8
\$\begingroup\$

Intel produces families of seemingly similar microprocessors. For example,

Does (or likely would) Intel produce processors like these three

  1. on separate manufacturing lines;
  2. on the same manufacturing line, but in separate runs on different days;
  3. on the same line all in one, indiscriminate run, and only later -- at a test stage -- grade the processors, sort them by grade, and assign them model numbers accordingly; or
  4. in some other way I do not understand?

All of these seem plausible to me except maybe option 1, but what seems plausible to me may have little to do with how a firm like Intel actually manufactures parts.

Please feel free to construe the question broadly. I am most curious to learn the basics of how such manufacturing is organized in modern practice.

UPDATE

@Shantam gives a better word to use with the search engine: binning, rather than grading. Searching with @Shantam's word, one finds @nik's interesting comments three years ago on Superuser.com:

Actually, manufacturers are a smart lot. They 'bin' their produces into different levels of failures. A partly failed cache in a processor instance could become the 'lesser cache, cheaper version' instead of going into the trash bin. Works quite well with the amount of failures seen in fabrication and the surface area of such memory modules (entire cores are 'wired down to sell the instance as a lower range processor -- the Phenom X3?). Nothing wrong in this, and the overclockers are happy to know such things.

The overclocker angle goes this way, if a processor cannot run (heats up) beyond certain frequencies, it is binned to a lower freq target. You get a E6300 C2D (which an overclocker can push up to a higher one with better cooling and maybe good luck on the manufacturers strict 'binning' policies that might have erred towards the lower frequency bin.

\$\endgroup\$
7
\$\begingroup\$

Most microprocessor manufacturing (along with countless other devices) undergo the process of binning: all similar products are made at once, and depending on their performance, are placed into "bins" (groups) of similarly performing products, and then packaged and sold accordingly.

In the case of Intel processors (AMD is similar), generally processors within the same line are manufactured together, and are binned according to their stable clock frequency. You can tell when a processor is part of the same "line," by looking at the core codename, or if that is not specific enough, the features of the core (as mentioned by embedded.kyle, the i5 doesn't have hyperthreading, but the i7 does, even though both in question are "Sandy Bridge").

Sometimes a higher-end processor that fails can still be sold as another. An example I know first-hand is that the M0 steppings of the old Northwood-core Pentium 4's (130nm process) were actually failed Gallatin-core processors (which was the core for the P4 "Extreme Edition"). Similarly, a lot of people had/have luck unlocking extra cores, caches and shader units on various CPUs and GPUs. For example, it is quite common to be able to buy a mid-high range video card (take for example, the AMD Radeon 6850) and flash it with the BIOS of the higher-level card (the Radeon 6870, in this case) and gain the extra things that card has (some extra shader cores). This also has to do with binning during the manufacturing process.

This sort of thing drives overclockers to take good note of the stepping, place of origin, and batch number of their processors. When word gets out that certain batches of processors are overclocking better than their not-as-potent brethren (same CPU, mind you, just made at a different time or place), they become more in demand.

If you're interested more, definitely search "CPU Binning," or read up at some forums. I'm a member at www.overclockers.com, and the forum there is quite welcoming and has a wealth of past and current knowledge (along with an abundance of fantastic members).

\$\endgroup\$
  • 1
    \$\begingroup\$ That is a very good example. Another is the AMD Athalon. You could unlock the old AMD Athalon XP chips by physically altering the exposed pins/pads on the top of the chip: sharkyextreme.com/guides/hwGuides/article.php/10709_1009731_7/… \$\endgroup\$ – embedded.kyle Jun 14 '12 at 12:52
  • \$\begingroup\$ I was going to include that (I vividly remember people taking x-acto knives to their processors), but figured that was slightly different than the type of binning i was describing. \$\endgroup\$ – Shamtam Jun 14 '12 at 23:10
  • \$\begingroup\$ Thank you for the helpful search term: "CPU Binning." (I had been searching "CPU Grading," which did not find the right things.) \$\endgroup\$ – thb Jun 14 '12 at 23:54
6
\$\begingroup\$

The i5-3320M and the i5-3360M would probably fall under Option 3. They are most likely being produced using the same die and testing determines the highest stable frequency that they will run at. Minor variations in manufacturing process will lead to some chips not being able to run stably above a certain clock frequency.

