Jukka Larja (roskakori2006@gmail.com) on 9/29/09 wrote:
---------------------------
>AM (myname4rwt@jee.male) on 9/28/09 wrote:
>
>> Seni was spot-on btw;
>
>For most part it seems Seni read my message in a vacuum, which sort of shows that
>this line of discussion has been going on for too long. I wouldn't expect anyone
>but you (AM) to really read these. At least not for the facts.
>
>> your (2) can be true
>> in many situations, but not in general (e.g.
>> caches which increase transistor count can
>> reduce power consumption)
>
>Is that something that doesn't apply to an in-order core? I didn't state it (part
>of the context that obviously needs more work), but of course there is a way to
>use transistors to reduce power. Just add some transistors that shut down the core :) .
>
>The point about register file ports is good, but isn't that just a case where OoO
>gives more performance (or tolerance to hazards) and thus gives you more leeway to implement the file?
>
>I have to admit, though, that I thought my example was simpler than it turned out to be.

Your example has very little to do with the subject of the discussion you joined. If you dispute anything of what I said, please point it out and as clearly and concisely as possible so we can concentrate on that w/o wasting time. Why change the subject?

>> (3) can be true in particular cases, e.g. for
>> comparison of similar devices, both thermally
>> or electrically constrained, but not in general,
>> when the opposite is often true as mentioned
>> (e.g. Vdd is adjusted to make full use of
>> available cooling solution or supply).
>
>I don't expect there to be actual situations where nanowatts (or whatever unit I used) matter. But power does matter.

This is not about that, discussion you "joined" was not about whether power matters or not; heck it was never even about in-order vs OoO in the same power budget.

>> There's nothing about OoOE that imposes
>> such hard constraint on design (at least
>> nothing I'm aware of).
>
>I wouldn't call anything that happens in everything else being equal land to be hard constraint.

It's not quite clear what you're trying to say here. Do you disagree with anything what I said? What's the essence you put in?

>> What's the connection with energy? Cycle time
>> is a function of, among other things, #gates
>> b/w the latches (roughly speaking), and the
>> latter is something that's determined by design.
>
>Out-of-ordering is an extra job and you can't do it without spending energy on
>it. You can't do any work without spending energy, unless you have 100 % efficiency,
>which you never have. (This is still just the not a nanowatt more argument.)

Yes, so what? Please, state clearly what you're trying to support/dispute using energy argument, because from what you say, the connection with your accusation of me "not getting any of the discussion" is far from obvious.

>> I recall I gave you some practical examples to
>> show that what you seem to think is true is
>> really not.
>
>The word "practical" should be enough for you to know that its not what I want.
>Your examples are high-level. I want to look at a single feature, as much in isolation
>as possible. Since that's not the way to actually design or manufacture real products,
>I find it very, very improbable that any examples exist.
>
>By the way, this isn't (as far as I can see) not what Jouni was after and not the
>reason I got into this discussion. I brought these up because it seemed you can't
>see OoOE to have any negative impact to clock speed, no matter how theoretical or
>how unrealistic (or rather irrelevant) in practise.

Once again: no, OoOE is not *bound* to have any impact on clock rate. Your line of thinking -- if I understand correctly -- that *anything* that increases power consumption must have a negative impact on clock rate, is ridiculously far from theory, from practice, and from what was being discussed in the thread you jumped in in the first place for quite clear reasons I gave you clues to. So, do you understand why or do you want a more detailed explanation?

>>> Well, if you aren't allowed to have greater
>>> power, and the clockspeed is the only
>>> variable you can adjust, I don't see what
>>> else to do.
>>
>> Actually, you not only can, but should adjust
>> Vdd along with it.
>
>Why do you think Vdd isn't part of the "everything else being equal"?

And why do you think clock rate is the only free variable to adjust, to begin with? This is not the case neither in practice, nor in theory -- this is either your understanding of chip design, or an attempt at changing the subject.

>> the point is that with such ifs as above,
>> you're changing the game very significantly.
>
>If with "the game" you are referring to what Jouni said, I certainly agree. Look above.

