How does Intel/AMD actually design CPUs?

We all know a pretty good deal about how the physical CPUs are made, however we don't have much info on the actual design process. My question is exactly how does this design process work? Do they first just brainstorm on improvements they can make? How do they decide exactly what is viable and what is not in a CPU design? To they re-cycle designs from generation to generation? About how many people are involved in the design process?

Do note by the design process I'm talking about the time/effort,etc that is used from the point of designing the die to actually running a ES Si run.
91 answers Last reply
More about intel design cpus
  1. Shadow703793 said:
    We all know a pretty good deal about how the physical CPUs are made, however we don't have much info on the actual design process. My question is exactly how does this design process work? Do they first just brainstorm on improvements they can make? How do they decide exactly what is viable and what is not in a CPU design? To they re-cycle designs from generation to generation? About how many people are involved in the design process?

    Do note by the design process I'm talking about the time/effort,etc that is used from the point of designing the die to actually running a ES Si run.



    Well a lot of it is going to be first seeing if they can push technology to a certain point - like 65nm, 45mn etc, and then they'll work out the yield, how much it will cost.

    Then, they have to design where everything goes. They could just plonk it down the same as last time, but I'd imagine for example, that the pipelines may not work having been shrunk, an increase in clock speed may have certain constraints on placements of certain pieces (for instance, two things you want to be close together have to be far apart because of this).

    So it's just a case of deciding what you want, getting it into as small a space as possible without it overheating, and while maintaining as much speed as possible - as the electrical charge will dissapate at distance - while keeping the pipeline small enough not to be too big.

    I imagine cost effective design is part of the equation too.

    Then there is some process cutting the die using UV sensitive materials and templates to etch the design into the silicon, pouring metal in and whatever.

    Just to make you aware, this is just a rough kind of guess, I've got no specific knowledge on it.
  2. *an increase in clock speed may have certain heat constraints on placements of certain pieces (for instance, two things you want to be close together have to be far apart because of this).

    Why can I not edit!
  3. Quote:
    You should ask this to intel technician rather than us.I believe answers of the question you asked can never be made public since Intel and amd guards this secret.

    He's asking about the design process, not the actual design.
  4. Brainstorming: how can the processor be improved? What is currently limiting performance? What features can we currently implement? How much will all this cost? What will yeilds be like? Etc.

    CPU's are just like any other product that improves over time: You take a look over what you have, determine what positive changes you can make, and go through the manufacturing process. CPU's may be very complicated, by they are no different as far as the design process.
  5. I am now picturing a very, very large logisim document...
  6. I think its Magic, Because When the magic smoke leaves a electronic device it stops working :)
  7. gamerk316 said:
    Brainstorming: how can the processor be improved? What is currently limiting performance? What features can we currently implement? How much will all this cost? What will yeilds be like? Etc.

    CPU's are just like any other product that improves over time: You take a look over what you have, determine what positive changes you can make, and go through the manufacturing process. CPU's may be very complicated, by they are no different as far as the design process.

    Ok, that makes sense. What about the actual designing of the CPU Core(s)? How do they know what goes where (ie L3,etc)?
  8. Design teams are eventually created as the designs become larger/featured.
    Its a break down in speciality to a large extent.
    Diversity in todays chips are so grand, no one team can do it, not in a fast paced environ
  9. Shadow703793 said:
    Ok, that makes sense. What about the actual designing of the CPU Core(s)? How do they know what goes where (ie L3,etc)?

    It be a dark art involving black magic, elves, goblins and pixie dust!
  10. And dont forget the lil dogs too!!!
  11. ^ Ok, that's just creepy....
  12. JAYDEEJOHN said:

    I can't comment on that pic due to the ToS! :lol:
  13. heheh, too true
    whats really creepy is, its true, the dog with the longest tongue
  14. What city do you live in? If you are in Tokyo, London, Munich, Frankfurt, Singapore, Beijing, Columbus or Las Vegas, I can sit down with you and give you the process. You'd have to buy me a beer first. And the coversation may take a couple of beers.

    I live in Austin, so that is an option as well. Way too much detail to put in a forum post.
  15. Again I could make a comment pertaining to videos and women with strange sexual fetishes but the ToS prevents me from doing so. :whistle:
  16. Ermmm.... :heink: Well ok then :lol: Anyway, i think companies should take more chances when designing them. Intel hasnt done much to the core architecture for years. Core to i7 wasnt as big of a jump as people think with arch, SB is a very similar arch, and going in both directions beyond them they havent done *much*, granted they have added quite a decent bit of performance. AMD hasnt really changed much since 2003? K7? Luckily BD will change things up a little, whenever it comes around... :sarcastic: But my point is, Intel has all the money in the world, and they have the vast market share, as maybe 2 in 100 people have heard of AMD, where as probably 30,50, maybe 70 people have heard of intel. Intel has such a considerable market lead, and they could have several netbursts in a row and still maintain the lead in market share. They should try something far from usual, change everything up, actually think. Not just shrink die, add transistors, fiddle with this, and then ship it out. AMD has done that for several years, and BD will be a welcome change. I just think they are too cautious...
  17. jf-amd said:
    What city do you live in? If you are in Tokyo, London, Munich, Frankfurt, Singapore, Beijing, Columbus or Las Vegas, I can sit down with you and give you the process. You'd have to buy me a beer first. And the coversation may take a couple of beers.

