Why does intel gimp some of their CPU lines?

[Homeles]

As someone with a fair knowledge of electronics manufacturing, I suspect they disabled some things after the chip failed testing. They were then marketed to a different segment.

That's the case sometimes, but I'd argue the majority of the time, it isn't.

 

[MagnusTheBrewer]

That's the case sometimes, but I'd argue the majority of the time, it isn't.

I don't know Intel 's numbers but Motorola averages only 27% to 30% "good " parts. I think anyway a company can eke out a few more salable parts, they will.

 

[Rakehellion]

I was NOT talking about being able to o/c one series better than another, or the chopping off cores/cache size or anything like that.

I was talking about CPU features that are removed, for whatever reason, and I cited virtualization as a specific example.

It's the same reason.

 

[greenhawk]

, if all is well ?

in some cases, it is not just all is well but the part that was disabled failed initial tests, so the cpu is dropped to a lower bin grouping instead of through it out.

Though some are disabled on purpose to meet market demand as too many cpus are good enough for higher binning.

you'd think the i7 would come with at least HD 5000. And yes, its all $$$.

It is also the market segment that a cpu is aimed at. A high end user is assumed to have a dedicated GPU.

Though then there are some features the higher end units does not have which might have been a useful thing to have like quick sync for encoding.

For when it doesn't make sense, I take my chip, disable some stuff, lock in lower frequencies, fuse off some caches and sell that version at $50. That way the cheapies in the world can buy my chip at $50 because it's "good enough". And then there's enough added value in the "premium" version for people to spend the extra $150.

that process is also seen in the GPU market as well.

 

[Matt1970]

No, its called crapping on your customers. Same as Microsoft with all their different Windows SKU's. But that gravy train is ending with Google and Apple putting the screws to them. ARM will eventually destroy the Intel we know today.

It's not crapping on your customers. It's just cheaper to build the same higher end processor and just block off the cores or VT or whatever that to completely redesign and retool for ever single GHz change or feature a processor has. That basically is the reason why we overclock. Your 2.4GHz processor is the same as the 3.2GHz processor. And it isn't just processors. I remember running the Omega drivers to unlock my video card and that was damn near 10 years ago. It's either that or charge everyone the same price and everyone gets the same processor. Are you going to be a winner or loser in that scenario?

 

[KingFatty]

Use a thought experiment and pretend that Intel did not have the two separate lines, and it was just one type of chip with everything enabled.

How could they price it fairly? Should they price it low for gamers? If so, they would lose a lot of money because businesses that needed virtualization would just stroll into Best Buy and purchase all the chips for cheap. To stop that, they would have to raise the price to professional levels. But then that's not fair for gamers, who could not afford it.

So by splitting it up into two different types, they can price one high for professionals and low for gamers.

Now, what's the cheapest and most efficient way to create two different types of chips? It turns out, you just make one type, and then you shoot lasers at it or otherwise cheaply cut off the fancy stuff.

So those two factors work together to create the situation that, in retrospect, seems unfair or destructive. Why are they disabling parts of a perfectly good chip? Well, it just ends up the best way to satisfy those factors above.

 

[BonzaiDuck]

As someone with a fair knowledge of electronics manufacturing, I suspect they disabled some things after the chip failed testing. They were then marketed to a different segment.

That's an interesting perspective. MRMT again touched on the logic of my own [rather wordy] analysis.

But an outfit that big -- as big as Intel -- has an army of cost-accountants and marketing staff. There's no doubt in my mind that product-differentiation evolved according to these other practices after Pentium in the '90s. They'd reap more profit with five or ten differentiated products addressing different market segments -- as I'd explained. These things are well-established in corporate strategies across several industries. Even if some justification would come from QC results, the rest of it would come from the non-tech business disciplines exploiting niches in an expanding market. Now that this market is mature, they've established their product line through several generations. Look how they've fine-tuned their products for mobile devices.

There's a seeming paradox -- for instance regarding the old E22x0 dual-core "Pentium" chips. "Why would they price a CPU at $90?" Or "How can they make a profit from such a low-end product?" But in fact, producing a low-end CPU adds to the profit they'll make from other market segments. Once these things are made en masse, routine production costs drop, testing costs may rise but with less significance. So -- yeah -- they would bin the chips, maybe disable some features, and capture the extra chump-change in a market that assures a decently high profit margin to begin with.

 

[MongGrel]

Sales. To make money targeted at various market segments.