I refer to the discussion that was taking place. If you're really disputing something from that, you shouldn't impose your own context in the first place.

>> In case you argue it makes comparison more
>> fair, then how about we factor in the time
>> to complete a task and thus measure energy
>> spent? This is as reasonable thing to do
>> as allocating the same power budget.
>
>Well, this isn't about fair or reasonable. Your suggestion does show that we think
>of very different things when talking about "everything else being equal". Of course, I already knew this.

Then I have a question: does your "everething else being equal" include die space or not, and why?

>> IOW, there are many different things which
>> can be specified additionally and lead
>> potentially to completely different answers
>> compared to the original setting.
>
>And by original setting you mean what Jouni said? See above.

By original setting I mean the context of the discussion you have jumped in claiming someone was not getting anything.

>> Well, finally some significant steps forward! :)
>
>Yeah, I've been trying to tell you the whole time that your argueing in wrong (or
>different, if that pleases you more) context (that's what those "I don't think you
>understand what I'm saying" were about).

Excuse me, but why does the "wrong" or "different" context apply to one and not the other side? Who are you to decide, let me ask, even if that were true?

>> If you want to specify the context, then this
>> is entirely up to you. Don't blame others for
>> not being able to read your thoughts or
>> uncover the real meaning you put in when it's
>> far from obvious or changes the problem
>> significantly.
>
>Sigh. So you still think that your default context is the right one? Your reading
>comprehension is perfect? That others are to blame if you don't understand what they mean?

Let's see. You jump in a thread w/o caring to read it through, as it turns out, and yet you think it's someone else who doesn't understand? Great. What makes you think you're the one who knows right, esp. considering you're changing the original context?

>You know, that just isn't very productive way of discussion. Instead of spending
>some time to understand my context, or what Jouni's point was, you have flooded
>him with nearly irrelevant examples and me with totally irrelevant examples, and
>flaunted around every meaningless mistake you have been able to squeeze out of us.
>Like, do you actually think Prescott, Pentium M and 90nm issue is somehow essensial?

Well, are you sure you're the one who understands properly everything? If that's really the case, then why do you start with insults rather than simple pointing out of misunderstandings?

>> Apologies then, I only used it because you did.
>
>Accepted. Its not the words, its how you use them.
>
>> Okay then, please fix all of my misunderstandings. :)
>
>Doing my best :) .

Doesn't look that way.

>> Btw, if you consider my style very
>> aggressive, you must have missed far
>> "merrier" times on this board.
>
>Actually, I was lurking around long before the board was split. Extremely aggressive,
>insulting and go **** yourself styles are of course more aggressive than yours :) .
>
>> Mind asking what you teach?
>
>Taught. It was something I did during my studies. Basics of networking (internet
>and telephone) and multimedia, a bit more advanced networking (internet and telephone),
>more advanced networking (internet) and basics of computer graphics (OGL).
>
>These were all part of five-year studies to get one's master's degree, but I don't
>think I can be more specific than that. I don't know enough about school systems of other countries.
>
>I wasn't doing any (or almost any) lecturing, but was deeply involved with exercises
>and often responsible for course bureaucracy, so I have quite some experience on
>dealing with students (and if you count them as people, I could just say experience dealing with people).
>
>> If you *really* seek insightful answers,
>> you should contact Hans as I suggested to
>> you many posts ago, for it seems he no
>> longer attends the board. I don't know
>> anyone here who might know GPU subject
>> better than he does.
>
>I don't usually bother people just to satisfy my curiosity. Now, if someone was
>to write (or point me to) a knowledgeable article about GPUs and OoO, I would probably start drooling (in a good way).
>
>By the way, you are talking about Hans de Vries, right?

Yes. Don't recall any other Hans posting here.

>> Read the thread, really. One of his fantastic
>> arguments was that the last Intel's CPU on
>> 90nm process was Prescott.
>
>I don't think that was especially important in the discussion. But there's really
>no point me trying to guess in detail what Jouni thought. I think he already posted on the issue anyway.

This is not a matter of thinking/context, just a clear example of how he developed his argument.