    I live in Austin, so that is an option as well. Way too much detail to put in a forum post.

    Damn it! That's like at least 1/2 way across the country (US. I'm in VA).
  18. Quote:
    Core to i7 wasnt as big of a jump as people think with arch, SB is a very similar arch, and going in both directions beyond them they havent done *much*, granted they have added quite a decent bit of performance.

    I wouldn't exactly call going from the memory controller in the NB to an IMC a small jump...
  19. And that and HT was basically it. Like i said, they do a die shrink every once in a while, shove some more transistors on there, increase clock speed, maybe add and tweak a feature, and thats it. I think they ashould actually think MORE outside of the box. Im not saying they arent changing anything, im just saying they just keep adding and tweaking, not so much reworking.
  20. Architecture changes that have great effects, such as from P4 Netburst to Core2, just do not happen that often.

    Feel free to think out of the box when you have your own semiconductor company. :)
  21. Did i miss something!!!!!!

    Here's on page which goes through the basics of chip making. That too the 32nm ones! The 45nm one's are more detailed and would give you a better idea.

    http://www.intel.com/pressroom/kits/chipmaking/index.htm
  22. hell_storm2004 said:
    Did i miss something:

    Here's on page which goes through the basics of chip making. That too the 32nm ones!

    http://www.intel.com/pressroom/kits/chipmaking/index.htm

    Where does it talk about the design process?
  23. It just gives you an idea of how chips are made from sand. Not the design and all. I don't think Intel would put those onto a public website for all to see! :)
  24. Why not? It's not like I'm gonna get some sand and make my own!
  25. ares1214 said:
    And that and HT was basically it. Like i said, they do a die shrink every once in a while, shove some more transistors on there, increase clock speed, maybe add and tweak a feature, and thats it. I think they ashould actually think MORE outside of the box. Im not saying they arent changing anything, im just saying they just keep adding and tweaking, not so much reworking.

    This and your other post are layman opinions. And you AMD angle has yet to be proven. lol
    The op asked a question about how they may go about design, and you look for a way to critique Intel VS Amd ? They may design many things VS what they come to market with also.
    Anonymous said:
    Architecture changes that have great effects, such as from P4 Netburst to Core2, just do not happen that often.

    Feel free to think out of the box when you have your own semiconductor company. :)

    +1
  26. notty22 said:
    This and your other post are layman opinions. And you AMD angle has yet to be proven. lol
    The op asked a question about how they may go about design, and you look for a way to critique Intel VS Amd ? They may design many things VS what they come to market with also.

    +1


    AMD angle? I wasnt critquing Intel VS AMD, i was saying they should think out of the box more, how is that putting the 2 against each other?
  27. ^I liked the doggie picture, this one hurts my thinky gland :P
    P.S. that model of dog is obsolete now having been replaced by the new damn big rottwieler edition :P
    Moto
  28. jf-amd said:
    What city do you live in? If you are in Tokyo, London, Munich, Frankfurt, Singapore, Beijing, Columbus or Las Vegas, I can sit down with you and give you the process. You'd have to buy me a beer first. And the coversation may take a couple of beers.

    I live in Austin, so that is an option as well. Way too much detail to put in a forum post.

    I live in Las Vegas, and I can buy beers.
  29. Quote:
    Why not? It's not like I'm gonna get some sand and make my own!

    Exactly! It cost a ton of money to build a fab.
  30. ScoobyJooby-Jew said:
    I live in Las Vegas, and I can buy beers.


    December, Gartner Data Center Conference. I like good beer ;)
  31. ares1214 said:
    i was saying they should think out of the box more


    Funny, with Bulldozer we went "outside the box" a lot more than any x86 design in history and yet people criticize the desgin and say that we aren't making real cores because they don't look like intel's.

    Just my 2 cents.

    And, actually, pitting the two of us against each other is how all of the advancements happen in the industry.

    Just think of the pace of change in CPUs in the past 7-10 years. Competition breeds innovation.
  32. jf-amd said:
    Funny, with Bulldozer we went "outside the box" a lot more than any x86 design in history and yet people criticize the desgin and say that we aren't making real cores because they don't look like intel's.

    Just my 2 cents.

    And, actually, pitting the two of us against each other is how all of the advancements happen in the industry.

