[HN Gopher] Interviewing Intel's Chief Architect of x86 Cores
       ___________________________________________________________________
        
       Interviewing Intel's Chief Architect of x86 Cores
        
       Author : ryandotsmith
       Score  : 134 points
       Date   : 2025-10-09 16:00 UTC (6 days ago)
        
 (HTM) web link (chipsandcheese.com)
 (TXT) w3m dump (chipsandcheese.com)
        
       | brucehoult wrote:
       | Oh em gee ... what a contentless interview.
       | 
       | "We made it wider and deeper".
       | 
       | Gosh. Why didn't anyone think about doing that before?
        
         | saagarjha wrote:
         | Because that costs power and area.
        
           | brucehoult wrote:
           | And it still does.
           | 
           | And the last generation was wider and deeper than the one
           | before it, also costing power and area.
           | 
           | The question that should be asked ... but which would never
           | be answered ... is "What was it that you changed that
           | REQUIRED and ALLOWED you to go wider and deeper?"
           | 
           | It's not a new process node every time.
           | 
           | Theres no NEED to have a massive reorder buffer unless you
           | can decode and dispatch that number of instructions in the
           | time it takes for a load to arrive from whichever level of
           | memory hierarchy you're optimising for. And there's no POINT
           | if you're often going to get a misprediction in that number
           | of instructions. Ok, so wider decode is one component of
           | that. Is there a difference in memory latency as well? Wider
           | decode past 3 or 4 instructions increasingly means that you
           | can't just end your packet of decoded instructions at the
           | first branch -- as you get wider you're increasingly going to
           | have to both parse past a conditional branch, and then have
           | to predict more than one branch in the same decode cycle.
           | You'll also get into branches that jump to other instructions
           | in the same decode group (either forward or backward).
           | 
           | There are all kinds of complications there, with no doubt
           | interesting solutions, that go far beyond "we went wider and
           | deeper".
        
         | porridgeraisin wrote:
         | https://chatgpt.com/share/68ef6cc3-1c48-8013-a545-905af89fbc...
         | 
         | I asked chatgpt to give a contentful summary of the interview,
         | it seems to be more or less accurate, albeit surface level. If
         | anyone is interested.
         | 
         | It gets the "why" but not the "how". Maybe someone here can
         | prompt it further to speculate on the "how". I don't think I'll
         | be able to verify its output well enough to do that.
        
           | mort96 wrote:
           | I'm not sure what you expect to get out of this. How do you
           | make a "contentful summary" of a contentless interview? Where
           | do you get the content from?
        
             | porridgeraisin wrote:
             | By using general knowledge to write e.g what adding a store
             | address unit accomplishes in the context of the rest of the
             | interview. Did you even read the chat?
        
               | MBCook wrote:
               | That doesn't add useful content. It adds definitions.
               | That's just padding.
               | 
               | Only the interviewee can add content.
               | 
               | I'm also of the opinion "I asked ChatGPT for a summary"
               | type comments are very low effort and don't add to the
               | discussion.
        
               | delfinom wrote:
               | Unfortunately our AI future involves many more people
               | refusing to use their brains for more than a few seconds
               | and depend on AI to generate summaries without knowing
               | what parts are hallucinated or even the point.
        
               | porridgeraisin wrote:
               | Or, they read the transcription, didn't have time to see
               | the video interview, and used an LLM to augment it to
               | make sense as prose as an aid to the casual reader. I
               | know a fair bit about the topic at hand:) but not enough
               | to be gung-ho about it on a tech forum frequented by
               | legends.
               | 
               | If you actually went through the LLM output, found
               | problems with it, and then commented this, it would be
               | fine. Until then it's an unfounded accusation.
        
               | MBCook wrote:
               | If they want to use an LLM, they can. I don't see what
               | posting it adds.
        
               | porridgeraisin wrote:
               | "If people want to read an article, they can search for
               | it themselves. I don't see what posting it here adds"
        
               | porridgeraisin wrote:
               | > don't add to the discussion
               | 
               | For sure, I'm against it as well, it's just that in this
               | case the transcription provided in the article was so
               | terse that it was more or less useless. LLMs are good at
               | expanding it to make more sense as prose. If you open the
               | link, that is what the prompt asks it do as well. I'd
               | argue that's useful and not just padding.
               | 
               | > Add content
               | 
               | Yes, I mentioned this in my original comment "not the
               | why" "surface level" etc
        