Tick Tock.

 

[PPB]

So I guess Toyota crapped on me, because there's a place on my car where fog lights go, but I didn't get them.

Or, it could be that I didn't pay for them.

So the fog lights are there, but they are disabled? If not, its yet another crappy car analogy that doesnt apply here. The logic in the CPU is there, its just artificially disabled. Ht is inherent to the core design, its just disabled later to further segmentate the market (and shaft their customers, as always).

 

[Homeles]

Or, you know, the more logical answer, which is that it's more economically sound to disable features and sell something for less than to not have a cut down version at all.

 

[Meghan54]

So the fog lights are there, but they are disabled? If not, its yet another crappy car analogy that doesnt apply here. The logic in the CPU is there, its just artificially disabled. Ht is inherent to the core design, its just disabled later to further segmentate the market (and shaft their customers, as always).

Well, it may not be as crappy an analogy as you put it. While the fog lights may be absent themselves, the wiring harness for them isn't missing, just taped off so the ends don't hang. But unlike a cpu with features permanently disabled, you can "re-enable" the fog lights.

 

[chrisjames61]

Yea! Because Apple would never make you pay for more memory on your iPhone or iPad, right? Or charge you more for 4G access on your iPad. Or segment the market into 5/5c/5s Iphones.

And they would never avoid a uniform charging standard, so you have to buy a charger, car charger, etc every time you get a new phone, right?

Why does Google charge for the 1Gbit internet if the 5Mbit is almost free? I mean, the fiber is already there, as is the modem. Why are they artificially limiting you?

Why aren't all new features in the Andriod OS backwards compatible? It is almost like they make you want to upgrade!


Sheesh...

Number one, all those Apple offerings are DIFFERENT hardware configurations. The charger was changed ONCE since the iPod was introduced in 2001.

 

[chrisjames61]

So I guess Toyota crapped on me, because there's a place on my car where fog lights go, but I didn't get them.

Or, it could be that I didn't pay for them.

No, the analogy would be that the fog lights were installed but DISABLED.....

 

[TeknoBug]

I7 3770 and 3770S has certain instructions or features a 3770K doesn't have, kind of strange.

 

[DrMrLordX]

Interesting thread from a business point-of-view. Bear in mind that it is desirable for businesses to keep customers happy no matter how poor the alternatives.

That being said, Intel has been using market segmentation for years. What was the difference between a PII-233 and PII-300? They were all Klamath cores. Binning was a factor, but really, the phsyical differences were minimal-to-non-existant. Intel's reasons for doing this should be obvious; furthermore, I don't think the OP is really complaining about that.

The issue with the modern K chips is that you can wind up paying more money to get fewer features, which makes no sense in your typical market segmentation scheme. The processor to which the 4770k should be compared is the 4770. They are practically the same CPU: they come off the same wafer and have nearly the same clockspeed (and, really, the difference in clockspeed and turbo settings is classic market segmentation at work). The k costs more for a 100 mhz bump in base clock speed and unlocked multipliers. Hooray! This makes sense. Pay more for more features. I could whine that all of Intel's chips should be overclockable, but bclk-based overclocking has been mostly gone for a long while (not counting +- 5 mhz bumps and different straps, I guess), and that's just the way it is.

Here's the sticking point: If I buy the 4770k, I would assume that I would get a 100% advantage over the guy with the 4770. I paid more money, so why not? In this instance, however, I don't: I've lost access to some instruction sets that actually could be a lot of fun for an enthusiast to use at home on his/her/its own time. What overclocker with deep pockets wouldn't explore the possibility of running other operating systems in VMs just for kicks (if not for some actually practical reason)?

There is precious little reason for Intel to cripple a chip that's positioned at a HIGHER price point unless:

1). They really think k-series chips are cutting into their workstation CPU sales in some revenue-negative fashion

-and/or-

2). They really expect home users to buy not one, but two Intel rigs for their personal use: one for overclocking and one for VMs.

It could be both. I don't think either line of thought represents rational thinking on Intel's part.

Bottom line is this: I can't buy a k-series chip with all the extensions enabled, no matter how much money I throw at them. Is there a k chip for $20-$100 (or more) available with the extensions available? No. Would they really lose any money on workstation CPU sales or from their inability to sell two CPUs to a person who only really wants to buy one? Probably not. Intel isn't allowing themselves to sell those k-series chips at a lower price point by gimping extensions. The consumer still pays more and gets less.

