Maynard Handley

@handleym99

ex South Africa; ex Cornell Physics; ex Apple QuickTime

Vrijeme pridruživanja: rujan 2011.

Tweetovi

Blokirali ste korisnika/cu @handleym99

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @handleym99

  1. Prikvačeni tweet
    17. srp 2019.
    Odgovor korisniku/ci

    It’s part of the general pattern of mainstream thought lagging reality by two generations. The lefty version is seeing the world in race and class. It’s why we look at Victorian colonialism (or Churchill post WW2) in amazement at what they were thinking.

    Poništi
  2. 2. sij

    I’d also be pissed off if I went expecting one type of theater experience and got something different. (And no trailers to let you know what to expect.) Seems to me a silly stretch, mostly, to convert people’s tastes into much more than that.

    Prikaži ovu nit
    Poništi
  3. 2. sij

    Second the usual references to “outrage” at new modes of theater. Lecturer seems determined to make a mountain out of a molehill. I, for one, hate realist theater — not bcs of outrage but because it bores the hell out of me.

    Prikaži ovu nit
    Poništi
  4. 2. sij

    Discussions of art that display women as seductive monsters treated as literal misogyny. Zero appreciation that most of this (and similar in literature) is *allegory*, frustration at the aspects of one’s personality that are uncontrollable, at the inner addict.

    Prikaži ovu nit
    Poništi
  5. 2. sij

    Listening to lectures on Euro history 19th, early 20th C, and frustrated by what seems to me so much shallowness in the analysis. Two examples:

    Prikaži ovu nit
    Poništi
  6. proslijedio/la je Tweet
    19. pro 2019.

    5. Because I was raised Catholic, forgiveness/atonement are foundational values for me. I have yet to find an online social space which supports this meaningfully.

    Prikaži ovu nit
    Poništi
  7. 16. pro 2019.

    83. Tech can be an iron cage, or it can be an exoskeleton. It is MOSTLY your choice, not tech's fault, not society's fault, not even your family's fault (though it may be your friends' fault... get better friends!) as to which it is.

    Poništi
  8. 16. pro 2019.

    82. This is not new. Every previous tech (print, phone, TV, ...) also falsified many models of humanity. But then every generation pretends the past is irrelevant to their new conditions, and reinvents what are mostly the same models.

    Poništi
  9. 16. pro 2019.

    81. Tech is not a brainwashing machine, it is an alethiometer. But lots of people are very unhappy to learn that their models of humanity have been aggressively falsified...

    Poništi
  10. 16. pro 2019.

    80. MIPS did something like this (MIPS MT ASE), but by the time this idea was released MIPS was no longer interesting to anyone so the idea seems to have disappeared, barely noticed.

    Poništi
  11. 16. pro 2019.

    79. Is the optionality of SFT worth it? Not for general throughput engine tasks, I don't think so -- you still have your sea of small cores. BUT it allows for lightweight launching of coroutines. And that has potential to make some programming styles more practical.

    Poništi
  12. 16. pro 2019.

    78. So, yeah: SMT dumb -- for PERSONAL devices (ie latency devices). SFT smart -- for PERSONAL devices.

    Poništi
  13. 16. pro 2019.

    77. Does this matter? For desktops and phones? Are you kidding? You have 4, 6, 8 high performance cores and you think it's a PROBLEM that you can't run two simultaneous independent apps on one of them?

    Poništi
  14. 16. pro 2019.

    76. What's the downside? Well the threads are (by design) not virtualized. So you don't get what looks like 2x as many CPUs. If you want to run two independent apps on one CPU, you have to resort to good-old-fashioned tim slicing.

    Poništi
  15. 16. pro 2019.

    75. Bcs OS is not involved, app can spawn extremely light-weight fibers that only have to execute a few tens of instructions to have a net performance win.

    Poništi
  16. 16. pro 2019.

    74. The wins of SFT over SMT are substantial. Security is no longer an issue (threads are same process). Sharing is much less of an issue. (Likely to be much more overlap of everything from TLB entries to I-cache and D-cache entries.)

    Poništi
  17. 16. pro 2019.

    73. OK, I said SMT is silly. But the devil is in the details. I have much more sympathy for "symmetric fiber threading". This is like SMT, but the threads are not visible to the OS (apart from trivialities like register save at context switching).

    Poništi
  18. 16. pro 2019.

    72. But for generic "PC"-type CPU? Nah, seems to me a silly design point.

    Poništi
  19. 16. pro 2019.

    71. So is SMT always dumb? No. If it's a hack around legalities (like per "cpu" licensing), well, what can you do? If the CPU is designed to have very few smarts, for a task that has little cache locality (eg in-order packet processor), it also makes sense.

    Poništi
  20. 16. pro 2019.

    70. It gets worse. There are the obvious security issues. You also start hurting your latency performance because it turns out naive partitioning resources between threads gives very low throughput boost. So you start adding logic (more FO4's, hurts frequency) for smart sharing.

    Poništi
  21. 16. pro 2019.

    69. So obviously L1$, but also branch and every other predictor. SMT shrinks all those (by 2, or 4, even 8). WTF? Why give yourself less of the constrained resource so you can share the unconstrained resource? Just throw down more small cores!

    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·