Dan's Data letters #115Publication date: 5 July 2004.
Last modified 03-Dec-2011.
Could you hazard a guess as to when we will see PCI Express motherboards for the Socket 939 Athlon 64? I have heard "Fall" from one or two review sites, but, strangely enough, this question does not come up very often so I cannot be sure.
For reference, my current system is an Athlon 2500+ (Barton) and Radeon 9700 with 1Gb (2x512Mb PC2700) RAM. In the never-ending pursuit of higher resolutions and frame rates, I reckon I am in dire need of a new video card, CPU, and more RAM, but I hate to buy yet another AGP video card and AGP motherboard. When might my suffering end?
I don't know when PCIe A-64 mobos will arrive, but it won't be a very big deal when they do. Not at first, at least, since the early boards will presumably be pretty much like the early Intel LGA775 boards. Those have an x16 PCIe video card slot, and zero-to-a-few of the tiny x1 PCIe slots, and a few regular PCI slots as well. X16 PCIe is twice as fast as AGP 8X, but that doesn't mean much; AGP speed bumps have never given much of a speed improvement, and neither does this (that page also has less than complimentary things to say about stopgap PCIe boards with AGP slots as well as x16).
PCIe is a point-to-point system versus the simple shared-bandwidth PCI parallel bus, so x1 PCIe cards will in theory be faster than PCI cards, but there's really no reason to get excited about them, especially since there are bugger-all of them on the market at this point.
So, basically, early PCIe Athlon Whatever boards will, in use, be little different from PCI/AGP boards. They'll probably have those little x1 slots that you likely won't be able to use for anything, and will probably have fewer PCI slots as a result, and there'll be that really fast x16 video card slot that accepts a somewhat more expensive graphics card that ends up running only marginally faster than the cheaper AGP 8X version of the same thing. Sure, you'll be able to upgrade, and all, but you know that by the time a twice-as-fast video card and twice-as-fast CPU have come out, your mobo will be in some other way incompatible - unable to run the RAM fast enough, for instance.
Assuming money is no object, it's just about worth your while now, performance-wise, to upgrade from what you've got at the moment to the fastest A64 and new-generation ATI or Nvidia card on the market at the moment. You could confidently expect at least a 1.5X performance improvement from doing that. If, of course, money is an object, and if you're not actually whipping your current system to death, it's a lot more sensible to sit about for a while, let the Socket 939 mobos increase in number and get their bugs worked out, and then, maybe, get a PCIe board, if something decent's available.
I read recently that under new management, Intel has been increasing the clock speed of its processors at a slower rate. Does this mean Moore’s law is dead, or is total processing power still doubling every 18 months?
Moore's Law wasn't originally expressed as such; it was just Gordon Moore noting in the mid-Sixties that minimum-price chip component counts (the counts for chips that cost the least per transistor) were roughly doubling each year, and that he thought the trend would probably hang on until at least 1975 (in '75, he observed that all the trend had done was slow down a bit).
The important bit was the "minimum-price" part, though - he was saying that there would soon enough be cheap chips with non-trivial processing power, all over the place, and that's precisely what we've got today. Many of these "embedded" chips are from a long way back on the Moore's Law graph; they use mildly refined versions of architectures a decade or two old (early PowerPC, 680x0, Z80...).
Moore wasn't really talking about the most complex processors, and he wasn't originally postulating the "18 month" period for transistor density doubling. Actually, he wasn't talking about actual processing power, per se, at all; processing power does track CPU transistor count to some extent, but there are important other factors that can massively skew transistor counts. On-die cache, the old CISC versus RISC battle, et cetera.
Personally, I think it was more impressive that Moore correctly predicted the popularity of home computers and mobile phones (and automatic cars, but never mind that for now...), than that he noticed how much more complex chips were getting.
You can read Moore's original 1965 paper (with its marvelling at 50 micron component spacing, and wild-eyed projection of as many as 65000 transistors on a chip by 1975...) in PDF format here.
I've got a 100Mb Zip drive, parallel port interface (yeah, I know, Old School). I know that the disks are a bit sensitive to damage that will result in the "Click of Death". I have a couple of those laying around. But recently, ANY disk I put in the drive isn't recognized. "Click... Click..."
It was reading the disk that was in it OK, but after an eject, this started happening.
Is this a problem I can overcome, or do I have a drive I should consider "recycling" (possibly to a museum)?
If the drive went south on me, did it damage any disks I put in after the fact?
