CPU "class system" pricing scheme

[Rove]

Anyone know why Intel & AMD build one part like the i7 3770K or the FX 8350 and then deliberately sabotage it into things like the i5, i3, FX 6350, FX 6300 and FX 4300?

I mean once they have done the R&D to make the nice parts, why spend more money to cripple them and make them worse?

So why the price difference and why bother to make the bad ones at all? Cause it costs them the same to produce a i3 as a i7 and a FX 4300 as a FX 8350. It's all the same materials, machines, labour and technology that goes into either one, the only difference is they spent a little extra on R & D to cripple some of them on purpose.

If they made all good parts as good as they could instead of the "good" and "bad" ones then they could just sell their best at a lower price and still make the same amount of money.

Why don't they do something sensible like that?

Also, then they could keep around older generation parts as "low price" solutions instead of crippling new ones to fill that role. I mean a old AMD Phenom II x6 1100T is still better than a newer AMD FX 4100 and unlike the FX 4100 (32nm) it can be made on old machines in 45nm without having to update or buy new factories to make a part that is actually worse than it.

The same can be said for 3rd gen i3 vs 2nd gen i7, why not just keep making the 2nd gen i7 on old equipment in 32nm instead of making the 3rd gen i3 in 22nm on new equipment that needed to be purchased or upgraded just to do that when the i3 is a worse part?

It'd also make building PCs sooo much easier if all the companies just did their best...

[_I_]

obviously you dont know how a business works, else ford, chevy ect.. would only make $100k cars

not everyone needs a rocket cpu when a slower one will work

and they already have the equipment to make new cpu dyes
nothing new is needed, they just change the mfg process and its done

[Rove]

_I_님이 먼저 게시:

obviously you dont know how a business works, else ford, chevy ect.. would only make $100k cars

not everyone needs a rocket cpu when a slower one will work

and they already have the equipment to make new cpu dyes
nothing new is needed, they just change the mfg process and its done

I have a dream...

Since it wouldn't cost them any extra and people will still have the same amount of $ per year to spend on CPUs why not? They would probably make just as much easily.

I think it's wrong, and don't even get me started on the car industry cause it's off topic. A old Model-T Ford made by hand out of metal that was mined by hand or very primitive machines adjusted for inflation in the present day would cost the equal to about $4000~. With mechanized mining and robotic manufacturing that should put entry level cars down below $1000 MSRP not including shipping or tax. I can't get a entry level subcompact for under $10,000, over double the adjusted price of a handmade, handmined Model-T adjusted for inflation. It's totally extortionate!

What we really need in both Computer and Automotive industries is a independant and capable competitor who is not afraid to do exactly what I have described. A competitor who will then be willing and able to radically undercut their prices because they focus on making only their best and nothing else.

[rotNdude]

There is no sabotage involved, it is all about what it takes to make a CPU. Do you know how a CPU is made? Do you understand that each step in a CPU manufacturing process requires different materials and purity for the end result? Just because the differences may seem rather small when comparing CPUs, there is a lot that goes into the entire process that makes the best CPUs more expensive than the lower cost CPUs.

http://www.youtube.com/watch?v=-GQmtITMdas

For example, cache memory on a CPU die is not the same as the transistors required for the CPU computing sections. Each CPU has to meet certain requirements for performance and this is why some CPUs are binned as a lower component. Nobody wants to throw silicon away if it can still perform as a reliable CPU at a lower frequency or have less cache memory to operate reliably.

When the CPU is designed, it takes into consideration the end price point based on materials, processing and labor involved. That's why you have i3, i5 or I7 CPUs.

[Rove]

Is it really just binning though?

Cause if they are making that many i3's vs i7 then they could really really use a better process.

I think it's deliberate and that they could just as easily stamp 2 cores (i3) or 4 cores (i7) every time the stamp goes down and every time the various other manfacturing machines run. I mean they need to package either one and honestly a few MB of cache is no big deal, AMD has more MB of cache in it's FX CPUs than Intel does in it's i7 and the FX are still much cheaper.

I heard that AMD for example delibertely used to cut cores off of 6 core CPUs to make 4 core CPUs and that they started cutting them because locking them off in the microcode got "unlocked" by various ambitous motherboard manufacturers and so they went and started laser cutting them deliberately to stop people from unlocking them and getting a free 2 cores upgrade. I also know that most of them (but maybe not all) were disabled just to make a 4 core, not because of any defect. That sounds like sabotage to me.

