i7's, that could change really fast. I could see a dual core i7 at 3.x Ghz pwning a triple core in most things thanks to hyperthreading.

 

[ShawnD1]

Originally posted by: Fox5
BTW, I think triple cores are a real sweet spot right now. Fast enough single-threaded performance, while still having the extra core for multitasking or things that can make use of it. Of course, if Intel ever launches dual core i7's, that could change really fast. I could see a dual core i7 at 3.x Ghz pwning a triple core in most things thanks to hyperthreading.

This will never happen due to the cost of the motherboard and DDR3 memory. People don't drop $300 on a motherboard (compared to $100) then cheap out and get a dual core processor for it.

The same thing (down-binning) happens with Intel chips which is why you can buy i7 920's that overclock just as high as i7 965's. Not all of them of course, but to be sure there are some perfectly capable 965's getting fused multipliers and sold as 920 as there are more than enough 965's rolling out of the fab to supply the 1k ASP demand curve.

That makes sense for Intel to do that since they're the top dog, but it wouldn't make much sense for AMD. It's not a great idea to screw around when your company is second place and losing crazy amounts of money. If they could sell those X3 as X4 and gain market share, they would. Remember how a $200 Athlon XP would destroy a $200 Pentium 4? They could have charged more, but capturing market share is a bigger priority when your company is #2 (or when you are #1 and want to destroy #2).

 

[Flipped Gazelle]

Originally posted by: ShawnD1

Originally posted by: Fox5
BTW, I think triple cores are a real sweet spot right now. Fast enough single-threaded performance, while still having the extra core for multitasking or things that can make use of it. Of course, if Intel ever launches dual core i7's, that could change really fast. I could see a dual core i7 at 3.x Ghz pwning a triple core in most things thanks to hyperthreading.

This will never happen due to the cost of the motherboard and DDR3 memory. People don't drop $300 on a motherboard (compared to $100) then cheap out and get a dual core processor for it..

You can already get X58 mobos for less than $200, before any sort of MIR.

LGA1366 mobos will be available for ~ $100 by summer's end, unless Intel just wants to keep prices high so as not to intrude on Core i5, due for release around that time.

DDR3 has already dropped precipitously in price, and will continue to do so. You can right now get a 2x2GB set for $55/shipped from Newegg. Yes, that's more than a 50% premium over DDR2, but much lower than it was last month.

 

[veri745]

Originally posted by: Fox5
Just because a large percentage of the cores in a batch are defective doesn't mean all are. They only test a few, if too high a percent of those are defective, the entire batch gets chucked, even if the 20% in the middle of the wafer were fine. Or something like that.
It's also possible to be just barely defective, like some random instruction that's unlikely to come up. If x87 instructions were defective, it wouldn't matter for x86-64, it drops those anyway, and I think they've largely been abandoned in 32-bit code as well.

This is actually not true in any way....

 

[Idontcare]

Originally posted by: veri745

Originally posted by: Fox5
Just because a large percentage of the cores in a batch are defective doesn't mean all are. They only test a few, if too high a percent of those are defective, the entire batch gets chucked, even if the 20% in the middle of the wafer were fine. Or something like that.
It's also possible to be just barely defective, like some random instruction that's unlikely to come up. If x87 instructions were defective, it wouldn't matter for x86-64, it drops those anyway, and I think they've largely been abandoned in 32-bit code as well.

This is actually not true in any way....

I concur. I can see how/why this perception would be created by an industry outsider, but the manner in which WLT is performed eliminates this as a possible outcome in all respects.

 

[shaolin95]

Originally posted by: dainthomas
The price difference between the X4 and X3 is so minimal why bother? Especially given the fact that even if the fourth core unlocks, it's probably not gonna OC for crap. The 810 will run at 3ghz no sweat.

I guess that gamble isn't worth $40-50 for me.

Very wrong amigo....I dont think I would call 3.6-3.8 not ocing for crap....

 

[Denithor]

Originally posted by: Idontcare
Nicely summarized. Should be a sticky. +1

Why thank you, thank you very much!

Originally posted by: shaolin95

Originally posted by: dainthomas
The price difference between the X4 and X3 is so minimal why bother? Especially given the fact that even if the fourth core unlocks, it's probably not gonna OC for crap. The 810 will run at 3ghz no sweat.

I guess that gamble isn't worth $40-50 for me.

Very wrong amigo....I dont think I would call 3.6-3.8 not ocing for crap....

Precisely. If that fourth core activates & is fully functional then it will OC just as well as the others.

As I mentioned above, some portion of the cores are defective while the balance are functional. If you get lucky and get a chip with four fully functional cores you've gotten a bargain.

But - if you need a quad-core CPU - don't play the lotto. Just buy an X4 and be done with it.

