The real strength of the book is a well-balanced and nuanced exploration of the tension between scientific and faith-based worldviews, wrapped up in a fictional shell. Many of the viewpoints on this topic found in Western culture and thought over the last 500 years are re-hashed out in this post-apocalyptic setting, and the story uses the setting to explore flaws and strengths in the thinking of religious people, scientists, and government figures. All this while remaining readable fun fiction.
Assessed just as fiction, it's a nice treatment of a few what-ifs. What happens to civilization after a global nuclear war? What happens to religion? What happens to science? It presents a plausible and thought-provoking scenario.
Skeptics who find the faith/science tension intriguing or thinking Christians, either of whom like fiction will probably find this book very enjoyable. It was my favorite discovery out of several college lit classes.
--LP
One small correction; DVD obsolescence
on
A 140GB CD-ROM?
·
· Score: 1
The rotating disks don't actually get 1 gigabyte per second read rates, it turns out that only the card media does.
And as commentary at Tasty Bits From the Technology Front points out, the most outstanding claim about FMD drives is the 1 gigabyte per second read rates, a full 200x faster than a 32x CD-ROM, and 40x faster than a 10,000 RPM hard disk. In comparison, capacity only improves 25x over the 5.2 GB DVDs.
Personally, I find the large capacities C-3D demonstrates just reinforce my perception that buying into DVD technology is just setting yourself up for obsolescence once higher-res HDTV versions of videos and movies become available on post-DVD media like C3D's in a few years (probably more securely next time though!;-)
To clarify: we agree that Kosovo was not a UN action, it was a US/NATO one, against a non-NATO nation.
I'm concerned that your argument is based on a false premise- that you can keep all the rights you have right now AND have your government involved with all these para-national organizations as currently constituted.
I'm not talking about replacing the US govt with these para-national bodies, obviously such bodies will augment and/or replace pieces of US govt function, just as national authority first augmented and partially replaced colonial state authority. My point is that even in such a partial scenario, you and I are fundamentally losing rights unless the para-national organization is properly constructed. Take for example your premise that you have the right to know what your government is up to. I'd agree. But obviously if the WTO sets some policy in a secret meeting that your government must adhere to, then your ability to know "what your government is up to" has been reduced. If part of your monetary policy is set in Brussels rather than by officials you elect, you fundamentally lose some of the power you previously wielded with your voting rights which have shrunk ever so slightly. You say this is OK "as long as I retain the rights I currently have," but my point is that once these situations occur, you have lost some of your rights and abilities to remedy the situation.
You can't have your cake and eat it too; unless you and I want to lose bits and pieces of our current rights, we have to push our elected officials to increase the transparency, accountability, and democratic process of these para-national organizations they are delegating their (and our) authority to.
I agree with and appreciate your observations on the hype over the protest. But I have a few bones to pick with the some of your other comments.
You seem rather quick to discard a principle (sovereignty) that has been around since the Treaty of Westphalia in 1648, underpinning the basis for relationships between nations at first in Europe and now around the world.
"I don't interfere with the way you run your country (unless you give me permission first)." It's a great principle, based in part on the golden rule (after all, I don't want your country interfering in our political process/internal affairs either.)
The EU and WTO technically don't interfere with sovereignity, since both countries agreed to the terms of those agreements up front. U.N. intervention in a country is more suspect, since U.N. membership is not predicated on such a contract. The U.S. invading Kosovo, IMHO, is a good example of violating another country's sovereignity (bombing a country that is killing a few thousand of its citizens.) There are various arguments about what heinousness of activity or what degree of multi-lateral assent from other countries or bodies like the U.N. justifies violating another country's sovereignity. But that's a whole other debate.
I care about my rights, and as long as they're preserved, why should I care whose running the show?
Because your rights aren't necessarily preserved. Can you vote rulers/officers of these para-national bodies out of office? Are the actions taken by such a body (actions that affect you) open to public, open review? Is there a legal appeals process and what are your rights under that process as a citizen? And beyond the protection of your rights under such a system, on a pragmatic basis, is power adequately distributed via a system of checks and balances to prevent or ameliorate corruption or tyranny?
These are all grave concerns that many people have with the multi-national bodies that have sprung up within the last 50 years. Sure, the UN/EU/WTO/etc have valuable uses for our global society; that's why they exist. However, many of the bodies appear to be less democratic than the nation-states that populate them. The concern is (as with software licensing), who controls the code?
Try the Amazon.com "Wish List" feature; it lets you keep track of books you want to buy "later", a list Amazon keeps track of. Very cool. You can tell Amazon to make that list public or private to people who know your email address, or tell Amazon to email your wishlist to someone else.
- Better than a Gift Certificate, since you get the book Christmas morning, less hassle - Better than getting someone else's book pick, at least on average - Easier than keeping track of some easily-lost self-created book list - Good for keeping track of books you want to buy for yourself, but don't want to binge on at the moment
I get a tad nervous about the lock-in this feature presents vis-a-vis Amazon, but it's too handy to turn down; I'd love to see it imitated elsewhere.
--LP, no relations to Amazon, just glad to see a feature I've wanted for the last year finally implemented
P.S. If you want to make it less obvious that you're giving a gift, try getting and wrapping up an old CD jewel case and putting a note in it telling the person where to find their *real* present. I've had fun with this and variants of this...
Too bad it'll never get moderated up at this point, due to the day-late nature of my post. One of the major flaws with the slashdot approach IMHO.
The only way I can think of to solve it would be to force moderators to read a certain set of posts that haven't been seen by other moderators (somewhat like the meta-moderation process.) I admit this would add an additional small discouragement factor to moderators.
--LP
Re:I don't think so, but an illuminating question.
on
New ATi 3D Chip
·
· Score: 1
Your description of graphics processors, including "XORs, fills, tight loops oriented towards shading" describes the bottlenecks in 2D graphics chips well, but does not describe the problems one would need to optimize in a 3D graphics chip. The 3D bottlenecks are not amenable to a Transmeta-style solution, IMHO.