Anyways, no I didn't understand it was a different amount of labour or materials. I think that is the same.

What I understood was that they wanted to make a class system, maybe to give people options, maybe to be more protective of their intelectual property, maybe to scare people into paying more than stuff was worth or maybe just to be "elitist".

I didn't, and don't really understand why they'd do it.

About the car thing, I know that is deliberate. They design a $100K car and while the R & D costs more than a 10K car the manufacturing is the same or at least very very similar.

I know that for cars, and apparently also for CPUs a large part of what they are selling appears to be "fashion" and that is what I am objecting to.

I don't like that they sell CPUs based on fashion and not actual value or cost of manufacturing or R & D involved. The whole i3, i5, i7 thing is to mimic BMW and you can't deny that cars are sold with fashion as the main value point.

And speaking of fashion, wow look at the i7 Exreme (2820-3970x) with their crazy high prices and performance that does not double in spite of the fact that price triples. I don't deny that they are great CPUs and I'm sure everyone owning one feels like (and maybe is) a boss for doing so. It's just outrageously bad value though and I'd rather see the value be better and that sort of quality be more mainstream, specially when and if it wouldn't cost them anything extra or lose them any profits.

Oh and one last question, what is a "badly binned" i7 3820 branded as? Cause there is no integrated GPU on that chip and it's a 4 core, not a 6 core like the rest of LGA 2011. I can't find anything about it being 6 core with 2 disabled cores, if that is what it is. Or is it genuinely minted as 4 cores? I think they bin all or nearly all chips perfect.

[rotNdude]

Rove님이 먼저 게시:

Is it really just binning though?

Cause if they are making that many i3's vs i7 then they could really really use a better process.

I think it's deliberate and that they could just as easily stamp 2 cores (i3) or 4 cores (i7) every time the stamp goes down and every time the various other manfacturing machines run

It is not just about binning. You evidently didn't understand what I said and what the video talked about. It is not a matter of just stamping, it is a very complex process that involves the design of the CPU for lithography, the materials required for getting that design onto the silicon, the conductor paths for the billions of circuits and the quality of the product after it's manufactured.

[_I_]

all of the same gen dekstop i5s are the same die, same with the i7's

they are all graded, and thats where you get the i5 from 3330 to 3570k
ones that have a broken or underperforming gpu get it disabled or downclocked
ones that preform better (use less current) get graded higher

amd also did that with the athlon ii and phenom ii
some were better and graded higher
others had bad cores and were disabled (all and phenom ii with l3 cache are quad dies)
some of the earlier amd pii x2/x3s were good quads with 1-2 disabled to fill inventory for the x2/x3

[Rove]

Okay but Phenom II used at least 2 types of die, a 6 core die and a 4 core die. Why not use only 6 core die?

Also I think at least i3 and i7 use different die.

Also Xeon 8 core and Xeon 4 core is almost certainly either different die or else 8 core die but then 4 cores to be disabled. They could just do same die and get more 8 cores than 4 cores but instead they do more 4 cores than 8 cores. It's the same socket Xeon that they do it with also. I'm pretty sure they are using a whole different die.

And I still want to know, what is a "badly binned" i7 3820 branded as, or is i7 3820 a "badly binned" i7 3930K or better?

[rotNdude]

You still don't seem to understand. If the materials and processing are twice the cost to produce an 8 core versus a 4 core CPU, you don't produce all 8 core CPUs. That just doesn't make sense from a manufacturing perspective. However, if an 8 core CPU doesn't test properly on all 8 cores, you can bin it as a 4 core CPU if the 4 cores test properly.

Regarding your question about a i7 3820 that is badly binned, that is definitely Intel proprietary information. An i7 3930K is a 6 core CPU with a stock speed of 3.5GHz. I seriously doubt that an i7 3820 with a stock speed of 3.6GHz is a binned failed 3930K, but I guess you'll never know.