Bean-counters are notorious for creating situations that make a lot of business sense while still annoying the stew out of customers perceptive enough to notice what's going on behind the scenes/under the hood/etc. This decision of Intel's seems to make no sense at all, even from the perspective of a bean-counter.

If someone can show evidence that Intel is losing money from k chips sales - even high-end k chip sales - cutting into Xeon sales in some revenue-negative fashion (or even in such a way that its producing an imbalance in chip inventories), then it might make sense. There seems to be little evidence supporting that claim.

edit: grammar/word usage

 

[zir_blazer]

Exactly. The whole issue is not classic market segmentation (For example, expecting that Xeon will have ECC Memory Support or any other feature useful for that market segment that isn't found on the Desktop, regardless if the die is physically the same). Is about that on the same line, there are too many possible sidegrades, instead of a more straightforward stair where the higher you go, the more features you have (Which is what AMD usually has).
Spending more on a 4770K and lose features in exchange for the Unlocked Multiplier is totally stupid, it should be the Unlocked Multiplier as an extra feature over what the normal 4770 already have. You are paying more for it.

 

[Imaginer]

The way I look at this, is that those who would be running a VM, would not be interested in stressing the CPU in overclocking and would rather have a guaranteed tested, advertised speed for computing. Overclocking, can affect stability in general and in a VM it is something that I do not think anyone would run in a risk of (generally, there would be many crucial software at work).

The locked multipliers, is a practice that has stayed - mainly because in my opinion in the past, many third party vendors would sell chips that pretty much falls into what some have done in the past as noted by this quote.

Eventually, someone in the Philippines discovered how to re-enable those chips and sell them as the more expensive model for a higher price -- a form of counterfeiting. [And I got one of those processors, too . . .]

If you think businesses are taking advantage of people, do not forget that people INHERENTLY take advantage of people, no matter what side of the river you stand on.

Overclocking is a percentage chance bonus anyways for minimal additional gain at the expense of stability and longevity and not exactly proven (would you complain that you cannot overclock as best as the guy that bought the same brand overclocking chip, they both overclock so all overclocking is equal)?

Running multiple processes of VMs and native OS on an untested spec out of the CPU line is risky and thus some virtualization features are "binned" out, because on that same boat, why should the VM guys pay more for the overclock or even need it to begin with?

What you should be wanting, is another differentiator, at a bit more of a price increase (a Z version if you would) that included both, and you can gladly risk of reducing your CPU life span in an overclock and with virtual machines running.

The next CPU I would get, would be a non K model - cheaper since I do not need to run things off spec much.

 

[Blitzvogel]

I admit to being an AMD fanboy but I'm not dumb enough to think of them as any kind of saviors. It more or less boils down to AMD providing relevant features to me at a good price and not letting Intel have all the marketshare.

 

[2is]

K series are actually higher quality.

What are you basing this on?

 

[DrMrLordX]

The way I look at this, is that those who would be running a VM, would not be interested in stressing the CPU in overclocking and would rather have a guaranteed tested, advertised speed for computing. Overclocking, can affect stability in general and in a VM it is something that I do not think anyone would run in a risk of (generally, there would be many crucial software at work).

To put it another way: there's no real market for VMs on overclocked chips, right? If that's the case, then there's no economic incentive for Intel to deliberately cripple VM-related instructions on the k chips since nobody's going to be using it anyway . . . right?

For that tiny number of people that overclock AND run VMs - you know, the ones that actually know how to stabilize an overclock for mission-critical apps - the k chip is a nice upgrade for some extra cash. Sticking it to them by making them choose between a k or non-k chip for their VM work doesn't make Intel any extra money. It just makes a few people mad.

The locked multipliers, is a practice that has stayed - mainly because in my opinion in the past, many third party vendors would sell chips that pretty much falls into what some have done in the past as noted by this quote.

K chips, by their nature, could be resold as something else entirely if, say, there were some higher-end chips out there that the k chips could be made to emulate via overclocking.

The way the k chips are set up, you don't really save much money buying them (especially not the 4770k) vs. some other part from the same wafer. Intel has mostly killed budget overclocking (boo!). People sure as heck aren't going to fall for a desktop k chip in a workstation masking as a Xeon (the people that, presumably, would have some real need to run VMs at work).

