Monday, 2 February 2004
9:29 - I was wrong again. In the third edition of PC Hardware in a Nutshell, I speculated that Intel would call their Prescott-core processor the Pentium 5 (or the Pentium V, which I prefer). Intel famously does not comment on unannounced products, but in this case one of my Intel contacts was good enough to warn me off gently. I sent my chapter draft to him for factual corrections. After he read the section of the draft that mentioned the "Pentium 5", he said "I'm not sure I'd say that...". I said it anyway, and it ended up being wrong. Today Intel introduced their Prescott-core processor, and they still call it the Pentium 4.
A lot of web sites jumped the gun on the NDA, which was 12 noon PST yesterday. Many of them damned the Prescott with faint praise, claiming that it's slower than the Northwood it replaces. As it happens, I have three 3.2 GHz Pentium 4 processor samples--Northwood, Prescott, and Gallatin--and so was able to test for myself using my own top-secret benchmark suite. As it turns out, I was wrong yet again.
Three weeks ago, I speculated that Prescott would be slower than Northwood, with a 2.8 GHz Prescott perhaps matching the overall performance of a 2.6 GHz Northwood. I expected that difference because Prescott extends the already-long 20-stage Northwood pipeline to 31-stages. A longer pipeline is a two-edged sword. On the one hand, it allows Intel to increase the core clock speed. On the other hand, a longer pipeline means slower performance for a given core speed. With a very long pipeline, pipeline stalls become more likely and more catastrophic in terms of impact on overall performance. All other things being equal, a processor with a 31-stage pipeline should be noticeably slower than an equivalent processor with only 20 stages.
That turns out not to be the case with Prescott, because all other things are not equal. Intel has obviously made some significant improvements in the branch prediction unit, whose job is to minimize the likelihood of pipeline stalls. Another major factor is that Intel boosted L2 cache from 512 KB on Northwood to 1 MB on Prescott. The result is that Prescott and Northwood provide essentially identical performance overall at the same clock speed.
That's not to say that individual benchmarks are identical, because they're not. The older Northwood-core Pentium 4 wins about a third of the benchmarks we ran by a few percent, but the new Prescott-core Pentium 4 wins a similar number of benchmarks by similar percentages, particularly those related to content creation, video processing, and so on. The remaining third of the benchmarks we ran were a dead heat.
The Gallatin-core Pentium 4 Extreme Edition is a Xeon wolf in Pentium 4 sheep's clothing. In effect, Gallatin is a Northwood-core Pentium 4 with 2 MB of L3 cache added, giving it a total of 2.5 MB of secondary cache. The Pentium 4 Extreme Edition blows the doors off both Northwood and Prescott, as it should for what it costs. Not many people other than hard-core gamers will buy the $1,000 Extreme Edition processor, though, so the real race is between Northwood and Prescott.
We think Prescott is the certain winner. Although Intel has made significant architectural changes with Prescott, including the move to a 90-nanometer process, the price and performance of Prescott matches Northwood so closely that in effect Intel is simply slipstreaming Prescott into their Pentium 4 lineup. Intel plans to ramp Prescott very quickly, so Northwood is likely to disappear soon, except in the Gallatin/EE variant, which is likely to remain Intel's fastest desktop processor for the remainder of the year.
Prescott also includes SSE3, codenamed PNI (Prescott New Instructions). For now, there isn't much software that takes advantage of SSE3. That will change, however, and when it does Prescott will run those SSE3-enabled applications noticeably faster than Northwood. With a major platform change to Socket 775 imminent, you can't really say that Prescott is a future-proof choice, but its support of SSE3 makes Prescott, in our opinion, the best choice in a Socket 478 processor.
We've gotten a surprisingly large number of emails from people asking if they can or should run out and upgrade to Prescott.
As to "can", it depends on the motherboard. Some Intel and third-party Socket 478 motherboards support Prescott, although they need a BIOS upgrade to accommodate the changes in L2 cache and so on. Other Socket 478 motherboards do not support Prescott, usually because their VRMs (voltage-regulator modules) are insufficient to support Prescott's higher current draw.
As to "should", the answer is probably not. If you're already running a fast Northwood P4 in a late model motherboard, upgrading to Prescott doesn't buy you much. If you're running a slower Northwood P4 in an older motherboard, chances are good that your motherboard doesn't support Prescott.
But if you're building a new system, no doubt about it. Use a Prescott.
11:05 - The New York Times has posted a fascinating William Safire column that details an immensely successful disinformation campaign waged by the US against the Soviets. Unable by their own efforts to keep up with US technology, the Soviets regularly stole technology from us. This article tells the story of a very bright guy who used the principles of aikido to let the Soviets do themselves in. It's well worth your time to read.
