• VinDuv (disco)
    Comment held for moderation.
  • RaceProUK (disco) in reply to VinDuv
    VinDuv:
    The main website seems down
    :sigh: Do we need to have servercooties.com monitor the main site as well now? :laughing:
  • aliceif (disco) in reply to VinDuv
    Comment held for moderation.
  • Boner (disco)

    CS Student knows more than employed computer "expert" shocker.

  • aliceif (disco) in reply to VinDuv
    Comment held for moderation.
  • boomzilla (disco)
    Comment held for moderation.
  • JBert (disco) in reply to boomzilla
    Comment held for moderation.
  • accalia (disco) in reply to RaceProUK
    RaceProUK:
    Do we need to have servercooties.com monitor the main site as well now?

    it can. ;-)

    want it to?

  • boomzilla (disco) in reply to accalia
    accalia:
    want it to?

    I doubt it. This is the first time I can recall this happening lately. I mean...previously we've had disconnects like not linking to comments or the sidebar feed being broken, but not a 500 on the whole page.

  • wft (disco)

    On actual article: that's why you want your server room to be supervised by someone with at least high school thermodynamics knowledge.

    I remember reading an article on how a supercomputer in Switzerland was reportedly plugged into the central water heating system, so that it would be cooled from incoming water and the heat it generates would be harvested and passed on to residential buildings in the neighbourhood.

    The incoming water was to be at something like 60°C, and the supercomputer's circuits would heat it to something like 61 or 62. I can almost hear Bob exclaim, "It's too damn hot, how could it possibly work?!"

    Yes, it's damn hot when you take a shower or try to drink a coffee at that temperature. And it's just chilling for a computer running at 70-80°C, if not hotter.

  • RFoxmich (disco)

    Oooh ooh wait I know -- the IT expert was the boss's nephew right?

  • flabdablet (disco)

    Scott installed a lock-box over the thermostat and told Bob that maintenance had done it.

    I like the cut of his jib.

  • flabdablet (disco)

    TRWTF: dedicated aircon for a room containing just two servers, attached by a doorway to a large space regularly occupied by humans. A simple exhaust fan and a vent drawing in air from the main shop floor would have been more than sufficient.

  • PWolff (disco) in reply to flabdablet
    flabdablet:
    TRWTF: dedicated aircon for a room containing just two servers, attached by a doorway to a large space regularly occupied by humans. A simple exhaust fan and a vent drawing in air from the main shop floor would have been more than sufficient.

    This could be easily explained by this room having been the boss' office before.

    Oh wait, that wouldn't diminish the WTFiness a bit.

  • dkf (disco) in reply to PWolff
    PWolff:
    This could be easily explained by this room having been the boss' office before.

    Or by the room being an ordinary office before being partitioned, with the AC in the server part.

  • HardwareGeek (disco)

    The small server closet, designed and installed by Bob, was supposed to be entirely redundant.

    Bob should have been made redundant.

  • CoyneTheDup (disco)

    A friend of mine once told to me what an "expert" is.

    "An 'ex' is a has-been," he explained, "And a 'spurt' is a drip under pressure."

  • HardwareGeek (disco) in reply to CoyneTheDup
    CoyneTheDup:
    A friend of mine once told to me what an "expert" is.
    My pappy used to tell me that 'fore you was even a twinkle in yo' mammy's eye. Now git offa my lawn!
  • DJSpudplucker (disco)

    TRWTF: allowing store personnel to mess with server hardware in any way. I used to work for a now-defunct big square box hardware chain and we often sent IT teams out to stores for upgrades, repairs, etc. We would not let some clowns design and/install their own server closet, even though these were dual unkillable PS/2 machines.

  • DocMonster (disco)

    TRWTF is that the "expert" wasn't let go after six screwups of the same thing after being told not to do it and that it was causing problems.

  • kupfernigk (disco) in reply to wft
    wft:
    I remember reading an article on how a supercomputer in Switzerland was reportedly plugged into the central water heating system, so that it would be cooled from incoming water and the heat it generates would be harvested and passed on to residential buildings in the neighbourhood.

    The numbers sound slightly off. My condensing heater system supplies water at 55C, and at full power on return the temperature is about 50C. 50-55C is fine for water cooling most CPUs, but 60C seems a little high. Perhaps someone can correct me on this? Unless it was an ecl supercomputer, where the maximum die temperature is around 85C, and a consistent operating temperature is more important than the absolute value (ecl is very sensitive to gate delay which is also sensitive to die temperature, so although the working temperature range is quite high, the optimum operating temperature window for a given clock rate can be quite small.)

  • wft (disco) in reply to kupfernigk

    Maybe my numbers are not exactly like this. I just remember that incoming "cool" water was too hot by human body standards, and that it contributed something like 1 to 3 degrees to the heating system.

    Which the likes of Bob would never understand.

  • HardwareGeek (disco) in reply to kupfernigk
    kupfernigk:
    ecl supercomputer

    What, a Cray-2 from 1985?