 

[Fox5]

Originally posted by: Idontcare

Originally posted by: veri745

Originally posted by: Fox5
Just because a large percentage of the cores in a batch are defective doesn't mean all are. They only test a few, if too high a percent of those are defective, the entire batch gets chucked, even if the 20% in the middle of the wafer were fine. Or something like that.
It's also possible to be just barely defective, like some random instruction that's unlikely to come up. If x87 instructions were defective, it wouldn't matter for x86-64, it drops those anyway, and I think they've largely been abandoned in 32-bit code as well.

This is actually not true in any way....

I concur. I can see how/why this perception would be created by an industry outsider, but the manner in which WLT is performed eliminates this as a possible outcome in all respects.

Alright, then how about what really goes on?

 

[VirtualLarry]

Originally posted by: Fox5
Just because a large percentage of the cores in a batch are defective doesn't mean all are. They only test a few, if too high a percent of those are defective, the entire batch gets chucked, even if the 20% in the middle of the wafer were fine. Or something like that.

I'm pretty sure that AMD and Intel test their chips 100%.

 

[veri745]

Originally posted by: Fox5

Originally posted by: Idontcare

Originally posted by: veri745

Originally posted by: Fox5
Just because a large percentage of the cores in a batch are defective doesn't mean all are. They only test a few, if too high a percent of those are defective, the entire batch gets chucked, even if the 20% in the middle of the wafer were fine. Or something like that.
It's also possible to be just barely defective, like some random instruction that's unlikely to come up. If x87 instructions were defective, it wouldn't matter for x86-64, it drops those anyway, and I think they've largely been abandoned in 32-bit code as well.

This is actually not true in any way....

I concur. I can see how/why this perception would be created by an industry outsider, but the manner in which WLT is performed eliminates this as a possible outcome in all respects.

Alright, then how about what really goes on?

For one thing, microchips are not paperweights -- you can't just test 5% of the production lot and assume all the rest of them are good. Every chip is tested in some way, at the very least for shorts/opens. I can't say too much more w/o talking to someone in Legal, but that 's a start.

 

[Idontcare]

Originally posted by: VirtualLarry

Originally posted by: Fox5
Just because a large percentage of the cores in a batch are defective doesn't mean all are. They only test a few, if too high a percent of those are defective, the entire batch gets chucked, even if the 20% in the middle of the wafer were fine. Or something like that.

I'm pretty sure that AMD and Intel test their chips 100%.

That's it in a nutshell. Every chip, every single chip on the wafer that reaches test, is tested for parametric yield and functional yield first and foremost.

Now there will be certain reliability flags built into the parametrics that if triggered will result in the entire wafer and/or lot to be pitched into the reclaim basket (not sold, not further binned, but entirely scrapped).

But barring the flagging of known reliability triggers, every chip that passes parametric and functional test proceeds to clockspeed test and binning.

Mixed in there somewhere (the sequence is not identical at every IDM) will be some degree of burn-in and voltage binning as well as fuse-blow (multiplier lock, redundant cache-line use, harvesting, etc).

This happens for every IC, logic of memory, analog or cmos. You can't get away with sampling and characterizing a sub-population for the purpose of determining whether the remainder of the untested population is deemed to meet specifications XYZ. IC's aren't steel or ball point pins. Way way too complicated.

(but for practicality reasons it can be cheaper to not bother testing all samples if the first handful of them are rejected, gate oxide integrity for example, if we have a single chip have a GOI issue at test we scrap the entire lot and don't bother risking sending undetected walking wounded chips into the field)

Boeing airplanes are the same way, every aspect of every single airplane to roll off an assembly line is tested to the fullest. Its not a "let's check the brakes on 1 of every 10 planes and call it day" situation.

 

[Denithor]

veri745 and IDC -

You guys seem to have some inside knowledge of how these lines run.

I'm curious about one thing - if they test all these thousands of chips, how do they perform the testing? Bolt each chip down on a motherboard and crank it up? Somehow I doubt it - cannot see them paying for HSF mounting all day long to ensure chips are functional. So how's it done?

 

[Idontcare]

You'll note I made use of the acronym WLT above...this stands for Wafer Level Test.

Each wafer goes into a machine generically called a "tester" which contains a "probe card" which itself can touchdown on one or more chips (depending on tester cost and chip complexity) and proceed to power them up and test them before die cut.

I've seen some pretty mammoth probe cards that could test 16 chips in parallel, the probe card itself (a consumable as the probe tips wear out) can cost more than $10k easy, the tester can be a couple million for the fast ones.

 

[veri745]

In a tester, the chips are not cooled passively. Depending on the type of tester, it may have a thermal head that uses compressed air, LN2, or water, and they can be attached to hydraulics to raise/lower them. Chips are NOT tested using passive thermals like a HSF; they're tested at a specific set point temperature.

Here is one example of a tester used by semiconductor manufacturers:
http://www.credence.com/xweb.n.../Sapphire?OpenDocument