I've done some research on the 3D graphics chips of UNIX workstations and PCs over the last five years and Transmeta's approach doesn't make any sense as a peak-performance graphics chip. While I must apologize to any real 3D chip designers out there for my generalizations and any misconceptions they may spot with their even greater experience, I'll try to summarize why programmable chips don't make sense to accomplish fast 3D:
1) If the algorithm a chip must execute is fixed (as it generally is with 3D algorithms), nothing is faster than a well-designed hard-coded ASIC that lays down precisely the circuits needed for optimally accelerating that algorithm. 2) If the algorithm varies substantially, a general purpose processor is more useful. In cases where there is significant unpredictable branching in the algorithm, a general purpose CPU will be optimal. In cases where there is strong data parallelism, a DSP will be optimal. 3) Run-time programmable logic, such as FPGAs (field programmable gate arrays), microcode, or Transmeta-style programmable logic, has traditionally been best for cases where 3A) you want to trade off top-notch ASIC speed for programmability in case your algorithm isn't debugged or you get flaws in the silicon 3B) you wanted to accelerate a broad, flexible set of functions faster than, say, software, but didn't want the expense of a general purpose microprocessor 3C) you really want to accelerate one algorithm now, but in few minutes, you want to accelerate another. With 3D you're changing the path picked through the 3D pipeline multiple times per frame, at every state change. You wouldn't be able to reconfigure your chip fast enough to optimize for that type of changing; at best, you might reconfigure your logic for a particular 3D game and the 3D primitives it uses.
Transmeta and FPGA chips can of course accelerate 3D logic, but what you have to realize about 3D logic is that 1) it is very branchy- lots of special cases depending on just whatever mode you're in (This makes DSPs and rasterization a poor fit, although Intergraph has used DSPs for geometry acceleration) 2) the front end geometry processing is primarily floating point (vertex) matrix manipulation, the back-end rasterization is primarily integer (pixel) manipulation; your architecture must provide both types of execution units, in the right proportion 3) the process of shipping all the various vertex, texture and triangle-to-pixel data is very timing-sensitive, requiring lots of dedicated FIFO buffers for optimal acceleration 4) many pixel operations such as pixel walking or gouraud shading generally require very simple increment operations that don't require a full integer unit as one would find in a CPU or DSP (FPGAs would be better here, ASICs better still) 5) the data paths between circuits on the chip grows practically exponentially as you go through the 3D graphics pipeline. Megabytes of vertex data turn into gigabytes of pixels. A general purpose CPU or FPGA are not typically optimized for this. 6) moreso than FPGAs or CPUs, a 3D chip has to be optimized for huge output bandwidth to the frame buffer, both read bandwidth (for Z buffer and blending operations) and write bandwidth, with a separate set of data lines for reading in the initial vertex/texture primitives. The backend frame buffer bandwidth typically requires more pins than you'd have in a CPU/FPGA package. And most CPUs, DSPs, and FPGAs don't have such split memory controller setups integrated into the package, requiring a more expensive external chip.
If this got too technical; I apologize for not having time to make it simpler and/or more precise. But nothing I've seen about Transmeta rings any bells as having promise for making a faster 3D graphics chip, something I'd be very interested in.
Just FYI, there's no cable Internet in Westchester county, north of NYC on the border of CT. I don't think there's cable Internet in NYC either, just on Long Island. You're not the only one.;-)
I think you should double-check the alternatives to BA.
I've signed webforms on a lot of DSL sites indicating interest and just got an offer last week from Flashcom for Internet access (w/ Northpoint as carrier) for cheaper than Bell Atlantic ($39.99/mo, free install, free DSL modem, with 1 year term, all for 410kbps down, 200kbps up.) It was a "special limited-time offer", (through Nov 15th?) but I'd be surprised if it doesn't reappear. On the email I received and web form, they said it was a two-year requirement, but when a sales rep called me the next day, he said I only needed to sign up for one year. So do ask about that.
A little slower than BA's 640/90 plan downstream, but $10/mo cheaper and faster upstream. This was in Westchester County about 45 minutes north of NYC, so you might check to see if that offer is available in your area too. I too have no love lost for Nynex/BA.
After BA said on their website in July that they'd have DSL in my area in "August," then "September, ", then "October-November," and now "November-December", I'm not eager get strung along further. And now especially that I'd be paying more. So I signed up w/ Flashcom on Monday. Provisioning is still problematic: 45 days to install they say, in part because they have to get Bell Atlantic, their competitor, to come to my place to do a line test. Screwy, isn't it?
If you define the OS without utilities, that is fine if you are an academic, but destined to fail in the marketplace, precisely because of the dev/admin/user constituencies' different needs.
Microsoft's market success can be partially explained by the degree to which they bundled in a lot of not-technically-OS-function utilities/applications that appealed to users (e.g. Edit in DOS as you mention, modem sw, Solitaire, etc.) If you don't bundle in such applications as part of the "OS", your OS looks less attractive and harder to use to John Doe users. (That why Linux distributions try to do this, although they overdo it on the bundled games, and often present the user with too many options.)
I agree that this blurs the line between applications and OS, something gravely troubling given Microsoft's desktop OS monopoly. Others call it an "operating environment" just to emphasize their OS+apps packaging, and I'm revert to that terminology often also. But basically, I agree with the pragmatic definition of the first poster, that an operating system is what operating system vendors sell.
An "Operating System" is what Operating System makers deliver.
I'd agree with this, but perhaps add:
An "operating system" is the set of interfaces and services that allow software developers (via APIs), system administrators (via admin tools), and users (via GUIs) to manipulate the hardware resources of a computer system. Whatever software an OS vendor bundles with every version of the OS is part of the OS, since it can be depended on by developers/admins/users. Anything else is optional, "layered" software aka middleware.
