26

Someone was telling me that lower-price or lower-spec computer parts such as graphics cards and central processing units are actually just higher-spec, higher-price units that have manufacturing faults.

So instead of the manufacturer throwing them out, they simply disable the faulty bits and sell it as something else.

For example if a CPU is made with 6 cores, but 2 cores are faulty (or don't meet quality control standards), they will disable access to the 2 cores and simply sell it as a quad-core processor.

Is there any truth to this? Could this person have meant something else?

Rowan Freeman
  • 1,176
  • 2
  • 10
  • 15
  • 2
    I worked as an engineer at Intel and I know there were multiple lines of chips. Some were older tech with refined manufacturing processes. It's possible that this happened at the server level chips with very large numbers of cores, but I don't know; even there there are multiple lines. – William Grobman Apr 06 '13 at 04:35
  • Welcome to Skeptics. Please read the [Welcome to new users page](http://meta.skeptics.stackexchange.com/questions/1505/welcome-to-new-users) to understand how this site works and link to a notable claim of someone saying this ("someone was telling me is not notable") – nico Apr 06 '13 at 10:28
  • 2
    Note, that this really applies mostly to integrated circuits created via photolitography with detail size well below light's wavelength. So it's not like it's generic case for all computer parts. – vartec Apr 08 '13 at 13:53

2 Answers2

32

This is true. I'm not sure it is true in all the examples you gave, but it is true for some multi-core processors.

From Gizmodo:

Part of AMD's multi-core Phenom blast today is the Phenom X3 8000, "the world's only triple-core x86 processor," which we heard about a few months ago. They're supposed to bargain chips for budget consumers, but they're a nicer bargain for AMD, actually, since it lets them dump bug-plagued quad-core Phenoms by disabling a core.

This article says the practice is used by Intel as well:

The reason for this is that the disabled cores are turned off for a reason: they failed factory tests. Cores can fail for any number of reasons, including defects in the silicon, problems running at full frequency, or a bug introduced during manufacturing. Both AMD and Intel disable CPU cores for this very reason.

This article shows the practice is also used by AMD with L3 cache:

When AMD produces a Phenom II die if part of the L3 is bad, it gets disabled and is sold as an 800 series chip. If one of the cores is bad, it gets disabled and is sold as a 700 series chip. If everything is in working order, then we've got a 900.

Intel has done this with entire on-board graphics processors (reference):

Intel recently released a couple of Ivy Bridge based processors that have disabled the integrated graphics completely, the 3350P being one of them. This allows Intel to sell processor die that might have a defect on the GPU portion to increase the relative yield rate of their 22nm process and also gives them another weapon to fight off any pricing competition from AMD.

IBM holds a patent (US7610437) on a particular method for doing this:

If a defect is detected in a single functional unit, such as a synergistic processor core, the defective synergistic processor core can be disabled and the multiprocessor can still be used with the remaining functional unit(s) as a partial good multiprocessor chip.

[...]

For example, if a single processing unit is defective, the defective processing core can be disabled and the multiprocessor can still be used with the remaining functional processing unit(s). Such a multiprocessor chip is a partial good multiprocessor.

One of their claims points to the mechanism for doing this:

blowing a fuse associated with each defective processor core to disable each defective processor core if the number of defective processor cores does not exceed the threshold.

From a presentation by Intel and University of Michigan:

Defective core options: disable or salvage. Disabling wastes entire core even for minor defect.

This is a common enough practice that some hardware research focuses on making use of those faulty cores: Putting Faulty Cores to Work. In that article, they make a statement about how common disabling is:

Due to the inherent irregularity of the non-cache parts of the core, it is well known that handling defects in these parts is challenging. A common solution is to disable the faulty core.

Community
  • 1
  • 5
    I would qualify this by saying it is true in this one case. This isn't necessarily demonstrative of the broader scope (or even that of all multi-core processors) - it sounds more like this was a local case. –  Apr 06 '13 at 06:00
  • -1: "some people have been able to re-activate that fourth core and occasionally it works fine". How many is some and how frequent is "occasionally"? The link speaks of "a korean user", that is hardly any proof. – nico Apr 06 '13 at 10:33
  • @Sancho: that is why I would like to see some references that are a bit more "strong". Of course I am not saying that such refs should be easily found, but still... – nico Apr 06 '13 at 17:03
  • @Sancho: much better now! +1 – nico Apr 06 '13 at 21:45
  • Is IBM patent about this? IIRC what made IBM CPUs unique was disabling faulty cores **at runtime**. Which isn't the case for x86 – vartec Apr 07 '13 at 10:51
  • 3
    I also remember reading that processors from the same series but with different clock frequencies often come from the same production line but had different test results. I.e. the top X% is sold as 3.33GHz, the next top Y% sold as 3.17GHx and so on. – MatsT Apr 08 '13 at 13:46
  • The same is true with memory modules that run at lower speeds. They're just higher-speed DIMMs that failed QA at high speeds. – forest Jan 19 '21 at 02:00
8

Generalizing user5582's answer, this is known as "part binning." It is common in the semiconductor industry, and in fact in many other industries where imperfect parts can still be sold.

Modern microchips often contain millions, even billions of transistors. The probability of them all working perfectly is low, so the parts are designed such that bits which don't work can be disabled. For example, if a CPU has four cores but two of them don't work, it can still be sold as a dual-core device. If it normally has 128kB of cache memory but part of it doesn't work, it may be possible to disable the broken bit and still sell it as a CPU with only 64kB of cache memory.

Binning is an effective way for manufacturers to keep costs down and manufacture high end parts.