Tuesday, 3 February 2004
9:20 - Our Internet connection went down yesterday morning, just I was about to publish the entry I'd made at 11:05. I fiddled around a bit to make sure the problem wasn't on our end, and then called Time-Warner Roadrunner tech support. After waiting on hold for 15 minutes, I was connected to a woman who told me there'd been no report of problems in the area and that they'd have someone out to make an on-site service call between 3:00 p.m. and 5:00 p.m. tomorrow (today). I told her that seemed a bit long to wait, and she said it was the earliest slot they had available. She also said that if the problem solved itself to give her a call back.
I power reset the cable modem several times over the next few hours. Finally, around 2:30 p.m., the cable light came back on. I called Time-Warner (why I even considered spending another 15 minutes on hold to let them know that their problem had fixed itself, I don't know). I got a busy signal, and the phone company kindly offered to keep trying for only 75 cents. Right.
Just as I hung up, I noticed that the cable modem was down again, so it's fortunate I didn't tell them it was working again. Throughout the afternoon, the cable modem was up and down. Sometimes it was up for minutes at a time, and others only a few seconds. Finally, late in the afternoon, it came up and stayed up. After what it had been doing all day long, I wasn't about to call and tell them it was fixed. After dinner, the phone rang. It was a guy from Time-Warner Roadrunner, telling me the problem was fixed and I should be live. He said he was looking at my connection as we spoke and it showed up as live. I checked to verify and I was indeed still live.
When I asked him what had happened, he said there'd been an area problem. I told him that the lady I'd spoken to had specifically told me there was no area problem and that it was specific to me. I wish they'd get their stories straight. It isn't as though they can't look and see for themselves whether it's just me down or a group of people. She could easily have checked any number of our neighbors who have Roadrunner to find out. I hate it when people lie to me.
What's really annoying is that I suspect they ignore individual problem reports. In fact, I know they do. I've been told by people in a position to know that most cable modem companies require a certain number of problem reports, usually three to five, before they'll even bother to look into the problem. That affects me more than most, because I work at home. More than once, I've been down for hours, and I suspect the reason it took so long to repair the problem was that Time-Warner waited until other people got home, had dinner, finally found out their service was down, and reported it.
I'd think about changing to DSL, but I'm afraid I'd be jumping from the fire into the frying pan.
The dogs went berserk a few minutes ago, shouting that there was an intruder on the property. They bark so loudly that our doorbell literally rings from sympathetic vibration. I opened the front door, looked out, saw nothing, and told them they were nuts. Then I headed for the kitchen to get some Coke. As I was returning to my office, I happened to glance out the back window and saw a large FedEx truck sitting in the driveway behind our house. I have most of their drivers trained to deliver at the front door, but, for some reason, some of their drivers insist on dropping stuff at the back door.
I opened the back door just as the driver jumped into her truck and fired it up. I was hoping whatever she'd delivered was small and light. Oh, joy. It was an Intel server chassis and motherboard. Sometimes I think FedEx delivers heavy stuff to the back door because they only have to haul it 15 feet instead of the whole length of the front walk. So I hoicked the server chassis up the stairs and then went back for the box with the motherboard and processor. That, at least, was small and light.
9:50 - I've gotten several email from several people who have recent Intel motherboards and aren't sure if they'll support the newer Prescott and Extreme Edition Pentium 4 processors. Check this page to determine whether your Intel motherboard supports Prescott- and Gallatin-core processors.
Wednesday, 4 February 2004
8:58 - Heads-down writing the rest of the week. I need to get some systems built, too, but I probably won't get to that at least until the weekend.
Things are starting to stack up around here. When Barbara arrived home yesterday, the Intel server chassis was sitting in the hallway near the foyer, still in its box. "What's that?", she asked. "An Intel server chassis," I replied. "Oh, okay," Barbara said. I'm always tempted to say something outrageous, like "an IBM S/390 mainframe". She'd accept that with equanimity, no doubt.
Twenty-five years ago, when my late friend Mel Tappan was writing Survival Guns, he mentioned more than once that every horizontal surface in the house was covered with guns, ammunition, and accessories. One time, Nancy opened the linen closet and found something lying on top of the towels she wanted. "What's this?", Nancy asked Mel. "A MAC-10 submachine gun," Mel replied. "Oh, okay," Nancy said. Knowing Mel, I'm sure it was.
Here's the best reason yet not to buy a TiVo. Why would I buy a product that allows the manufacturer to keep track not just of what programs I watch, but the details about how I watch them? If the copyright pigs continue to buy legislation as they have been doing, I can envision a day arriving when TiVo users receive a retroactive bill for ten years of skipped commercials. You might think that's ridiculous, but then you probably wouldn't have believed something like the DMCA could be passed, either.
It's bad enough to have Big Brother keeping track of your TV viewing habits, but having to pay every month for the privilege is truly outrageous. Just Say No to TiVo. How's that for a slogan?
Thursday, 5 February 2004
8:58 - Boy, I hate this:
Tracing route to ttgnet.com [188.8.131.52]
over a maximum of 30 hops:
1 <10 ms <10 ms <10 ms [IP redacted; RBT local router]
2 <10 ms 15 ms <10 ms 10.32.240.1
3 * * * Request timed out.