If you think about it, the tri-partite constituencies (developer, admin, user) explain a lot of OS marketplace dynamics.
To honestly answer your question "What I want to know is what sort of crazy setup would have 4 NICS feeding into the same segment? "
an engineering workgroup wanting fast Ethernet for throwing around big CAD files
Picture a four 100 Mbit ethernet segments hanging off one mondo server. The workstations on the workgroup are split among the four segments- a dozen or so workstations hang off each switched or shared 100Mbit segment. This gets you high network fileserving performance without having to add costly gigabit ethernet (and fiber cabling, until recently) to every desktop.
This is a real-life production UNIX (NFS) fileserver example, BTW.
--LP
P.S. On a slightly related rant: many people (especially journalists, but also novices) and articles seem to confuse "corporate computing" with "enterprise computing". Enterprise != corporate. And all computers in an enterprise aren't necessarily "enterprise computers" or "enterprise servers". Enterprise computing is substantially more demanding than your basic company LAN. It's typically multi-site, and if systems go down, serious money is lost, creating a lot more havoc than a coffee break and a few lost files-in-progress.
No journalling in Windows NT 4.0's NTFS 4.0. Journalling in Windows 2000's NTFS 5.0.
Journalling in both Linux and NT today is alpha/beta level, neither supported nor shipping. Journalling has been in conventional UNIXes for years, with AIX (!) introducing it earliest in 1991.
You are correct that XFS isn't available for Linux. The author's comment there is a red herring. At the very least he could point to the alpha/beta tests of ReiserFS and Stephen Tweedie's ext2 journaling extensions where working code is available to the public.
The NTFS 4.0 in Windows NT 4.0 doesn't have journalling. The NTFS 5.0 in Windows 2000 does include journalling.
Note that none of these journalling filesystems, Linux or NT, are shipping, supported products, so as far as I'm concerned, it's a wash. Conventional UNIXes remain better than Linux or NT, at least for a little while longer.;-)
The article says that the IBM machine was running a large amount of UNIX code. That and the fact that it is called a supercomputer suggests that the machine in question was an IBM SP running AIX, not an S/390 running OS/390 or MVS. Thus Beowulf here is replacing a UNIX "supercomputer" server and the article has literally nothing to do with mainframes.
What applications do you run where integer performance is a bottleneck?
Every single application I run. Word processing, browsing, the OS itself, most spreadsheet work, webserving, distributed number crunching bits of Pi.
You seem to misunderstanding what integer performance represents; it does not have to do primarily with integer number crunching. Remember, branching, string manipulation, pointer handling, GUI/pixel operations, are all built entirely out of integer operations. Integer benchmarks such as SPECint95 typically measure all of these and make a good proxy for general purpose computing (CPU) performance.
Unless you are specifically doing intensive loops of floating point math for physics/math/financial number crunching, FP is mostly irrelevant. Even 3D, the one potentially general purpose use for FP, is about equally dependent on integer performance for all the other operations that the app performs (handling input, AI, sound, other app logic, etc.) Not to mention that once you are using a 3D card with a geometry engine (e.g. Nvidia's GeForce 256 or 3Dlabs Oxygen GVX1), FP becomes practically irrelevant again.
Most CAD and financial apps, the bread and butter of RISC workstations, often used fixed point (i.e. integer-based) arithmetic to avoid past processor limitations in floating point performance; one survey I read indicated that CAD code was 70% integer. Obviously analysis jobs like FEA or CFD tend to be floating-point bound.
Admittedly, I don't need a 700 MHz computer these days, so its true that in some sense, integer performance isn't my bottleneck; my network connection is the real bottleneck. But within the PC, integer is far more of a limiting factor than floating point in the vast majority of usage scenarios.
Yeah, extrapolating G4 specs from G2/G3 is bogus. I tried to put in enough hedges so people wouldn't take it too seriously ("*if* you extrapolate", chips are "mildly similar", focused on comparing with a more recent Power RS-II chip rather than the older 604e, etc.) Perhaps I should have stuck to pointing out that when vendors don't publish performance results on industry standard benchmarks, it's generally because at best, they have no advantage on such benchmarks or at worst, their performance is significantly worse than competitors.
Frankly, Apple was trying to claim the G4 was a supercomputer and 2-4x faster than Pentium III in its marketing ("for certain tasks", aka a highly selected and tuned group of Photoshop tasks and a few obscure Intel signal processing benchmarks (but not Intel's MMX media benchmarks interestingly)). Any benchmarks that showed the chips even close to being comparable would have undercut their hypy marketing. So they didn't release broad-based industry standard performance metrics (or even relevant niche performance metrics such as performance on desktop publishing and/or more common photoshop tasks.) And any performance edge would have been more exposed as Intel cranked up clock rates and overtook them.
I guess the point of my extrapolation which I didn't drive home was that the G4 improvement over IBM's respectable March 1999 Power RS-II chip would require a further 50% improvement in SPECint per clock cycle to come close to matching today's topline 733 MHz PentiumIIIs. (24->36 SPECint) Unlikely given the similarities of timeframe between G4 and RS-II designs (March 99 vs. Sept 99) and similar process technologies. I just don't think the G4 is 50% better than the Power RS-II on a clock-per-clock basis. But this is admittedly a highly imprecise guessing game caused by Apple not releasing standard performance metrics.
Perhaps I should have skipped the issue entirely, but since the previous post specifically asked about G4, I attempted to make the best assessment I could with the data available. I stand by my statement that I don't see any reasonable way a 500 MHz G4 can match a 733 MHz Pentium III on any broad benchmark of general purpose computing tasks. There may not be proof that is true, but the evidence I see all points that direction.
