Is it time to stop research in Computer Architecture ?
|
From: Rob Warnock on 22 Oct 2009 20:56 Robert Myers <rbmyersusa(a)gmail.com> wrote: +--------------- | Andrew Reilly <andrew-newsp...(a)areilly.bpc-users.org> wrote: | > On Wed, 21 Oct 2009 13:56:13 -0700, Robert Myers wrote: | > > As crazy as that sounds, it's the only way I can make sense of Intel's | > > idea that Itanium would replace x86 as a desktop chip. | > | > I don't think that it's as crazy as it sounds (today). �At the time | > Microsoft had Windows NT running on MIPS and Alpha as well as x86: how | > much effort would it be to run all of the other stuff through the | > compiler too? | | Maybe Rob Warnock would tell us a little more candidly than he did | before just how hard it was to wring those SpecFP numbers out of | Itanium. I think he already said you needed a black belt in | compiling, or something like that. +--------------- Sorry, that must have been someone else. I wasn't ever close to the SGI compilers, and *definitely* not the Itanium compilers. [Networking periphs, I/O interconnects, yes. Compilers, no.] -Rob ----- Rob Warnock <rpw3(a)rpw3.org> 627 26th Avenue <URL:http://rpw3.org/> San Mateo, CA 94403 (650)572-2607
From: Bill Todd on 22 Oct 2009 21:18 Robert Myers wrote: .... > You can ask different questions. Where did the delusion originate > (your question)? Why did Intel hang onto that delusion so long in > spite of evidence that it was a delusion (my question)? The answer to your question seems obvious: once you've placed such a large bet, it's really, really difficult to walk away from it when there's still a chance you can bluff your way into winning it all. I doubt that Intel remained deluded for long - certainly no longer than it took to see the first McKinleys in action internally and recognize that, while a vast improvement over Merced, they weren't nearly what had been hoped for. But the remarkable thing was that most of the rest of the world continued to buy into the original hype well beyond that point - and given such encouragement it's hardly surprising that Intel (and in some ways even more HP) tried to capitalize on that perception of invincibility. Montecito did offer another chance at the brass ring, albeit one of much-diminished dimension - and they hyped it for all it was worth right up until the time that the failure of Foxton to deliver the level of performance expected became a fait accompli. Again, though, this was hardly surprising: given any chance at all that Montecito would be a world-beater, even if only by a small margin, they were hardly likely to downplay its potential before the fact. The fact that Itanic came so close to world domination *despite* its abject failure to deliver on the promises that had seemed to make that domination inevitable tends to prove that the attempt to bluff its way to success was a daring and risky move but hardly an insane one. The rewards for success would have been large enough to justify even the side-tracking of x86 development that allowed AMD to gain so much more traction than it had ever had before - especially given that Intel was strong enough to accept the costs of failure should that occur (as it did). In groping > for an answer to my question, I have wandered all over the history of > Itanium, going so far as to speculate that Intel should have known > better in 1994-95. My speculation could be way off target, but it is > just that: speculation. I think a good case can be made that Intel made at least a reasonable bet in 1994-5 - their major error being to believe what HP (which at that time was a leader in VLIW research) was telling them (another significant error was the way they dove into initial development with a somewhat callow team, but that can be to some degree excused by the fact that they knew that a *very* experienced HP team was working in parallel on the next-generation product). The prospect of a mass market processor unfettered by the existing x86 cross-licensing agreements and which could drive out the existing enterprise-level processor competition at the same time offered such high potential reward that significant levels of risk must have seemed eminently tolerable. I think a good case can be made that Intel *did* know better at least before the end of Y2K (and in some ways before the end of 1996, given how quiet they - especially in the person of Albert Yu - became at about that time regarding world domination) - but while the potential reward by that time had greatly diminished it was still very respectable indeed and justified throwing a good deal more money into the pot to see whether the world could be bluffed into folding (which, of course, significant parts of the world like SGI's MIPS, Alpha, and PA-RISC in fact did, whether because they really believed the hype or just found it convenient to for internal reasons). If they had succeeded after all, the success would rightly have been considered a triumph of business strategy and marketing rather than a triumph of technology. There are profound lessons to be learned in all three areas from Itanic's voyage - and now that the hype machine has finally quieted down perhaps they can be. - bill
From: Robert Myers on 22 Oct 2009 21:22 On Oct 22, 8:56 pm, r...(a)rpw3.org (Rob Warnock) wrote: > Robert Myers <rbmyers...(a)gmail.com> wrote: > +--------------- > | Andrew Reilly <andrew-newsp...(a)areilly.bpc-users.org> wrote: > | > On Wed, 21 Oct 2009 13:56:13 -0700, Robert Myers wrote: > | > > As crazy as that sounds, it's the only way I can make sense of Intel's > | > > idea that Itanium would replace x86 as a desktop chip. > | > > | > I don't think that it's as crazy as it sounds (today). At the time > | > Microsoft had Windows NT running on MIPS and Alpha as well as x86: how > | > much effort would it be to run all of the other stuff through the > | > compiler too? > | > | Maybe Rob Warnock would tell us a little more candidly than he did > | before just how hard it was to wring those SpecFP numbers out of > | Itanium. I think he already said you needed a black belt in > | compiling, or something like that. > +--------------- > > Sorry, that must have been someone else. I wasn't ever close to > the SGI compilers, and *definitely* not the Itanium compilers. > [Networking periphs, I/O interconnects, yes. Compilers, no.] > You're right, and thanks for correcting me. It was "Mortenb, Systems Engineer, SGI Oslo, Norway," on April 20, 2004 under the thread "Sun has switch to Itanium?" on comp.arch. Robert.
From: eternal september on 22 Oct 2009 21:38 "Robert Myers" <rbmyersusa(a)gmail.com> wrote in message news:12b64227-a088-45dd-843a-0df2e78f3a62(a)o21g2000vbl.googlegroups.com... > > Intel couldn't clock it fast because of all the architectural > features, and, if you can't take advantage of those features, it's a > dog. To be fair, it is not the architectural features, so much, as the optimal uarch implied by the naive implementation of those features. The main ones being: 1) Single cycle L1D (in-order core loses huge performance per clock on the L1D, compiler is unable to work around) 2) Huge register file (this is architecture) 3) Huge instruction dispatch matrix (that comes from being 6 wide) 4) For a long time - global stall Ned
From: "Andy "Krazy" Glew" on 22 Oct 2009 23:19
Stephen Fuld wrote: > That all makes sense, but I am bothered by what seems two contradictory > ideas here. Andy says the problem is an algorithm Excel uses that is > O(N**2) compared with one he could write himself that is perhaps O(N) or > O(N log N). I presume this algorithm isn't in the parsing but in the > calculation themselves. On the other hand, Paul said the bulk of the > time is probably in the parsing rather than the recalculation of the > parsed results. I verified that the performance problem is not in the parsing. |