>> I don't recall you pointing out a *single*
>> pair of contradicting statements from me,
>> I think it was me who asked for clarif.
>> once -- smth. about OoOE being a waste.
>
>Yeah. There I was trying to point out something contradictory. And it turned out
>you totally misunderstood me. I tried to explain more (don't really recall. Perhaps
>at that point I already asked you to just take a look at the three statements, because
>those made my point more clearly (and not clearly enough as we have seen)), but that didn't lead anywhere.

It didnt't lead anywhere because with that you're just changing the problem. Why would you do it?

>> Well, since you referred to theory, I asked
>> you for some references many times, only to
>> see nothing. You shouldn't refer to
>> "theory" when expressing your opinions or
>> views, that's a bad practice that hardly
>> anyone respects.
>
>Well, now that you have seen what theory I (probably) was talking about, does it help? :)
>
>To be honest, since particular details of this whole "the theory" discussion may
>be just some loose language on my side, I don't think there's much point going on.
>I wouldn't want to say I was wrong, because I don't know that, but I rather do that
>than go searching what was said verbatim.

The way it goes from your side, there's very little point really. If you set out to dispute something of what was said, you should begin with pointing it out. The "three basic truths" hardly count as theory, this is just a specific context (same power budget) you have chosen to show how someone else was "wrong" or "misunderstanding".

>> Well, if you like to correct misunderstandings,
>> you're absolutely free to do so. By all means,
>> please point out what you think were the
>> different issues.
>
>What do you think I've been trying to do? Obviously this isn't something I can
>just point out (although I thought it was).
>
>>You know, I decided to check your first post:
>
>Oh my, how rude I was.

Rudeness is one thing, another thing is you say that someone doesn't get something when apparently you didn't even care to read the discussion in full to understand where it comes from. I don't quite understand the reason of such advocacy of someone who resorted to lies (sorry, must be typos) to defend the point, but I don't care that much either. It's simply surprizing and very frustrating to see such cheap shots from someone with self-proclaimed academic background and admiration of theory.

>> "I don't think you have gotten any of the
>> discussion. It is obvious that Jouni has
>> been talking from theoretical standpoint.
>> Sure, you have pulled him in with some
>> idiotic comparisons of real chips, but
>> the fact is that no one has ever made an
>> out-of-order chip by taking and in-order
>> design and slapping out-of-order on it.
>> ..."
>>
>> Theoretical standpoint usually means
>> something abstracted from reality/details,
>> but sometimes it's simply used to defend a
>> totally bogus argument. The arguments were
>> just that -- bogus. OoOE can't "take 2
>> extra stages" etc. as asserted. This is
>> not a "theoretical standpoint", but a load
>> of BS.
>
>As far as that particular "2 extra stages" issue goes, I never had a problem with
>it in context. I mean, you can make a processor without pipeline stages, without
>pipeline. How great is that! No, wait...
>
>The thing is, you can find that "2 extra stages" or something equivalent from many
>sources (it doesn't really matter how many stages it is). If someone says it's in
>H&P, I have no trouble believing that. What I do have trouble believing is that
>some knowledgeable source would assert "2 extra stages no matter what". Perhaps
>that's why I never considered Jouni to mean that either.

Then maybe you simply should get a bit more of the context of the discussion? His 2-stage and other BS was *not* about particular implementations -- where it may hold -- it was OoO in general. That's exactly why he got from me a fair and well-deserved share of examples. Starting to get it now?

>Pipeline stage can be pretty high level issue. Think about classic (5 stage) RISC-pipeline
>used widely to teach the idea of pipelining. Now take Prescott's pipeline. Think
>about the difference in a pipeline stage of classic RISC and that of Prescott. If
>someone just talks about stages, the context is quite important,
>
>Perhaps my interpretation of context was wrong, perhaps not. Either way I don't
>think you can just assert that yours is the only correct one, and then pick on some details to make your point.

Actually, I never tried to switch the context. You are the one who does -- your "three basic truths".

>> Yes, and one doesn't have to look far: in
>> this very thread you can find numerous
>> examples where communication doesn't seem
>> to work or words like "first" are
>> mistakenly interpreted as "last". :)
>
>Should I interpret this to mean that you don't think understanding can be an issue here :) .