My fault; I presumed AMD still wasn't publishing SPEC scores, but when I went to look, I discovered that hey, they released some with the Athlon launch in September! No 700 MHz results, but they do have 650 MHz Athlon scores, and Intel has released 650 MHz Pentium III scores. So here, hopefully better late than never:
Thus the answer to your question is "Yes, Alpha remains the fastest", with the important caveat that the 10% performance advantage over Intel comes at a significantly higher price. All other RISCs are slower than the fastest Intel systems, at least in terms of uniprocessor integer performance, the best single predictor for most CPU-limited applications.
Note that Apple G4 performance, and performance of IBM's latest S80 (450 MHz Power RS-III) aren't discussed by their respective vendors. If you extrapolated the G4 performance from the mildly similar 340 MHz Power RS-II, performance of a 500 MHz part would be around 23.53 SPECint with SPECfp at 31.8. IBM's Power PC 604e parts have slightly lower integer performance and much lower floating point performance at the same clock rates as the RS-II (375 MHz 604e runs 15.1 SPECint, 10.1 SPECfp,) so even if there are some other G4 improvements, I doubt the 500 MHz G4 will be beating a 733 MHz Pentium III.
Note that these benchmarks don't measure performance of vector-processing chip features like MMX used by a few apps like Photoshop.
--LP
P.S. (Sidebar: The minimal performance value-add of RISC over Intel is is why I think Linux's highly touted multi-architecture support for RISCs is exactly a glowing scalability feature as some apparently make it. I guess 64-bits and floating point comes in handy for a few apps. Other than that, it's a nice plus for legacy hardware that grows less relevant by the day. )
If you actually want the answer to "what features does it have," go download the free 75+ page paper examining commercial UNIX features from D.H. Brown Associates. Click on "1998 report (download PDF)." It's a little old, but quite thorough if you want to know more about Tru64 (and other $$$ UNIX) features. I think they have another white paper focusing just on the improvements in the newer Tru64 5.0 release somewhere on Compaq's website. Search, search, search... ok, here's more on Tru64 v5.0 features from them.
The big deal is that once one knows exactly how the encryption scheme works, one can write encrypted DVD movies onto a disk that *can* be read by regular DVD players. If you can write data to a DVD disk (e.g. via a DVD-RAM drive on your computer), you can write encrypted data to a disk. If you can write encrypted data in the proper format with the proper keys to a disk, you can make the DVD disk appear just like a movie DVD disk (AFAIK.)
The one other caveat is that current DVD-RAM drives are 2.6 GB capacity. Most movie DVDs are use 4.7 GB capacity disks. Writing a 4.7 GB movie onto a 2.6 GB disk loses quality and you don't have a resulting disk that appears just like a fresh 4.7 GB regular DVD movie. The 4.7 GB writable DVD drives for PCs are scheduled to come out the first half of next year. Once they are out, you could effectively create your own "regular DVD movie disks", with the proper software.
With encryption details publicly understood, anyone can create such "proper software" for writing DVD movie content; without it, one would need to get license the encryption techniques which are held very closely and given out under stringent terms to prevent just such potential piracy.
At least that's how I understand it; corrections welcome.
Keep media prices high. If the media prices don't drop enough for the larger 4.7 GB (or even 2.6 GB) media, a widespread copying "problem" will not materialize, or will materialize late enough to put in place a more secure "DVD-II" sucessor. Copying is highly unlikely if media is $25/disk, and still pretty rare if the media is over $10/disk. At $5/disk and below is where copying starts to run rampant.
One might ask, "How could content producers ensure that media costs stay high?" The answer is, like before, to try to coerce DVD device/media manufacturers via a contract to do so. Such a contract would tack on a "royalty" payment to each piece of media. The device/media manufacturers would sign such a contract to keep the DVD market from collapsing or fragmenting under the threat of the content producers halting publishing on DVDs. Something like the CD-R royalty on media, but without the fatal flaw of an exception clause for data-oriented media.
That would prevent 4.6GB DVD from taking off as a data storage medium, painful for device manufacturers, but not necessarily the worst scenario, which would be a total withdrawl of current future video content on DVD. And the content producers might throw the device manufacturers a bone- you can let 2.6 GB media and writers get cheap, but "high-quality" 4.7 and 9.4 GB media must be tarriffed until a DVD-II scenario can be devised (say, with SDMI watermarking).
Perhaps I'm paranoid, but this scenario seems like a likely short term fix to expect from the DVD powers-that-be, if they can pull it off. And they then rush their SDMI watermarking efforts to market as fast as possible.
With any luck SDMI will be rushed enough that they screw that one up too, right?;-)
Does the multi-head support in XFree86 4.0 include support for a single large 3D graphics window being stretched across three screens? (e.g. like the fighter-plane in the Powerview 290 ad?)
When reading a totally unrelated article after my last post, it occurred to me that perhaps Alan Cox was confusing mainframes with supercomputers? Beowulf is replacing supercomputers.
In any case, I find it very odd and mildly disturbing that Alan Cox appears to have so little clue what a mainframe is, even the most basic minimal knowledge.
Hate to accuse someone I admire of gross ignorance, but that's sure what it looks like. Is there any other explanation [for his statement]?
Slashdot Mirror
Assessed just as fiction, it's a nice treatment of a few what-ifs. What happens to civilization after a global nuclear war? What happens to religion? What happens to science? It presents a plausible and thought-provoking scenario.
Skeptics who find the faith/science tension intriguing or thinking Christians, either of whom like fiction will probably find this book very enjoyable. It was my favorite discovery out of several college lit classes.
--LP
The rotating disks don't actually get 1 gigabyte per second read rates, it turns out that only the card media does.
;-)
And as commentary at Tasty Bits From the Technology Front points out, the most outstanding claim about FMD drives is the 1 gigabyte per second read rates, a full 200x faster than a 32x CD-ROM, and 40x faster than a 10,000 RPM hard disk. In comparison, capacity only improves 25x over the 5.2 GB DVDs.