So, while your observation has merit, it's not entirely related to the point being made by the OP: Intel probably isn't losing money because people are pushing 4770k chips on workstation customers in Singapore (or wherever) that are really trying to buy Xeons for VMs. Any customer clueless enough to run a desktop chip in a workstation thinking it's a Xeon probably wouldn't notice the missing instruction sets anyway . . .

If you think businesses are taking advantage of people, do not forget that people INHERENTLY take advantage of people, no matter what side of the river you stand on.

In this particular instance, I think Intel is annoying a minority of enthusiasts in such a way as to make no significant profit.

Running multiple processes of VMs and native OS on an untested spec out of the CPU line is risky and thus some virtualization features are "binned" out, because on that same boat, why should the VM guys pay more for the overclock or even need it to begin with?

Nobody's saying that the majority of VM users should be paying extra for overclocking features that they don't want. If they want a desktop chip for VM use, get a 4770 and call it a day. If they want the Xeon, great, it's there for them too. Leaving all instruction sets active on the k chips doesn't affect that at all.

What you should be wanting, is another differentiator, at a bit more of a price increase (a Z version if you would) that included both, and you can gladly risk of reducing your CPU life span in an overclock and with virtual machines running.

That might work, but super-segmentation would be pretty unnecessary, and the expense involved adding another SKU to their supply chain would probably not be offset by the extra $30+ per chip they'd be making selling to the minority of overclockers that have an avid interest in robust, high-performance VMs. It would have made more sense to just leave all instructions active on the k chips.

The mere existence of k chips is due to Intel throwing a bone to overclockers who feared that all overclocking would be dead on Intel platforms after Nehalem. There are plenty of people (notably AMD fans) that enjoy griping about that. Needlessly stripping away a feature that most overclockers tend to ignore is just another bullet point on the list of complaints that overclockers could conceivably make about modern Intel platforms.

edit: grammar
edit2: forgot the word "be", grr.

 

[Lepton87]

I wonder if those extra instructions that Intel disables on K models could potentially limit OC, it seems unlikely but if that's the case then disabling those instructions makes some sense.
ps. I noticed that people focused on virtualization but with Haswell Intel also disables TSX, which has nothing to do with virtualization, anyway as of now the point is moot because nothing uses those instructions yet.

 

[witeken]

I wonder if those extra instructions that Intel disables on K models could potentially limit OC, it seems unlikely but if that's the case then disabling those instructions makes some sense.

I don't think there are technical limitations. Statement of Intel architect on the subject.

I had not noticed this. I don't work on the SKUing, but I agree with your reasoning. I'll make a case for it thanks for bringing it to my attention.

 

[zir_blazer]

The way I look at this, is that those who would be running a VM, would not be interested in stressing the CPU in overclocking and would rather have a guaranteed tested, advertised speed for computing. Overclocking, can affect stability in general and in a VM it is something that I do not think anyone would run in a risk of (generally, there would be many crucial software at work).

Which is untrue, cause LGA 2011 Core i7 have both VT-d and Unlocked Multiplier. There is not any technical reason for not having both features on LGA 1150/1155 K parts.

Running multiple processes of VMs and native OS on an untested spec out of the CPU line is risky and thus some virtualization features are "binned" out, because on that same boat, why should the VM guys pay more for the overclock or even need it to begin with?

What you do with your computer, is your issue, not Intel. If I want to run something mission-critical on a highly overclocked Desktop part, I can, and no one can stop me of doing so - even through the logical choice would be a stock Xeon with ECC RAM. If it crashes, I'm the one responsible for making the poor choices.
If you want to run out-of-spec, you're responsible for testing it and making sure that stability is rock solid. It doesn't matters what you are going to do with it. The VMs may not even be important at all if you're running a experimental lab and not a production one. Heck, even when overclocking for gaming purposes requires pretty much tested, reliable 24/7 stability, because if you're into online gaming you KNOW how deadly a crash may be, so any type of Multiplayer or online gaming is considered mission-critical for me.

 

[bononos]

I7 3770 and 3770S has certain instructions or features a 3770K doesn't have, kind of strange.

I think disabling vt-d on the k-cpu's makes sense in Intel's market segmentation strategy if Intel doesn't want overclocked cpu's in servers which would affect sales of Xeon.

 

[DominionSeraph]

Bottom line is this: I can't buy a k-series chip with all the extensions enabled, no matter how much money I throw at them.

Yes you can. You just have to step up to LGA 2011. An enthusiast will have an i7 4960X.