    Filed under: An iPad 2 has the same LINPACK benchmark results as a Cray-2

  • FrostCat (disco) in reply to wft
    wft:
    On actual article: that's why you want your server room to be supervised by someone with at least high school thermodynamics knowledge.

    7 times, Bob managed to do that? I find that unlikely, as someone who's managed to accidentally freeze his AC by forgetting to change the filter. Spending a couple of hours with it 95° inside while you wait for the ice to melt focuses your mind.

  • superjer (disco) in reply to wft
    wft:
    60°C

    Yes, it's damn hot when you take a shower or try to drink a coffee at that temperature.

    ???

  • Dragnslcr (disco) in reply to HardwareGeek
    HardwareGeek:
    >The small server closet, designed and installed by Bob, was supposed to be entirely redundant.

    Bob should have been made redundant.

    Er, he was. That's why Scott was there. It's just that in this case, the backup was better than the primary.

  • HardwareGeek (disco) in reply to Dragnslcr

    #re·dun·dant /rəˈdəndənt/ adjective adjective: redundant BRITISH (of a person) no longer employed because there is no more work available. "eight permanent staff were made redundant"

  • dkf (disco) in reply to kupfernigk
    kupfernigk:
    My condensing heater system supplies water at 55C, and at full power on return the temperature is about 50C. 50-55C is fine for water cooling most CPUs, but 60C seems a little high. Perhaps someone can correct me on this?

    Supercomputers run rather hotter than most computers, as their design (and the design of the code they run) is intended to get absolutely as much speed as possible. They also tend to have fewer safety cutouts (e.g., thermal limiters) in place than most; everything is sacrificed for speed. If your code isn't CPU bound (after tuning for what is probably a weird parallel architecture) GTFO.

  • aliceif (disco) in reply to superjer

    Celsius.


    Filed under: Why can't we all just use Kelvin?

  • Buddy (disco) in reply to aliceif

    Still, 60C is some mild ass coffee.

  • aliceif (disco) in reply to Buddy
    Comment held for moderation.
  • Buddy (disco) in reply to aliceif
    Comment held for moderation.
  • Watson (disco) in reply to dkf

    I don't think I've seen any commercial building more than five years old in which the floorspace is still being used in the manner assumed when the aircon was designed and installed.

  • Watson (disco) in reply to CoyneTheDup

    I was told that x is an unknown quantity.

  • Vault_Dweller (disco) in reply to HardwareGeek
    HardwareGeek:
    BRITISH

    Well, there's your problem

  • kupfernigk (disco) in reply to HardwareGeek
    Comment held for moderation.
  • HardwareGeek (disco) in reply to kupfernigk
    kupfernigk:
    the Cray wasn't the only ECL supercomputer

    I didn't say it was; however, it was the most recent one I found in a few minutes of searching. Its immediate successor as "world's fastest computer" was CMOS. It's likely there were a few more supercomputers built using ECL, but as CMOS speed has continued to follow Moore's Law, ECL became obsolete pretty quickly.

    kupfernigk:
    He didn't say how old the system was
    You're right. My brain failed to make the context switch from Bob's server closet to the Swiss supercomputer of unknown vintage.
  • FrostCat (disco) in reply to kupfernigk
    kupfernigk:
    If you were able to design a program that accomplished useful work and kept an iPad compute bound, I doubt it would last very long before overheating caused it to slow down dramatically.

    Conveniently, there's sites out there that have confirmed this: [image]

    Ouch.

  • kupfernigk (disco) in reply to FrostCat
    FrostCat:
    Ouch.

    It's actually doing rather better than I expected, but then we don't know what the actual CPU/GPU utilization is; it could be a lot less than 100% overall.

  • kupfernigk (disco) in reply to HardwareGeek
    HardwareGeek:
    It's likely there were a few more supercomputers built using ECL, but as CMOS speed has continued to follow Moore's Law

    Your comment is fair enough except for this; Moore's Law is about transistor density, not speed. CMOS hasn't really got faster for quite a long time, with 2-3GHz being the typical range of commercial CPUs. However, die shrinkage has increased component density more or less in line with Moore's D= 2^(t/18), where t is months. At the same time, various other improvements have maintained clock speed with reduced power. It's the reduced power of CMOS that made it an ECL killer, and ultimately made first laptops and then mobile phones feasible.

  • FrostCat (disco) in reply to kupfernigk
    kupfernigk:
    It's actually doing rather better than I expected, but then we don't know what the actual CPU/GPU utilization is; it could be a lot less than 100% overall.

    I wasn't prepared for any kind of detail, just the shape of the curve.

    I mean, the name of the benchmark's there in the title so you could look it up if you want.