I suggest you should interpret it as the smile suggests -- I started putting them more often to help you get the meaning right. To put it even more clearly for you, attempts at defending one's argument with made-up facts (Jouni) or theory (you) is hardly a matter of misunderstanding. Word "trolling" comes to mind.

>> Well, if your goal was *really* to fix
>> misunderstanding, all you had to do was
>> point them out. But instead, you say
>> particular person doesn't get anything
>> and yet you don't point out any
>> misunderstandings. This is what makes
>> me wonder as to what your goal really
>> is.
>
>Not getting the context actually is quite close to what not getting anything means,
>but I should have used more polite language. If I would have thought of a word "context"
>there, I would have dropped it in.

No, what you're actually doing is trying to switch the context of original discussion. It was never about in-order vs OoO in the same power budget -- please read it.

>> Heh, I can tell you that you can find at
>> least one post where one person asked
>> what was meant, but no clarification
>> followed.
>
>If you are referring to that GPU, OoO and waste thing, I said that it isn't useful
>to discuss that unless we can find some agreement on the three statements. If we
>can find an agreement there, we can go back to the other issue, though I don't think
>that will be necessary (after all, I don't think there's a factual disagreement
>there, at least not after we have found any agreement with the three statements).

No, I'm talking about the discussion you jumped in w/o caring to read it. It was me asking Jouni to clarify what he said.

As for your statements, it's a completely different setting from what was discussed, something I have already explained you.

>>> What I have done is spent hours talking to
>>> students and working out wordings to make
>>> exercises (both in Finnish and English) as
>>> unambiguous as ever. Only to see someone
>>> completely ignore the context,
>>> misunderstand the point and come yelling
>>> to us that we are idiots.
>>
>> Well, it appears it only shows that careful
>> working out the wordings was a waste. :)
>> Someone ignored the context anyway, while
>> others probably wouldn't need the extra
>> work you put in to understand things
>> correctly (or just ask if in doubt).
>
>Funny, but your making assumptions out of context again :) (context I didn't give,
>so no wonder. People usually have very weird ideas about teaching. Often very idealistic
>and simplyfied, ignoring the practise. Many ideas that are theoretically speaking sound, simply don't work /selfirony).

I didn't assume anything for what was said, certainly not #students as you suggest below. Just made a direct observation that there were people for whom extra work was unneeded and could have well been unnecessary for others. Yet another example of how you prefer to blame others for "out-of-context assumptions" when it's not even true.

>If a student decides to start argueing what the exercise was about (after we have
>graded it), it can easily take several man hours to resolve the issue. Clearing
>the exercise up after someone asks about it takes less time, but what should you
>do with those students, who already returned their answers and had made the same
>mistake. Whatever you decide, your going to spend quite some time on it. Thus, it
>is actually beneficial to use quite many hours to make assignments as clear as possible.
>
>The context your actually missing is that the number of students I'm talking about
>is from 100 to 300. If you take into account that the exercises are sometimes reused,
>that number goes up n fold. From anecdotal evidence I can deduce that about 15 %
>of students need some extra attension, if you just go with an assignment that doesn't
>have any actual errors and politely answer when one of them asks to specify (if
>you are impolite, they won't ask again, or so I have been told). Getting a percentage
>point or few out of that saves several man hours of work. Exact point where to stop
>optimizing isn't clear, but "several (man)hours" is clearly below it.[1]
>
>Sure, we could have better students, or we could just tell them that they are wrong
>and should stop whining. Or tell them that once they get a real job, they will have
>to deal with much more ambiguous assignments. Some people with much more expertice
>on teaching have decided that that wouldn't be a good idea. I tend to believe them
>for what its worth, but don't ask me cite studies on the subject. My knowledge of the matter is practical :) .
>
>[1] Other important thing to consider here is that if students have to spend time
>deciphering the details of an assignment they are less effectively studying what
>they should. The lost time, of course, doesn't show in any budgetory calculations
>of the university, so it is harder to justify for your superiors than the view I took above.
>
>-JLarja