[Rove]

_I_님이 먼저 게시:

all of the same gen dekstop i5s are the same die, same with the i7's

they are all graded, and thats where you get the i5 from 3330 to 3570k
ones that have a broken or underperforming gpu get it disabled or downclocked
ones that preform better (use less current) get graded higher

amd also did that with the athlon ii and phenom ii
some were better and graded higher
others had bad cores and were disabled (all and phenom ii with l3 cache are quad dies)
some of the earlier amd pii x2/x3s were good quads with 1-2 disabled to fill inventory for the x2/x3

Right so why not only use i7 die? It's obviously better and then they would produce all i7 3770s, i7 3770, i73770K parts only. Am I missing any?

Obviously the i7 die is somehow "so good" that they don't make very many mistakes. There are no "bad" ones. If it's true that they all use the same die then they are only different because of how the micro-code is tuned I think. For overclocking K series, for normal use but preventing overclocking (perfect for kids, employees or clumsy people) the standard series, for low power the S series.

I wonder if it would be possible (without factory equipment) to edit your micro-code and turn a S series to a K series by deleting S series micro-code and installing K series? You'd probably have to be able to speak the hardware language (without micro-code translator) to do it otherwise the chip would just be unresponsive after you deleted it's micro-code.

I know you could use overclocking control to tune a K series to be a like a S series by undervolting and underclocking and stuff.

The question is can you do it with micro-code.

Maybe more importantly, is there any place to legally aquire this micro-code from Intel or would you have to rip it out of a CPU?

I notice standard i7 and S series i7 (3rd gen) are both the same price.
I assume that K series only costs more because of the greater warranty risk they take by selling it with higher clockspeed. Also with overclocking and thus the risk of someone voiding the warranty and breaking it by overclocking but then lying and submiting a false claim.

If you could change a S series to a K series it would probably void the warranty and Intel would at most only warranty it as S series.

The question is, would it be legal to own a K and S series and swap their micro-code?

I guess all this would be impossible without a Dual CPU Socket 1155 board though? So that you could have one in the board for editing while the other ran the system. If you could even edit it through the pin interface and if it isn't permanent non-writable memory that stores the micro-code.

[Rove]

I'm just wondering how they know they will make the most money off of a class system pricing scheme?

After all, no CPU manufacturer that I know of has recently tried for even 1 generation to focus on doing only their best and providing a few convenient options with no "downgrades" or deliberate sabotage to fit a lower price bracket.

[Rove]

I'm not suggesting that they throw any CPUs away or waste anything. If it was true that all CPUs weaker than i7 were only badly binned that would also mean they (i5 and below) were very very rare because yields are pretty high and i7 would be by far the most common CPU, which it's not.

Only that they use only copies of their best die(s) and no inferior ones. IE i7 Extreme dies, i7 normal dies and no i5, i3, Pentium or Celeron dies.

i3, Pentium G, and Celeron are definetly not made from the same die as i7, they don't make THAT many mistakes when making CPUs. They make those CPUs weaker on purpose.

AMD isn't putting out 4 core FX chips that have 4 disabled core I don't think. They have a 4 core die, a 6 core die and a 8 core die and the actual quality of the chips is such that those of the lowest standard still meet the quality advertised for the 8350, the 6350 and the 4350 or whatever they are currently running through production.

That's why they are all unlocked and you can overclock them all, some perhaps slightly more than others.

I believe both they and Intel handle their binning by deliberately advertising and clocking (in the factory) products lower than their potential so that even the lowest binning passes the official specs gauranteed by the brand. Some chips overclock better than others, that is the only real effect of binning.

i3 and i5 aren't just "salvaged" i7s, neither are FX 4100 and 6100 just "salvaged" 8150s. They are made separetly and deliberately in different clases of performance.

[rotNdude]

Once again, there are three major costs associated with making a CPU; material, labor and the capital cost of equipment. Also, don't forget about the R&D. You don't make everything to maximum standards and then bin them down if they are still good overall. When you know the marketplace won't pay the money for the best, you manufacture to the design specs of the specific CPU and the marketplace. Cache memory, number of cores, hyperthreading, etc., all come into play on the design and manufacture.

[Rove]

Okay well I was assuming that manufacturing and materials costs for a 4 core FX or a 8 core FX and a i3 or a i7 are almost exactly the same?

What I currently think is:

1. Materials:
Unless they are using gold or something then materials are very cheap because the CPU is very small. I mean 50 micrograms (or whatever) more copper or other non-precious metal to fill 2-4 more cores worth of CPU, so what? Silicon and packaging size is the same because it needs to fit same socket. Even pin count is the same.