Personally, I find the large capacities C-3D demonstrates just reinforce my perception that buying into DVD technology is just setting yourself up for obsolescence once higher-res HDTV versions of videos and movies become available on post-DVD media like C3D's in a few years (probably more securely next time though!
--LP
To clarify: we agree that Kosovo was not a UN action, it was a US/NATO one, against a non-NATO nation.
I'm concerned that your argument is based on a false premise- that you can keep all the rights you have right now AND have your government involved with all these para-national organizations as currently constituted.
I'm not talking about replacing the US govt with these para-national bodies, obviously such bodies will augment and/or replace pieces of US govt function, just as national authority first augmented and partially replaced colonial state authority. My point is that even in such a partial scenario, you and I are fundamentally losing rights unless the para-national organization is properly constructed. Take for example your premise that you have the right to know what your government is up to. I'd agree. But obviously if the WTO sets some policy in a secret meeting that your government must adhere to, then your ability to know "what your government is up to" has been reduced. If part of your monetary policy is set in Brussels rather than by officials you elect, you fundamentally lose some of the power you previously wielded with your voting rights which have shrunk ever so slightly. You say this is OK "as long as I retain the rights I currently have," but my point is that once these situations occur, you have lost some of your rights and abilities to remedy the situation.
You can't have your cake and eat it too; unless you and I want to lose bits and pieces of our current rights, we have to push our elected officials to increase the transparency, accountability, and democratic process of these para-national organizations they are delegating their (and our) authority to.
--LP
You seem rather quick to discard a principle (sovereignty) that has been around since the Treaty of Westphalia in 1648, underpinning the basis for relationships between nations at first in Europe and now around the world.
"I don't interfere with the way you run your country (unless you give me permission first)." It's a great principle, based in part on the golden rule (after all, I don't want your country interfering in our political process/internal affairs either.)
The EU and WTO technically don't interfere with sovereignity, since both countries agreed to the terms of those agreements up front. U.N. intervention in a country is more suspect, since U.N. membership is not predicated on such a contract. The U.S. invading Kosovo, IMHO, is a good example of violating another country's sovereignity (bombing a country that is killing a few thousand of its citizens.) There are various arguments about what heinousness of activity or what degree of multi-lateral assent from other countries or bodies like the U.N. justifies violating another country's sovereignity. But that's a whole other debate.
I care about my rights, and as long as they're preserved, why should I care whose running the show?
Because your rights aren't necessarily preserved. Can you vote rulers/officers of these para-national bodies out of office? Are the actions taken by such a body (actions that affect you) open to public, open review? Is there a legal appeals process and what are your rights under that process as a citizen? And beyond the protection of your rights under such a system, on a pragmatic basis, is power adequately distributed via a system of checks and balances to prevent or ameliorate corruption or tyranny?
These are all grave concerns that many people have with the multi-national bodies that have sprung up within the last 50 years. Sure, the UN/EU/WTO/etc have valuable uses for our global society; that's why they exist. However, many of the bodies appear to be less democratic than the nation-states that populate them. The concern is (as with software licensing), who controls the code?
--LP
Try the Amazon.com "Wish List" feature; it lets you keep track of books you want to buy "later", a list Amazon keeps track of. Very cool. You can tell Amazon to make that list public or private to people who know your email address, or tell Amazon to email your wishlist to someone else.
- Better than a Gift Certificate, since you get the book Christmas morning, less hassle
- Better than getting someone else's book pick, at least on average
- Easier than keeping track of some easily-lost self-created book list
- Good for keeping track of books you want to buy for yourself, but don't want to binge on at the moment
I get a tad nervous about the lock-in this feature presents vis-a-vis Amazon, but it's too handy to turn down; I'd love to see it imitated elsewhere.
--LP, no relations to Amazon, just glad to see a feature I've wanted for the last year finally implemented
P.S. If you want to make it less obvious that you're giving a gift, try getting and wrapping up an old CD jewel case and putting a note in it telling the person where to find their *real* present. I've had fun with this and variants of this...
Too bad it'll never get moderated up at this point, due to the day-late nature of my post. One of the major flaws with the slashdot approach IMHO.
The only way I can think of to solve it would be to force moderators to read a certain set of posts that haven't been seen by other moderators (somewhat like the meta-moderation process.) I admit this would add an additional small discouragement factor to moderators.
--LP
Your description of graphics processors, including "XORs, fills, tight loops oriented towards shading" describes the bottlenecks in 2D graphics chips well, but does not describe the problems one would need to optimize in a 3D graphics chip. The 3D bottlenecks are not amenable to a Transmeta-style solution, IMHO.
I've done some research on the 3D graphics chips of UNIX workstations and PCs over the last five years and Transmeta's approach doesn't make any sense as a peak-performance graphics chip. While I must apologize to any real 3D chip designers out there for my generalizations and any misconceptions they may spot with their even greater experience,
I'll try to summarize why programmable chips don't make sense to accomplish fast 3D:
1) If the algorithm a chip must execute is fixed (as it generally is with 3D algorithms), nothing is faster than a well-designed hard-coded ASIC that lays down precisely the circuits needed for optimally accelerating that algorithm.
2) If the algorithm varies substantially, a general purpose processor is more useful. In cases where there is significant unpredictable branching in the algorithm, a general purpose CPU will be optimal. In cases where there is strong data parallelism, a DSP will be optimal.
3) Run-time programmable logic, such as FPGAs (field programmable gate arrays), microcode, or Transmeta-style programmable logic, has traditionally been best for cases where
3A) you want to trade off top-notch ASIC speed for programmability in case your algorithm isn't debugged or you get flaws in the silicon
3B) you wanted to accelerate a broad, flexible set of functions faster than, say, software, but didn't want the expense of a general purpose microprocessor
3C) you really want to accelerate one algorithm now, but in few minutes, you want to accelerate another. With 3D you're changing the path picked through the 3D pipeline multiple times per frame, at every state change. You wouldn't be able to reconfigure your chip fast enough to optimize for that type of changing; at best, you might reconfigure your logic for a particular 3D game and the 3D primitives it uses.