Yes, this could be the "click of death" problem, indicating a Zip drive that's screwed up and will now screw up any disk you write to in it, but it could also just be worn out disks - yes, a whole bunch of old disks could have gone stale on the shelf. It could be something weirder, too, like mildew on the disk surface; Zip cartridges aren't well sealed.
A Zip drive that can't read a disk will retract and reinsert its heads over and over to clean them (floppy drives do similar things), making that clicking noise; this is by design. So it's not necessarily the drive's fault; a Zip drive that's clicking may just have bad media, and is not necessarily an Anything Of Death. If the drive does the same thing with a brand new preformatted disk, though, then it's definitely the drive.
The "click of death" furore that caused such bad publicity for Iomega back in the day (not helped at all by their lousy customer support...) wasn't actually, in my opinion, much of an epidemic. Even assuming that everyone who thought they had Click Of Death actually did (and hadn't, say, barbecued their cartridges on the dashboard of their car in summer, or something), quite a lot of complaints can be expected for pretty much any product as popular as the Zip drive. Iomega sold tens of millions of the things, after all, and I wouldn't be surprised if there are a few hundred million cartridges out there. With that kind of market penetration, a one-in-ten-thousand failure rate can fill a football stadium with unhappy customers.
My Sydney apartment building's in a DSL "black hole", and I can't get cable broadband because they don't serve apartments, but we've now got wireless broadband provided by BigAir. They're offering one megabit per second plans (with a 1Gb download limit, nine cents per megabyte excess fee) for $AU49.90 per month (3Gb per month is $AU79), which seems quite reasonable and would actually result in a cost saving for me because I could get rid of my phone line, which I very rarely actually use for phone calls. From what I have been able to read about it, BigAir seems all OK - what do you think?
I don't trust any ISP that doesn't publish pricing details on their Web site.
BigAir's site is one big shiny brochure, with nary a dollar sign, or indeed any technical data to speak of, to be seen. I suppose this is partly because they don't sell direct to the public - a building gets a connection, then the tenants can pay to sign up - but it's still a bit odd.
Poking around on the Whirlpool forums turns up some people who are happy enough with BigAir's service - the BigAir inter-building network uses standard 802.11 wireless networking, which means it should work well unless a windstorm misaligns the antennae, or it's absolutely urinating with rain, or someone hits your building with a search radar. BigAir are too new to even be listed in the generally excellent Whirlpool Broadband Choice section, though.
BigAir's symmetric network is a very nice feature for some people - having upload speeds the same as download is a big deal for Web publishing and virtual private networking (useful for working from home), and also for file sharing, not that you're going to be doing any of that on a 1Gb monthly limit. You're sharing your bandwidth with everyone else on your "cell", and there are various other bottlenecks that mean you'll probably seldom see full speed, but it's still nice to not be limited at your end.
The nine cents per excess megabyte fee seems to be pretty standard for BigAir (on a 12 month contract; move out and you'll still be paying the monthly fee until the term expires). So if you go for the 1Gb plan, your second gigabyte (invariably defined as 1000Mb) of downloads for a month will only cost you another $AU90. But you could still end up with a nasty surprise in store if you actually use that high download speed they offer. In the unlikely event that you managed to suck down data at full speed, you could use a 1Gb monthly allowance in two hours and 20 minutes.
If you're only going to do light, occasional Web browsing and e-mail (or games, even; online action games only need something like 20 to 30 megabytes an hour) 1Gb a month is fine. It's surprising how easily you can use it up, though, even with ordinary work-from-home stuff.
As DSL and cable are out, though, this does still seem to be the best deal you're going to get in your building. And yes, it'll certainly beat dial-up.
Paying $AU50 a month for a 1Gb download allowance still sticks in my craw, though. When iiNet's next free upgrade rolls through on the 12th of this month, the second-from-the-top residential ADSL plan that I'm using at the moment, which costs $AU110 a month, will be up to 1.5 megabits per second download speed, 256 kilobits per second upload, and 48Gb per month total download allowance. 24Gb on peak, 24Gb off-peak. If you manage to exceed the limit they just reduce the connection speed to a mere twice that of a dial-up modem; you don't pay any more. That's the way civilised ISPs do it.
IiNet's $50-a-month plan's only 256k/64k speed, but even that's going to offer 24Gb monthly downloads after the 12th. Their $60-a-month plan won't have a limit at all - they'll just speed-throttle the biggest downloaders when necessary.