    The funny thing is that if you read the article on Anandtech I swiped the picture from, before they released an update, the chip didn't throttle hardly at all, and apparently the back of the phone was incandescently hot. :smile:

  • HardwareGeek (disco) in reply to kupfernigk
    kupfernigk:
    CMOS hasn't really got faster for quite a long time, with 2-3GHz being the typical range of commercial CPUs.
    Clock speeds have plateaued at 2-3GHz, but as transistors have gotten smaller, they have also gotten faster (partially offset by lowering the operating voltage for reduced power). It also depends on your definition of "long time." When I started1, gate delays were speced in ns; now they're in ps.

    1 Yeah, that's a "long time." Shaddup, and get off my lawn.

  • dkf (disco) in reply to HardwareGeek
    HardwareGeek:
    Clock speeds have plateaued at 2-3GHz

    There are good reasons for that. In particular, it's getting to the point where clock propagation across the chip is a real issue (because of the local speed of light) and you're also starting to hit a major wall with electromagnetic emissions.

  • HardwareGeek (disco) in reply to dkf
    dkf:
    the local speed of light

    As opposed to the non-local speed of light? ;P

    Filed under: Light travels faster on HSR, but we don't have any in the US.

  • abarker (disco) in reply to HardwareGeek
    Comment held for moderation.
  • kupfernigk (disco) in reply to HardwareGeek
    HardwareGeek:
    As opposed to the non-local speed of light?

    The propagation velocity of an EM wave in a conductor is lower than c in free space - from work I did with delay lines in the early 80s, I vaguely recall it's around 210^8 m/s rather than 310^8 if you are using a 50 ohm coax cable, it's a bit different again in circuit boards and on ICs.

    abarker:
    Some of those have been available for a couple years. Please exit your time pod carefully.
    Which is why I was careful to qualify, as we were actually discussing mobile phone CPUs in the context of Cray 2 versus iPad. Yes, I know there are CPUs up to 6GHz and beyond, but they don't tend to run Android.
    FrostCat:
    I mean, the name of the benchmark's there in the title so you could look it up if you want.
    I did, and their website wasn't at all helpful so I gave up, as I wasn't *that* interested.
  • abarker (disco) in reply to kupfernigk
    kupfernigk:
    Which is why I was careful to qualify, as we were actually discussing mobile phone CPUs in the context of Cray 2 versus iPad. Yes, I know there are CPUs up to 6GHz and beyond, but they don't tend to run Android.

    You think I'm going to read that far back in the convo? I got what I deemed sufficient context for my pendantry, and I pounced on it.

    Now that I have been informed of the full context, let me try again:

    dkf:
    HardwareGeek:
    Clock speeds have plateaued at 2-3GHz
    There are good reasons for that. In particular, it's getting to the point where clock propagation across the chip is a real issue (because of the local speed of light) and you're also starting to hit a major wall with electromagnetic emissions.

    Actually, it has more to do with thermodynamics. With a desktop or server, there is more space to provide active cooling mechanisms, so you are able to get clock speeds significantly higher than in tablets and phones. Because everything is about space in a mobile device, those systems must rely on passive cooling, so the speeds must be throttled down.

    The examples I provided earlier are a very clear indication that the local speed of light (aka, the EM propagation velocity) is not the constraining factor when dealing with mobile platform clock speed.

  • powerlord (disco) in reply to CoyneTheDup

    Did your friend ever explain politics to you?

    poly = many ticks = blood-sucking creatures

  • HardwareGeek (disco) in reply to kupfernigk
    kupfernigk:
    The propagation velocity of an EM wave in a conductor is lower than c in free space - from work I did with delay lines in the early 80s, I vaguely recall it's around 2*10^8 m/s rather than 3*10^8 if you are using a 50 ohm coax cable, it's a bit different again in circuit boards and on ICs.

    That's about right for coax, though it depends somewhat on the dielectric (and maybe other things; my knowledge comes from ham radio, and it's been almost that long since I did any antenna phasing sort of stuff). And not just conductors; index of refraction is the ratio of the speed of light in a vacuum to the speed in a transparent medium. Light travels at just under 3E8 in air (n=1.000293), about 2E8 m/s in water (n=1.3330), 1.7E8 to 1.97E8 in glass (n=1.52 - 1.75, typ.), and 1.24E8 in diamond (n=2.417). (Some materials can have fractional, or even negative, indices of refraction, but I haven't had enough caffeine yet this morning to wrap my head around what this means.)

  • kupfernigk (disco) in reply to abarker
    abarker:
    I got what I deemed sufficient context for my pendantry,

    You make bling out of clock speeds?

    abarker:
    Actually, it has more to do with thermodynamics. With a desktop or server, there is more space to provide active cooling mechanisms, so you are able to get clock speeds significantly higher than in tablets and phones. Because everything is about space in a mobile device, those systems must rely on passive cooling, so the speeds must be throttled down.

    That was where we came in. This thread has now become circular. (Though I'm not sure where thermodynamics comes in, it's just the physics of heat removal.)

Leave a comment on “The Big Box Hot Box”

Log In or post as a guest

Replying to comment #:

« Return to Article