Transmeta and FPGA chips can of course accelerate 3D logic, but what you have to realize about 3D logic is that
1) it is very branchy- lots of special cases depending on just whatever mode you're in (This makes DSPs and rasterization a poor fit, although Intergraph has used DSPs for geometry acceleration)
2) the front end geometry processing is primarily floating point (vertex) matrix manipulation, the back-end rasterization is primarily integer (pixel) manipulation; your architecture must provide both types of execution units, in the right proportion
3) the process of shipping all the various vertex, texture and triangle-to-pixel data is very timing-sensitive, requiring lots of dedicated FIFO buffers for optimal acceleration
4) many pixel operations such as pixel walking or gouraud shading generally require very simple increment operations that don't require a full integer unit as one would find in a CPU or DSP (FPGAs would be better here, ASICs better still)
5) the data paths between circuits on the chip grows practically exponentially as you go through the 3D graphics pipeline. Megabytes of vertex data turn into gigabytes of pixels. A general purpose CPU or FPGA are not typically optimized for this.
6) moreso than FPGAs or CPUs, a 3D chip has to be optimized for huge output bandwidth to the frame buffer, both read bandwidth (for Z buffer and blending operations) and write bandwidth, with a separate set of data lines for reading in the initial vertex/texture primitives. The backend frame buffer bandwidth typically requires more pins than you'd have in a CPU/FPGA package. And most CPUs, DSPs, and FPGAs don't have such split memory controller setups integrated into the package, requiring a more expensive external chip.
If this got too technical; I apologize for not having time to make it simpler and/or more precise. But nothing I've seen about Transmeta rings any bells as having promise for making a faster 3D graphics chip, something I'd be very interested in.
--LP
You're contradicting yourself.
- analysts predict the death of the PC and rise of NC
- analysts seem to take MS promises at face value
Perhaps your brush is too broad. Do you find it more easy to remember predictions that were wrong rather than predictions that were right? I do.
--LP
Just FYI, there's no cable Internet in Westchester county, north of NYC on the border of CT. I don't think there's cable Internet in NYC either, just on Long Island. You're not the only one. ;-)
--LP
I think you should double-check the alternatives to BA.
I've signed webforms on a lot of DSL sites indicating interest and just got an offer last week from Flashcom for Internet access (w/ Northpoint as carrier) for cheaper than Bell Atlantic ($39.99/mo, free install, free DSL modem, with 1 year term, all for 410kbps down, 200kbps up.) It was a "special limited-time offer", (through Nov 15th?) but I'd be surprised if it doesn't reappear. On the email I received and web form, they said it was a two-year requirement, but when a sales rep called me the next day, he said I only needed to sign up for one year. So do ask about that.
A little slower than BA's 640/90 plan downstream, but $10/mo cheaper and faster upstream. This was in Westchester County about 45 minutes north of NYC, so you might check to see if that offer is available in your area too. I too have no love lost for Nynex/BA.
After BA said on their website in July that they'd have DSL in my area in "August," then "September, ", then "October-November," and now "November-December", I'm not eager get strung along further. And now especially that I'd be paying more. So I signed up w/ Flashcom on Monday. Provisioning is still problematic: 45 days to install they say, in part because they have to get Bell Atlantic, their competitor, to come to my place to do a line test. Screwy, isn't it?
--LP
If you define the OS without utilities, that is fine if you are an academic, but destined to fail in the marketplace, precisely because of the dev/admin/user constituencies' different needs.
Microsoft's market success can be partially explained by the degree to which they bundled in a lot of not-technically-OS-function utilities/applications that appealed to users (e.g. Edit in DOS as you mention, modem sw, Solitaire, etc.) If you don't bundle in such applications as part of the "OS", your OS looks less attractive and harder to use to John Doe users. (That why Linux distributions try to do this, although they overdo it on the bundled games, and often present the user with too many options.)
I agree that this blurs the line between applications and OS, something gravely troubling given Microsoft's desktop OS monopoly. Others call it an "operating environment" just to emphasize their OS+apps packaging, and I'm revert to that terminology often also. But basically, I agree with the pragmatic definition of the first poster, that an operating system is what operating system vendors sell.
--LP
An "Operating System" is what Operating System makers deliver.
I'd agree with this, but perhaps add:
An "operating system" is the set of interfaces and services that allow software developers (via APIs), system administrators (via admin tools), and users (via GUIs) to manipulate the hardware resources of a computer system. Whatever software an OS vendor bundles with every version of the OS is part of the OS, since it can be depended on by developers/admins/users. Anything else is optional, "layered" software aka middleware.
If you think about it, the tri-partite constituencies (developer, admin, user) explain a lot of OS marketplace dynamics.
--LP
To honestly answer your question "What I want to know is what sort of crazy setup would have 4 NICS feeding into the same segment? "
an engineering workgroup wanting fast Ethernet for throwing around big CAD files
Picture a four 100 Mbit ethernet segments hanging off one mondo server. The workstations on the workgroup are split among the four segments- a dozen or so workstations hang off each switched or shared 100Mbit segment. This gets you high network fileserving performance without having to add costly gigabit ethernet (and fiber cabling, until recently) to every desktop.
This is a real-life production UNIX (NFS) fileserver example, BTW.
--LP
P.S. On a slightly related rant: many people (especially journalists, but also novices) and articles seem to confuse "corporate computing" with "enterprise computing". Enterprise != corporate. And all computers in an enterprise aren't necessarily "enterprise computers" or "enterprise servers". Enterprise computing is substantially more demanding than your basic company LAN. It's typically multi-site, and if systems go down, serious money is lost, creating a lot more havoc than a coffee break and a few lost files-in-progress.