The Whirlpool nerds are pretty much all serious leechers like me, so they all consider BigAir to be a complete joke. If you have no interest in downloading even a lot of MP3s, though (don't even think about movies...), it'll do.
I have 2 speakers wires (both with stripes) hanging in my car boot, and have no idea which one is positive and which negative - is there any way to figure it out without a multimeter and without pulling out the head unit?
Initially I soldered the wires to my new speakers in what I thought was the correct way, but found the bass pretty poor. So then I switched the wires, but I'm not able to tell any big difference, out of phase or not. However, I would like to get it right, of course.
The only thing I can think of is to see whether the driver goes out-in, or in-out on bass hits. But then again, I'm not sure which way is correct (I'm thinking it should be out-in). Any ideas?
You can't get absolute phase right without seeing both ends of the wires, but it's possible to get the speakers relatively correct without carpet-ripping and dashboard-dismantling, which is good enough. You just want them all set up with the same phase. It doesn't matter whether that phase is "correct" or 180 degrees out (well, it matters to audiophiles, but they're badly lost if they're installing a car stereo), as long as they're all the same.
To match rear speakers to front speakers, you will of course need to at least remove a front speaker so you can see what its wiring looks like. Car audio setups often have mildly confusing wire colours, but positive and negative generally have a consistent colour/stripe relationship.
To identify the terminal polarity of a speaker that doesn't have polarity markings on it (or does have markings, but which you suspect was manufactured by a company that didn't really care...), take a 1.5 volt battery with a couple of wires on it (rough up the ends of a normal D cell with coarse sandpaper and you can solder wires right onto it - this can be a surprisingly useful object to have in one's workshop) and connect it briefly (so you don't barbecue any voice coils; the little voice coils in car-speaker coaxial midrange and treble drivers can be quite fragile) to the speaker terminals. If the positive terminal of the battery's connected to the positive terminal of the driver, the cone will move outwards. If it moves inwards, you've got battery positive connected to speaker negative.
We are about to stop operating a rather old mass spectrometer at our labs, which has been controlled from a Digital DECstation 5000/25 running custom software on Ultrix. This obsolete machine interfaces with its obsolete storage media (an old HP C1536A external LynxDAT tape drive) via an interface with a cable that has 25-by-2 pins on either side of a notch, and has "H8574-A" printed on it. Google tells me this is SCSI.
Now that we have a new GCMS which interfaces with a PC we'd like to be able to access our old files on those tapes on the PC. Do you know of a (preferably cheap) way that we might be able to convert that cable to serial/parallel/USB, etc, and have a PC talk to that old tape drive? Or do I need a SCSI card in the PC? Will our PC even be able to read those Ultrix files?
Based on the seven seconds of research I just did, I think you ought to be able to get the files off the tapes, though you'll need someone with basic familiarity with the system to do it. Yes, the tape drive should be a SCSI-2 device, but a device driver for it for modern OSes will probably be hilariously hard to find. Fortunately, the 5000/25's spec sheet says the machine also has built-in Ethernet. If it's only 10Base2 (which was going out of fashion in the PC world when I put my old networking page up, four years ago...), but that's not a big deal; if you don't have a suitably equipped PC handy, a PCI 10BaseT/2 NE2000-compatible card for a PC should work without extra drivers on all current PC OSes, and cost you ten bucks, tops. A couple of T-pieces and terminators, a length of cable, and you're connected; TCP/IP config shouldn't be too painful (says the guy who isn't going to have to do it).
And, as it turns out, it's probably 10BaseT anyway; that's what's on the 5000/25 sitting in the basement of the reader who just e-mailed me.
Anyway, once you've got a TCP/IP connection to a modern computer, you should be able to shift the files via FTP. Swapping tapes for X hours to transfer all of the data to the newer system sounds, to me, like a great job for the work experience kid.
At least the price is reasonable, at €59!
Like various other audiophile swindles, this one requires you to wait some time (20 days!) before the full sonic character of the tweak is revealed. Then you've got cured lacquer all over your semiconductors and would have to scrape it off again to get back to the initial state, and re-paint and wait another 20 days to re-compare. Only if you've got two identical amplifiers (or whatever) to test with could you do a proper A/B comparison - and then, of course, Altmann would probably insist that no two components sound alike in the first place.
The 59 Euros the lacquer costs is, as you say, not much to pay compared with cables-that-cost-as-much-as-a-car. I venture the opinion that the value customers are getting for their money is, however, exactly the same.