    Just think of the pace of change in CPUs in the past 7-10 years. Competition breeds innovation.

    I guess you better really step outside the box on the next one then, how about some sort of Isolinear/Dark matter hybrid CPU using a Plasma injector interface?
  33. Not until we can work out the interface to the flux capacitor.
  34. Oh yeah, I keep forgetting that bit, have tried using some unobtainium?
  35. It has been obtained.
  36. jf-amd said:
    December, Gartner Data Center Conference. I like good beer ;)

    I'll be around. With plenty of Milwaukee's Best and Mickey's Bigmouth :D
  37. Ping me in December, I go out for beers with folks all around the world if they are interested in talking processors. However, you won't get any non-public data.
  38. jf-amd said:
    Ping me in December, I go out for beers with folks all around the world if they are interested in talking processors. However, you won't get any non-public data.


    Surely that depends not so much on the quality of the beer, but the quantity? ;)
  39. I Life is too short for cheap beer. I only drink quality, which is more important that quantity.
  40. A large quantity of quality beer might be more persuasive :whistle:
  41. Yeah jf, how bout you throw on Bulldozer performance and TDP. Why not socket capabilities too! :lol: Yeah, i doubt it...all i know is mid to late october, and around 30%.
  42. ^ Do realize that JF works with the Server division. He can't tell you the release dates,etc (due to NDA,etc) so don't bother.

    Quote:
    Ping me in December, I go out for beers with folks all around the world if they are interested in talking processors. However, you won't get any non-public data.

    In that case, can you give us the gist of the design process?
  43. Shadow703793 said:
    ^ Do realize that JF works with the Server division. He can't tell you the release dates,etc (due to NDA,etc) so don't bother.

    Quote:
    Ping me in December, I go out for beers with folks all around the world if they are interested in talking processors. However, you won't get any non-public data.

    In that case, can you give us the gist of the design process?


    Yes i was being sarcastic as i know he cant divulge this sort of info :lol:
  44. ares1214 said:
    Yeah jf, how bout you throw on Bulldozer performance and TDP. Why not socket capabilities too! :lol: Yeah, i doubt it...all i know is mid to late october, and around 30%.


    TDPs will be identical to existing server products. Sockets will be the same as existing server products.

    A 16-core Interlagos will have 50% more throughput than a 12-core Magny Cours.


    Who says I can't share data?
  45. jf-amd said:
    TDPs will be identical to existing server products. Sockets will be the same as existing server products.

    A 16-core Interlagos will have 50% more throughput than a 12-core Magny Cours.


    Who says I can't share data?


    Now, we were told this a long time ago, 10-100 watts, and 50% more thruput. At the time, i wasnt very impressed, as 33% more cores got you 50% more thruput, so it wasnt exactly the most impressive numbers. Now im wondering, is the 16 "core" really an 8 module?
  46. "Then throw some light towards Ati 6000 series as well.Since AMD is making all its profit from Ati graphics.Becoz phenom cpu are all worthless compared to intel i series and only results in losses."

    Where have you been?
    We are talking a new arch here
    Its not like Intel was still making P4s, or AMD still making P2s.

    So, to expect the same performance in a radically new design is beyond me.
    My guess would be everyone will appreciate the new arch in MT scenarios, its power and for some, its scalability.
  47. JAYDEEJOHN said:
    "Then throw some light towards Ati 6000 series as well.Since AMD is making all its profit from Ati graphics.Becoz phenom cpu are all worthless compared to intel i series and only results in losses."

    Where have you been?
    We are talking a new arch here
    Its not like Intel was still making P4s, or AMD still making P2s.

    So, to expect the same performance in a radically new design is beyond me.
    My guess would be everyone will appreciate the new arch in MT scenarios, its power and for some, its scalability.


    From what ive seen from the arch, and AMD blogs about it, what seems like people will appreciate BD for the most is price to performance. AMD said i believe 80% of work has nothing to do with the FP? So they took 1 out for every 2 cores. I think that will make things perform pretty well, but most of all be a very efficient CPU, as far as for your $. Just me though. I could be wrong. BD could blow everything away and be priced at $1000, or just not compete with whats out at the time. Who knows... Besides JF :hello: :whistle: :lol:
  48. ares1214 said:
    Now, we were told this a long time ago, 10-100 watts, and 50% more thruput. At the time, i wasnt very impressed, as 33% more cores got you 50% more thruput, so it wasnt exactly the most impressive numbers. Now im wondering, is the 16 "core" really an 8 module?



    33% more cores adding 50% more performance doesn't sound interesting?

    When Intel added 50% more cores with Westmere they got 33% more performance.

    Those numbers clearly lay out that on a per-core basis the core performance is getting better. And that is a comparison of server workloads on a fully utilized server, don't try to do client comparisons because the workloads are different.
Ask a new question

Read More

CPUs