No journalling in Windows NT 4.0's NTFS 4.0.
Journalling in Windows 2000's NTFS 5.0.
Journalling in both Linux and NT today is alpha/beta level, neither supported nor shipping. Journalling has been in conventional UNIXes for years, with AIX (!) introducing it earliest in 1991.
--LP
You are correct that XFS isn't available for Linux. The author's comment there is a red herring. At the very least he could point to the alpha/beta tests of ReiserFS and Stephen Tweedie's ext2 journaling extensions where working code is available to the public.
;-)
The NTFS 4.0 in Windows NT 4.0 doesn't have journalling. The NTFS 5.0 in Windows 2000 does include journalling.
Note that none of these journalling filesystems, Linux or NT, are shipping, supported products, so as far as I'm concerned, it's a wash. Conventional UNIXes remain better than Linux or NT, at least for a little while longer.
--LP
The article says that the IBM machine was running a large amount of UNIX code. That and the fact that it is called a supercomputer suggests that the machine in question was an IBM SP running AIX, not an S/390 running OS/390 or MVS. Thus Beowulf here is replacing a UNIX "supercomputer" server and the article has literally nothing to do with mainframes.
--LP
What applications do you run where integer performance is a bottleneck?
Every single application I run. Word processing, browsing, the OS itself, most spreadsheet work, webserving, distributed number crunching bits of Pi.
You seem to misunderstanding what integer performance represents; it does not have to do primarily with integer number crunching. Remember, branching, string manipulation, pointer handling, GUI/pixel operations, are all built entirely out of integer operations. Integer benchmarks such as SPECint95 typically measure all of these and make a good proxy for general purpose computing (CPU) performance.
Unless you are specifically doing intensive loops of floating point math for physics/math/financial number crunching, FP is mostly irrelevant. Even 3D, the one potentially general purpose use for FP, is about equally dependent on integer performance for all the other operations that the app performs (handling input, AI, sound, other app logic, etc.) Not to mention that once you are using a 3D card with a geometry engine (e.g. Nvidia's GeForce 256 or 3Dlabs Oxygen GVX1), FP becomes practically irrelevant again.
Most CAD and financial apps, the bread and butter of RISC workstations, often used fixed point (i.e. integer-based) arithmetic to avoid past processor limitations in floating point performance; one survey I read indicated that CAD code was 70% integer. Obviously analysis jobs like FEA or CFD tend to be floating-point bound.
Admittedly, I don't need a 700 MHz computer these days, so its true that in some sense, integer performance isn't my bottleneck; my network connection is the real bottleneck. But within the PC, integer is far more of a limiting factor than floating point in the vast majority of usage scenarios.
--LP
Yeah, extrapolating G4 specs from G2/G3 is bogus. I tried to put in enough hedges so people wouldn't take it too seriously ("*if* you extrapolate", chips are "mildly similar", focused on comparing with a more recent Power RS-II chip rather than the older 604e, etc.) Perhaps I should have stuck to pointing out that when vendors don't publish performance results on industry standard benchmarks, it's generally because at best, they have no advantage on such benchmarks or at worst, their performance is significantly worse than competitors.
Frankly, Apple was trying to claim the G4 was a supercomputer and 2-4x faster than Pentium III in its marketing ("for certain tasks", aka a highly selected and tuned group of Photoshop tasks and a few obscure Intel signal processing benchmarks (but not Intel's MMX media benchmarks interestingly)). Any benchmarks that showed the chips even close to being comparable would have undercut their hypy marketing. So they didn't release broad-based industry standard performance metrics (or even relevant niche performance metrics such as performance on desktop publishing and/or more common photoshop tasks.) And any performance edge would have been more exposed as Intel cranked up clock rates and overtook them.
I guess the point of my extrapolation which I didn't drive home was that the G4 improvement over IBM's respectable March 1999 Power RS-II chip would require a further 50% improvement in SPECint per clock cycle to come close to matching today's topline 733 MHz PentiumIIIs. (24->36 SPECint) Unlikely given the similarities of timeframe between G4 and RS-II designs (March 99 vs. Sept 99) and similar process technologies. I just don't think the G4 is 50% better than the Power RS-II on a clock-per-clock basis. But this is admittedly a highly imprecise guessing game caused by Apple not releasing standard performance metrics.
Perhaps I should have skipped the issue entirely, but since the previous post specifically asked about G4, I attempted to make the best assessment I could with the data available. I stand by my statement that I don't see any reasonable way a 500 MHz G4 can match a 733 MHz Pentium III on any broad benchmark of general purpose computing tasks. There may not be proof that is true, but the evidence I see all points that direction.
--LP
My fault; I presumed AMD still wasn't publishing SPEC scores, but when I went to look, I discovered that hey, they released some with the Athlon launch in September! No 700 MHz results, but they do have 650 MHz Athlon scores, and Intel has released 650 MHz Pentium III scores. So here, hopefully better late than never:
650 MHz Athlon, Microstar MS-6167 mboard:
29.4 SPECint95 / 22.6 SPECfp95
650 MHz Pentium III, SE440BX2 mboard (results from Intel, not SPEC directly):
31.6 SPECint95 / 22.9 SPECfp95
I'm not sure where you get your statement that "AMD performs faster than the Pentium III in almost all tasks." What tasks, exactly?
Personally, I find differences of 5-10% irrelevant between any two chips anyway. Athlon and PIII appear quite comparable performance-wise.
--LP
The best fair metric for chip performance seems to remain SPECint95 and SPECfp95 (www.specbench.org).