2. Labor:
is similar because they have all the same parts in the same nm process size, just 1/2 as much. The machine will do full or 1/2 in the same amount of time though.

3. Equipment:
The machine that makes the i3 or FX 4000s could instead make i7 or FX 8000s if it was fitted with a different die, which probably costs them as much for any type as it does for the other.

4. R & D
is the same because i3 and i7 use same core architecture, as do FX 4000s and FX 8000s.

Can anyone come up with more in-depth comparison of these items using hard facts?

From what I understand we're all just speculating, one way or another and don't have any hard facts to surely say that a machine making a i3 can't make a i7 at very similar cost instead, or a machine making a FX 4000 can't make a FX 8000 for nearly the same cost to the manufacturer.

I guess I'd have to OEM some chips and get quotes from founderies to see if a batch of i7 or FX 8000s would actually cost me more than a batch of i3 or FX 4000s in manufacturing or just in royalties to Intel or AMD.

One thought I can come up with to support their current system, rather than my proposed one is:
They need a way of testing their products reliability and in the computer field with technology moving so fast doing 3-5 years of QA work is maybe not a good idea. Meanwhile they can do their QA work in your home using the more expensive, less common products which are a small percentage of their revenue and thus a small risk to the company if they ever require a recall. So their lowest end products might use their "safest" and most reliably tested designs, which are also their most obsolete and lowest performing. Meanwhile their highest end and most rare and costly products would be much more experimental aqnd their pricing might factor in the need for more potential RMAs and customer service.

[rotNdude]

Rove님이 먼저 게시:

Okay well I was assuming that manufacturing and materials costs for a 4 core FX or a 8 core FX and a i3 or a i7 are almost exactly the same?

What I currently think is:

1. Materials:
Unless they are using gold or something then materials are very cheap because the CPU is very small. I mean 50 micrograms (or whatever) more copper or other non-precious metal to fill 2-4 more cores worth of CPU, so what? Silicon and packaging size is the same because it needs to fit same socket. Even pin count is the same.

2. Labor:
is similar because they have all the same parts in the same nm process size, just 1/2 as much. The machine will do full or 1/2 in the same amount of time though.

3. Equipment:
The machine that makes the i3 or FX 4000s could instead make i7 or FX 8000s if it was fitted with a different die, which probably costs them as much for any type as it does for the other.

4. R & D
is the same because i3 and i7 use same core architecture, as do FX 4000s and FX 8000s.

Materials are dictated by the amount of materials you place on the silicon in each step of the process. An eight core is ~twice what it takes for a 4 core. You have to remember that contamination or imperfections in any step of the process reduce yields.

Labor is pretty much the same, yes.

The machines are not the same due to the density of the components being produced.

R&D is spread across the architectures. An i3 is not the same architecture as an i7 and an FX 4000 is not the same as an FX 8000.

Rove님이 먼저 게시:

Can anyone come up with more in-depth comparison of these items using hard facts?

From what I understand we're all just speculating, one way or another and don't have any hard facts to surely say that a machine making a i3 can't make a i7 at very similar cost instead, or a machine making a FX 4000 can't make a FX 8000 for nearly the same cost to the manufacturer.

I'm not speculating, I've been involved with this aspect of electrical engineering for nearly 40 years.

Rove님이 먼저 게시:

I guess I'd have to OEM some chips and get quotes from founderies to see if a batch of i7 or FX 8000s would actually cost me more than a batch of i3 or FX 4000s in manufacturing or just in royalties to Intel or AMD.

One thought I can come up with to support their current system, rather than my proposed one is:
They need a way of testing their products reliability and in the computer field with technology moving so fast doing 3-5 years of QA work is maybe not a good idea. Meanwhile they can do their QA work in your home using the more expensive, less common products which are a small percentage of their revenue and thus a small risk to the company if they ever require a recall. So their lowest end products might use their "safest" and most reliably tested designs, which are also their most obsolete and lowest performing. Meanwhile their highest end and most rare and costly products would be much more experimental aqnd their pricing might factor in the need for more potential RMAs and customer service.

You go right ahead and look at the pricing per 1000 and you'll soon see the difference.

[_I_]

Rove, how about you apply for a job at intel or amd