It's these minor manufacturing differences that can also explain why two identical setups that are being overclocked cannot achieve the same frequency. Even the same model chip will have manufacturing differences from one lot to the next, and to a lesser extent, one chip to the next. When you push the limits of the chip, these differences become more noticeable in that they affect the chip more.

For instance, I've read of a few people being able to run the same processor as I have stably at frequencies above 4.4GHz. However, I, and most others that have tried, have trouble running anywhere above 4GHz. Though there are also differences in cooling solutions and peripherals to take into consideration, all else being equal, I doubt I'd be able to attain 4.4GHz because of luck of the draw. I didn't get a "perfect" chip. If you hunt around on overclocker forums, you can read a lot more about this.

In terms of manufacturing processes in general, Option 3 is also how resistor tolerances are determined. All 1KΩ resistors come off the same line. They are then tested and packaged at 5%, 1%, and 0.1% tolerances depending on how they test.

The i7 is a different die and could fall under Option 1 or 2 though most likely Option 1 in order to keep up with demand and reduce costs. Any time you have to shut down a line to change out a die, you're losing money. For an outfit like Intel, it's cheaper to run separate lines.

\$\endgroup\$
  • \$\begingroup\$ My search-engine searches had not been turning up answers to my question, but I had never thought to add the word overclocking to the searches. Adding this word turns up some interesting results. \$\endgroup\$ – thb Jun 13 '12 at 21:15
  • \$\begingroup\$ Thank you for the interesting information that the i7 is a different die than the i5. Can you tell me how you know? The reason I ask is that, with a search engine, I only come up hits like this: pcmag.com/image_popup/0,1740,iid=312413,00.asp \$\endgroup\$ – thb Jun 13 '12 at 21:26
  • 2
    \$\begingroup\$ In that picture, look at the bottom. There is more to the chip than what they show there. The main difference between an i5 and an i7 is that the i5 has it's Hyper Threading disabled. Now weather they do that in the actual die or if they do that via some processes further down the line, I don't know. I'd wager that it's done via some supplemental process. But the different dies come into play in the cache size. \$\endgroup\$ – embedded.kyle Jun 13 '12 at 21:36
  • \$\begingroup\$ Though now that you've got me thinking about it, it's entirely possible that they start off the same. Use one die on one line to etch the main processor then split the line into i5 and i7. The i5's get etched with one cache die and the i7's with another. Then the i7 undergo another process to enable their Hyper Threading. \$\endgroup\$ – embedded.kyle Jun 13 '12 at 21:38
  • 1
    \$\begingroup\$ In the case of multi-core microprocessors, a 3-core processor may just be a 4-core processor in which one of the cores failed to meet specs during testing. Refer to this article on Tom's Hardware site. Read the comments to the article for some additional twists (i.e. the 4th core may not even have tested bad, but the manufacturer was short of 3-core processors). \$\endgroup\$ – tcrosley Jun 13 '12 at 22:10
2
\$\begingroup\$

Two example from an older era, but interesting in context:

The 80486SX was an 80486 with an intentionally destroyed FPU, probably salvage from 80486DX that had faulty coprocessors. The 80487 was a complete 486DX, and mainboard circuitry disabled the 486SX completely when a 487 was plugged in.

Any classic 40 pin 8051 (ROM/OTP) microcontroller can be (and likely has been, if there was overstock of 8051 with unwanted code loaded into it) perfectly sold as an 8031 - it behaves as an 8031 if you wire it up like an 8031. This also applies to the 80C517/80C515 and to many chips in the 8096/80196 family.

\$\endgroup\$
  • \$\begingroup\$ It's too bad that your answer, responding to my five-year-old question, won't get more attention and upvotes. It's an interesting answer, +1. Wire it up like an 8031? That's funny. \$\endgroup\$ – thb Jun 20 '17 at 15:56
  • \$\begingroup\$ See the datasheet and how the EA pin works ... :) Actually tried it with 8051s salvaged from VCRs, Monitors, Keyboards - they did fine in 8031 circuits :) \$\endgroup\$ – rackandboneman Jun 20 '17 at 16:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.