4 <10 ms 15 ms <10 ms srp10-0.gnboncsg-rtr1.triad.rr.com [184.108.40.206]
5 * * * Request timed out.
6 * * <10 ms son0-0-2.chrlncsa-rtr6.carolina.rr.com [220.127.116.11]
7 <10 ms 16 ms 15 ms pop1-cha-P0-3.atdn.net [18.104.22.168]
8 * <10 ms 15 ms bb2-cha-P0-2.atdn.net [22.214.171.124]
9 16 ms 15 ms 16 ms bb2-atm-P6-0.atdn.net [126.96.36.199]
10 16 ms 31 ms 15 ms pop1-atm-P1-0.atdn.net [188.8.131.52]
11 16 ms * 15 ms pos8-0.er1.atl4.us.above.net [184.108.40.206]
12 31 ms 31 ms 47 ms so-3-0-0.mpr1.iah1.us.above.net [220.127.116.11]
13 32 ms 47 ms 31 ms 18.104.22.168.ev1.net [22.214.171.124]
14 47 ms 47 ms 62 ms ivhou-207-218-245-35.ev1.net [126.96.36.199]
15 46 ms 63 ms 47 ms rocket.mazin.net [188.8.131.52]
There's not even any point to calling Roadrunner tech support. I'd end up talking to some clueless first-level person who'd want me to unplug my cable modem and would never have heard of a traceroute. I'd waste an hour on the phone with no results. They'll fix it when they fix it, I guess. I have no idea when I'll be able to publish this. Not that it matters much, because I'm still doing heads-down writing. Which is a lot harder when I can't get to the Internet to look things up.
8:58 - I called Roadrunner tech support finally, but it was futile as I knew it would be. I spoke with a polite young lady in Virginia, but I might as well have been connected to someone in India. I tried to explain to her in simple terms that my problem wasn't the connection between my cable modem and Roadrunner, but something in their internal network. I explained that using traceroute and ping showed real problems on their end, and that the packet loss was upwards of 90%.
In response to that, she read what must have been the first question on her list: "Are you using Windows or a Mac?", to which I replied, "No." After a slight pause, I took pity on her and told her I was running Linux. "Oh," she said, "we can't support you if you're running Linux." I explained again to her that I didn't need support for my local connection, that I needed Roadrunner to fix the problem in their internal network. "Well," she said, "I can't even refer you to someone else unless you're using Windows or a Mac." Jesus.
I'm following the case of the little girl in Florida who was snatched. They've caught the guy who almost certainly did it, but he's not talking. That means it's very likely although not certain that the little girl is dead. She may be imprisoned somewhere, dying as I write this.
If it were me, I'd do whatever was necessary to find out. I'd strap the guy down to a solid table and wire up his testicles to an electrical generator. I'd soon find out where he'd put her, and I wouldn't spare any sympathy for his suffering. I have no moral problem with using torture under such circumstances. A little girl's life may be at stake, and the man in custody is almost certainly the perpetrator. Her life matters. His comfort doesn't.
The problems with that, of course, are (a) the court would immediately throw out the evidence obtained by torture (not to mention punishing the torturers), and (b) the guy may not have done it.
As to (a), it seems to me that in such circumstances the courts should give the police more flexibility. If they find evidence in such circumstances by using torture or other illegal means, that evidence should be admissible. If the police correctly suspect someone and evidence obtained by illegal means confirms their suspicions, the guy shouldn't walk. If it turns out the guy did it, that should be the only thing that matters.
But what if he didn't do it? We want to discourage the police from torturing innocent people. So, as to (b), I suggest asking for volunteers. There shouldn't be any lack of them. Cops have daughters, nieces, and grandchildren, too. Ask for 100 volunteers who (1) agree that if the guy turns out to be innocent, they will pay him $10,000 out of their own pockets, and (2) that one of the 100 volunteers will be selected by lot to undergo the same torture as was used on the suspect. If the cops were as sure as they probably are in this case, there wouldn't be a lack of volunteers because they wouldn't be taking much risk. If they're 99.9% certain the suspect did it, for example, the expected value of any one volunteer's risk is $10, and his probability of having to undergo torture would be 1 in 100,000. Pretty good odds.
If the suspect was indeed innocent, he should have the choice of accepting the $1,000,000 in exchange for not pressing charges, or of foregoing the $1,000,000 and taking his chances in criminal court by filing assault and battery charges and in civil court by claiming damages. I suspect what few innocent suspects there'd be would mostly choose the former.
Torture is distasteful, certainly, but less so than giving up on the little girl or, alternatively, striking some sort of deal with the probable criminal.
Friday, 6 February 2004
Saturday, 7 February 2004
Sunday, 8 February 2004
Copyright © 1998, 1999, 2000, 2001, 2002, 2003, 2004 by Robert Bruce Thompson. All Rights Reserved.