% | $
Quotes you view appear here for quick access.

Intel Corporation Message Board

  • marsavian marsavian Jan 3, 2013 9:22 AM Flag

    Nehalem/Westmere/Haswell/Broadwell Intel Cpu Architect reddit session


    great info and insights


    This topic is deleted.
    SortNewest  |  Oldest  |  Most Replied Expand all replies
    • I like the parts where he/she describes the accuracy of the pubs.

      "I also read Semiaccurate for its humor value and to level set."

    • What do you think, Marsavian. Do you still see Intel as a goos stock to own? I think you mentioned once that you thought it would see 40 or 50/share one day.

      • 1 Reply to intelonly_please
      • Yes and Yes. Stock is ridiculously undervalued. x86 is reducing its power greater than ARM is raising its performance and wave after wave of Tick-Tock of both Core and Atom over the coming years will roll back East's cocky ARMy back into the cheap slots. Thus the current P/E is unwarranted and too pessimistic. The stock might take a year to start recovering though as the OEMs complete changing their inventory to the new form factor mixes and the new low power Ivy Bridges and Atoms and Haswells get up to full volume and Windows 8 gets accepted. Just a matter of time though, everybody's roadmap is out there to see for years to come and Intel's is the strongest by some measure.

    • and here's some ...

      'How well do you think journalists (Anandtech in particular) cover your latest architecture? It seems to go over a majority of the tech-news writers' heads...
      When journalists attempt to pry into your CPUs to determine some undisclosed facts that you never released, how often do they hit the nail on the head? Is withholding this information from them just a cat and mouse game?
      Where do you read your tech news?
      Are you worried at all about reverse engineering efforts by competitors (ie is this common in your field)? Are protection mechanisms designed at the manufacturing/materials level or logical/layout level?
      Was Global Foundary's split from AMD a wise move? Purchase of ATI wise?
      Pentium 4 era was obviously a cluster-f. Interesting enough, AMD has started to wander down this path in their latest processors -- hoping to get to higher frequencies with longer pipelines. There has to be SOME technical justification for AMD to 'repeat' the past's mistakes. What technical detail is elusively close yet fails to be reached time and again?
      How many transistors/gates are hand-laid vs computer synthesized on a modern processor? Old Pentium Pro era videos shown at colleges indicated a heavy reliance on manual design for large sections, although I assume this isn't the case anymore.
      Would an undergraduate EE have any possibility of an interview? Or is a Master essentially required?


      I will thank them all for you. The space race is what got me interested in engineering so I know where you're coming from. Here are your answers (some long because you asked some great questions):

      Anandtech and Real World Tech (sometimes The Tech Report) are the best sites with the most accurate information. Especially with Real World Tech, we are sometimes surprised at the accuracy of many of the inferences. Anandtech's latest Haswell preview is also excellent; missing some key puzzle pieces to complete the picture and answer some open questions or correct some details but otherwise great.

      They get close (see above). There are a couple of things to note here: sometimes the architectural information is not enough, the circuit implementation is incredibly important and that is not often discussed. I guess it's lower on the totem pole. Sometimes we do keep some information from the press that end up in patents, conference papers, etc... But eventually we disclose everything, I think is because we try to outdo ourselves every generation as well as being proud and wanting to share our accomplishment. Ask Apple for a disclosure of Swift.

      I like Real World Tech the most and find that Anandtech and The Tech Report do good jobs too. I also read Semiaccurate for its humor value and to level set.

      No and there are little protection mechanisms once it's in customer's hands. By the time they're able to reverse engineering, we're on to the next thing. And even then their implementations tend to not be as good (see AMD power gate efficiency and leakage). Here I referred only to hardware/circuits because security features are a different matter.

      They didn't have a choice if they wanted to stay in business. They do not have enough silicon revenue to sustain it. In retrospect the ATI purchase was necessary, the sad part is they did overpay by a large margin. Also execution missteps in coming out with their "APUs" allowed us to come very close.

      In my mind, Netburst, much as it's maligned, brought some very good things internally for Intel design teams. First, unbelievable circuit expertise (the FP logic was running at 8GHz in Prescott stock!). Next, the trace cache which you can see reimplemented in Sandy and Ivy Bridge. Also, SMT. Building a validation team that could validate the beast pre- and post-silicon. The power-perf thinking i.e. frequency through power savings. Finally, the development of tools and project management required to do that kind of extreme design. All of these learning continue to this day and it's a very large contributor to why in client and server CPUs Intel can sustain the roadmap we have.

      I can't say. But the most important, performance and power sensitive parts are still hand-drawn. Otherwise you can't get past around 1.8GHz on Intel 22nm without losing too much perf from overhead.

      Yes, we have tons of UG interns and most of our hires have BS. MS is always helpful, but do it for your own personal growth and interest, not to get a job. If you're interested PM me.'

      'AMD had to sell their fabs, otherwise they wouldn't be in business today. There are advantages to having fabs. You'll see many things, especially with Broadwell, that you cannot do without owning a fab.'

      ' What are the PC's inside intel that you guys use on a day to day basis like? Are they using hardware that has not been released to the public?

      Are there working prototypes of future processors that aren't even supposed to come out for a couple more years (such as broadwell)?

      As a person who just bought an Ivy-bridged based system, is there anything you can tell me to convince myself to save up for a haswell or broadwell system?


      Yes and no. For day-to-day, our laptops have regular Ivy Bridge processors. Most of the heavy lifting though happens in our datacenters. There we do have cherry picked, specially-fused parts to run our high compute workloads. The most interesting scenario I can recount is we used our Haswell A0 silicon to tape out the subsequent steppings, thus "validating" it.

      Yes. Many in the labs, used in publications. For products, we have things in the labs well in advance because it takes time to get them up to snuff.

      What do you usually do with your system? If you like to overclock, Haswell is worth it (can't tell you why but read the Haswell Anandtech preview very carefully for buried treasure). On-die graphics is improving quite a bit as well. If you're into energy efficiency or even more graphics, Broadwell. I think the tech community will be very pleasantly surprised with Broadwell. But I'm biased, so we'll just going to have to prove it the hard way.'

      'Q:I've been told that Intel CPU's are still completely hand laid out. Is this true? Do you see Intel transitioning blocks to Place & Route for CPUs in the near future? Does intel use a custom toolset or vendor supplied like synposis, cadence, etc...?

      A: All of the analog circuitry, arrays, and performance-sensitive parts are definitely hand-drawn (schematics) and hand laid out. We're one of the few places that actually still do this (apparently Apple does too). You can tell which parts were laid out by hand if you look at die photos.

      As for tools, we definitely use many custom tools. For some things we do use Synopsys and Cadence tools when it doesn't make sense to develop internally. But, I think more interesting, is we develop on Linux systems and use all sorts of Open Source software.'

28.565+0.345(+1.22%)11:46 AMEST