         | jng wrote:
         | He is no Jim Keller, and the mostly[1] automated transcript
         | makes it read cringe, but it is not at all devoid of content.
         | 
         | Some examples of very interesting, non-obvious content:
         | 
         | * Even if store ports are kept fixed (2 in his example), adding
         | store address generators (up to 4 in his example) actually
         | improves performance, because it frees up load port
         | dependencies. * Within the same core, they use two different
         | styles of load/address address contention mechanisms which he
         | describes as two tables, one with explicit "allows" and the
         | other one with explicit "denies" -- which of course end up
         | converging (I understand it refers to two different encodings
         | which vary in what is stored). * Between cores, they have
         | completely separate teams which reach different designs for
         | things like this. * It was interesting to me to discover how
         | isolated the different core design teams work (which makes
         | sense) * It was interesting to me to picture the load/store
         | address contention subsystem, which must be quite complex and
         | needs to be really fast.
         | 
         | And I stop listing, re different types of workloads, gaming
         | workloads being similar to DB workloads, and even more similar
         | between them than to SPEC benchmarks and so on.
         | 
         | Just go read the interview if you're interested in CPU design!
         | 
         | [1] mostly automated: at least the dialog name labels seem to
         | be hand-edited, as one of them has a typo
        
           | brucehoult wrote:
           | You're right the things you list do contain fresh
           | information. Though the similarity between game logic and
           | business logic is not a new observation ... and web browser
           | in the same ballpark too. I think it's a code size vs data
           | size thing. SPEC programs mostly have a relatively small
           | amount of code, gcc being an obvious exception. And I guess
           | Blender in 2017 FP.
        
           | pixelpoet wrote:
           | I did the transcription, but not the dialogs and labels etc.
           | So I can say with certainty that it wasn't automated :)
           | 
           | What made the transcription "cringe"? I'd like to believe
           | it's accurate.
        
             | jng wrote:
             | Oops, sorry about carelessly throwing the "cringe" label at
             | that. Thanks for the transcript which allowed me to enjoy
             | the content, which I did find very interesting.
             | 
             | I haven't watched the video so I am not sure how he
             | actually talks, but what read cringe to me was things like
             | the following paragraph:
             | 
             | "Stephen Robinson: Yeah. So let's, let's break it down into
             | address generation versus execution. So, when you have
             | three load execution ports, you need three load address
             | generators. And so that's there. On the store side, we have
             | four store address generation units. But we only sustain
             | two stores into the data cache."
             | 
             | Which reads weird. "let's" repeated twice, probably a
             | stutter, could be transcribed just once. The "So" or "And
             | so" the interviewee uses all the time at the start of
             | sentences can also be removed for clearer and easier
             | reading most of the time, without loss of meaning. Some
             | sentences can almost be removed completely as they provide
             | no actual information. The previous paragraph could be
             | transcribed like this:
             | 
             | "Stephen Robinson: Let's break it down into address
             | generation versus execution. When you have three load
             | execution ports, you need three load address generators.
             | That's there. On the store side, we have four store address
             | generation units. But we only sustain two stores into the
             | data cache."
             | 
             | I hesitate to remove "That's there." so I left it. But
             | everything else I removed, it makes it clearer, and I think
             | I'm not being unfaithful to the original. Removing the
             | duplicate "let's" is a given as it's normal to stutter when
             | speaking, but you don't really want to transcribe that
             | unless the goal is to transcribe the talking imperfections
             | we all have. And all the other things I removed, "Yeah",
             | "So", "And so", are basically the same type of thing.
             | 
             | I thought this was automated because it had so many of the
             | meaningless go-to words and hesitations from the original.
             | Now that you mention it, automated transcription would
             | probably never have produced something this good. And
             | otherwise we are talking about stylistic preference here,
             | always subjective -- although I'd definitely prefer the
             | style of transcription suggest here.
             | 
             | Thanks again. I read chips and cheese with interest, quite
             | often, and enjoy it quite a lot. Keep up the good work. And
             | sorry for the careless put-down.
        
               | Neywiny wrote:
               | The way you backpedal when getting called out, now that's
               | cringe lol
        
               | verall wrote:
               | Users willingness to apologize is one of the better parts
               | of this forum.
               | 
               | Lets not discourage it with trite comments like this :)
        
         | misja111 wrote:
         | Well isn't Intel mostly alive by capital injections from the US
         | government and NVidia nowadays? How much content did you expect
         | from a straw puppet.
        
       | norin wrote:
       | yeah strange sort of
        
       | BoredPositron wrote:
       | Odd read especially after that preamble >> The transcript has
       | been edited for readability and conciseness.
       | 
       | Not a lot of novel information either.
        
       ___________________________________________________________________
       (page generated 2025-10-15 23:02 UTC)