By that metric, the fastest recorded RISC and CISC speeds (1 per architecture) are the following:
700 MHz 21264A Compaq AlphaServer GS60E:
39.1 SPECint95 / 68.10 SPECfp95 (or w/ minimum optimizations, 34.7 SPECint_base95 / 54.5 SPECfp_base95)
733 MHz Pentium III (i840) Intel:
35.6 SPECint95 / 30.4 SPECfp95 (no _base figures available, figures from Intel, not yet on SPEC website)
440 MHz PA-8500 HP N4000:
34.0 SPECint95 / 51.4 SPECfp95 (30.8 int_base / 48.7 fp_base)
450 MHz UltraSPARC-II Sun Ultra 60 Model 1450:
19.7 SPECint95 / 27.0 SPECfp95 (16.2 int_base / 23.90 fp_base)
300 MHz MIPS R12000 Origin 2000 2-way:
18.4 SPECint95 / 34.4 SPECfp95 (18.1 int_base / 30.1 fp_base)
340 MHz PowerPC RS64-II IBM H70:
16.0 SPECint95 / 21.2 SPECfp95 (13.7 int_base / 20.2 fp_base)
Thus the answer to your question is "Yes, Alpha remains the fastest", with the important caveat that the 10% performance advantage over Intel comes at a significantly higher price. All other RISCs are slower than the fastest Intel systems, at least in terms of uniprocessor integer performance, the best single predictor for most CPU-limited applications.
Note that Apple G4 performance, and performance of IBM's latest S80 (450 MHz Power RS-III) aren't discussed by their respective vendors. If you extrapolated the G4 performance from the mildly similar 340 MHz Power RS-II, performance of a 500 MHz part would be around 23.53 SPECint with SPECfp at 31.8. IBM's Power PC 604e parts have slightly lower integer performance and much lower floating point performance at the same clock rates as the RS-II (375 MHz 604e runs 15.1 SPECint, 10.1 SPECfp,) so even if there are some other G4 improvements, I doubt the 500 MHz G4 will be beating a 733 MHz Pentium III.
Note that these benchmarks don't measure performance of vector-processing chip features like MMX used by a few apps like Photoshop.
--LP
P.S. (Sidebar: The minimal performance value-add of RISC over Intel is is why I think Linux's highly touted multi-architecture support for RISCs is exactly a glowing scalability feature as some apparently make it. I guess 64-bits and floating point comes in handy for a few apps. Other than that, it's a nice plus for legacy hardware that grows less relevant by the day. )
If you actually want the answer to "what features does it have," go download the free 75+ page paper examining commercial UNIX features from D.H. Brown Associates. Click on "1998 report (download PDF)." It's a little old, but quite thorough if you want to know more about Tru64 (and other $$$ UNIX) features. I think they have another white paper focusing just on the improvements in the newer Tru64 5.0 release somewhere on Compaq's website. Search, search, search... ok, here's more on Tru64 v5.0 features from them.
--LP
The big deal is that once one knows exactly how the encryption scheme works, one can write encrypted DVD movies onto a disk that *can* be read by regular DVD players. If you can write data to a DVD disk (e.g. via a DVD-RAM drive on your computer), you can write encrypted data to a disk. If you can write encrypted data in the proper format with the proper keys to a disk, you can make the DVD disk appear just like a movie DVD disk (AFAIK.)
The one other caveat is that current DVD-RAM drives are 2.6 GB capacity. Most movie DVDs are use 4.7 GB capacity disks. Writing a 4.7 GB movie onto a 2.6 GB disk loses quality and you don't have a resulting disk that appears just like a fresh 4.7 GB regular DVD movie. The 4.7 GB writable DVD drives for PCs are scheduled to come out the first half of next year. Once they are out, you could effectively create your own "regular DVD movie disks", with the proper software.
With encryption details publicly understood, anyone can create such "proper software" for writing DVD movie content; without it, one would need to get license the encryption techniques which are held very closely and given out under stringent terms to prevent just such potential piracy.
At least that's how I understand it; corrections welcome.
--LP
Keep media prices high. If the media prices don't drop enough for the larger 4.7 GB (or even 2.6 GB) media, a widespread copying "problem" will not materialize, or will materialize late enough to put in place a more secure "DVD-II" sucessor. Copying is highly unlikely if media is $25/disk, and still pretty rare if the media is over $10/disk. At $5/disk and below is where copying starts to run rampant.
;-)
One might ask, "How could content producers ensure that media costs stay high?" The answer is, like before, to try to coerce DVD device/media manufacturers via a contract to do so. Such a contract would tack on a "royalty" payment to each piece of media. The device/media manufacturers would sign such a contract to keep the DVD market from collapsing or fragmenting under the threat of the content producers halting publishing on DVDs. Something like the CD-R royalty on media, but without the fatal flaw of an exception clause for data-oriented media.
That would prevent 4.6GB DVD from taking off as a data storage medium, painful for device manufacturers, but not necessarily the worst scenario, which would be a total withdrawl of current future video content on DVD. And the content producers might throw the device manufacturers a bone- you can let 2.6 GB media and writers get cheap, but "high-quality" 4.7 and 9.4 GB media must be tarriffed until a DVD-II scenario can be devised (say, with SDMI watermarking).
Perhaps I'm paranoid, but this scenario seems like a likely short term fix to expect from the DVD powers-that-be, if they can pull it off. And they then rush their SDMI watermarking efforts to market as fast as possible.
With any luck SDMI will be rushed enough that they screw that one up too, right?
--LinuxParanoid
Does the multi-head support in XFree86 4.0 include support for a single large 3D graphics window being stretched across three screens? (e.g. like the fighter-plane in the Powerview 290 ad?)
--LP
When reading a totally unrelated article after my last post, it occurred to me that perhaps Alan Cox was confusing mainframes with supercomputers? Beowulf is replacing supercomputers.
In any case, I find it very odd and mildly disturbing that Alan Cox appears to have so little clue what a mainframe is, even the most basic minimal knowledge.
Hate to accuse someone I admire of gross ignorance, but that's sure what it looks like. Is there any other explanation [for his statement]?
--LP