Poor bandwidth? How, exactly, is the P4's 3.2GB/s of bandwidth poor as compared to the Athlon's 2.1GB/s or the PIII's 1.06GB/s of bandwidth? Of all things that is actually one of the real strong suits for the P4! (at least until AMD brings out their Hammer chips, which have boatloads of bandwidth as far as x86 chips are concerned).
As for the thermal throttling issue, that is as much of a non-issue as the issue of Athlons melting when their heatsink is ripped off. Heatsinks don't fall off, and P4's don't throttle back to lower speed when operating properly (though I would be more then a bit worried about throttling on some of those "laptops" that use desktop P4 processors).
The main reasons why the P4 is sometimes slower then the PIII is mainly because it's L1 data cache is quite small (8K, vs. 16K for the PIII or 64K for the Athlon) and it has a 20-stage pipeline which causes long delays any time it stalls.
The best car-comparison I've been able to find with MHz is the displacement of the engine.
Will a 3.0L engine give more HP then a 2.0L engine? All else being equal, usually yes, but that's not always the case. Most 3.0L engines only produce ~200HP these days, while Honda's S2000 produces 240HP out of a 2.0L engine, and that's without even taking into account things like turbochargers.
ie all else being equal, higher MHz does tend to give a faster processor, but this is not always the case, particularly when looking different processor designs.
Even once you get the HP figure though, will that tell you which car is faster? Not really, an 18-wheeler will have WAY more HP then the above-mentioned S2000, but which is faster? Things like aerodynamics, suspension, transmission, brakes, etc. etc. all play into determining which vehicle is "faster".
ie the processor is only one small part of the system, other components like hard drives
Besides all that, just how do you define "faster". A 1/4 mile time or 0-60mph time will tell you which car has the best acceleration, but will it show you which will be fastest on a rally course? Which will handle best on really rough terrain? Which will be fastest carrying a 2000kg load of cargo?
ie only looking at a benchmark for a single application will ONLY tell you which computer is faster for that one application, and you should really focus on benchmarks that are representative of what you want to do with your computer.
A bit of a side-note to the above, I have tried an nForce, and I HIGHLY recomend it to anyone looking for a reliable chipset with drivers that don't suck.
I've had problems with drivers from virtually EVERY chipset manufacturer out there. Intel had previously been the best in my mind, but they were FAR from fault-less. The early PIIX4 drivers (ie when the 430TX and 440LX chipsets first came out) were TERRIBLE! If you installed them in Win95 with the wrong order of patches your only recourse was to format and reinstall everything. They also OFTEN did not end up working, and when they did they often decreased performance over the stock Windows drivers. Similarly the early i8xx series of drivers were piss-poor as well, though at least these just caused the system to fail to boot in most cases.
The upside for Intel though is that they did eventually get their drivers right after 6 months or so. VIA's drivers usually aren't quite as bad right from the get-go, but they just never quite get the bugs worked out of them, even 3+ years later. SiS has been better, though a bit up and down, and besides, they mostly just stick to Microsoft drivers for most things (it's kinda scary when Microsoft writes the most reliable software for other companies hardware).
nVidia's nForce drivers aren't 100% perfect, I did have one slight problem with them when I first installed my motherboard, but that was using the pre-release drivers shipped with the motherboard (yeah, cutting myself on the bleeding edge again). However I was able to download the release drivers which were available at that time from nVidia, and they worked flawlessly (as have the 1.05 drivers, as well as several versions of their video drivers for the integrated video). Their Linux drivers have had a somewhat bumpier road to stability, and the audio driver still leaves a lot to be desired, but at least they HAVE Linux drivers, which is a good start (though it would be nice if they were open-source).
For Photoshop, the dual-1GHz Macs are pretty much always beaten out by fast single processor AthlonXP or P4 processors, and when you throw a dual-processor AthlonMP or Xeon chip into the mix, it just starts to look messy for Apple.
There's a VERY good reason why Apple/Motorola have NEVER published any cross-platform benchmarks of their chips, and that is because they are SLOW! The fastest G4's are now slower then the slowest PC chips being sold today, and it's only getting worse as time goes by. Even in Apple's last hold-out, Photoshop, the P4 and the Athlon are very solidly in the lead in terms of performance if you look at ANY independant tests.
Note: you have to always be VERY careful with Photoshop benchmark comparisons. It is BY FAR the easiest program to skew results in favor of one system or another, by VERY large margins. By simply chosing different filters, different images and different weightings to put on it all, you can pretty much get a test that will show that ANY chip is faster then any other chip.
Actually, the hard drive manufacturers are about the only people who got it right.
1MB really IS 1,000,000 bytes! exactly as the hard drive manufacturers advertised them. However with computer's we tend to use the term "megabyte" to refer to what is really a "mibibyte" (MiB), ie 2^20.
It's become considerably more complicated then even just that.
For example, teh 486 had something like 8K of L1 cache and usually no L2 cache. The AthlonXP has 128K of L1 and 256K of L2 cache, while the P4 has about 30K of L1 (it's cache is measured somewhat differently) and 512K of L2 cache. These are VERY important figures in determining overall system performance.
Also, we've now got all kinds of DDR and QDR buses. For example, the PIII used to run off of a 133MHz bus, but until recently the P4 only used a 100MHz bus (now up to 133MHz). However the P4's bus is a QDR bus, so it's effectively 400MHz (or 533MHz with the new P4s). On the other hand, while DDR memory runs at only 133MHz, RDRAM runs at 400MHz, but while the RDRAM would seem 3 times faster, you have to also take into account that it only offers 1/4 of the width of pipe, so DDR actually offers more total bandwidth.
Long story short, there is definitely not any single number that you can use to compare the performance of a system these days (though I don't really think that there ever was). Best bet, read the reviews and try to find some benchmarks that as closely as possible match what YOU are planning on using the computer for. Anandtech.com usually does a good job of benchmarking systems, while Tom's Hardware usually does a piss-poor job of it and the trade rags are even worse, but I suppose that at least they're a start, and certainly better then just looking at "MHz".
Absolutely NO ONE would have ever considered the consequences of ripping a heatsink off the processor if it hadn't been for that ridiculous Tom's Hardware video where they did just that (hint to the general readership: Tom measured the P4 at a constant 29C, the overheat protection circuitry on a P4 kicks in somewhere south of 60C... do you see a problem with his results?).
Heatsinks do NOT fall off, or if they do, your processor burning up is probably the least of your worries as compared to, for example, the severe physical thrashing that your computer case is receiving to cause that heatsink to fall off.
If the fan fails, any processor will shut the system down, either by using the RPM detection circuitry built into all modern motherboards, or by using the temperature sensors in the motherboards. Sure, the P4 could keep running without a fan, though if it's not throwing an obvious error message, this might actually be worse then completely shutting down. Why? Because then you've got a server running at 50% (or less) of it's normal performance for no immediately obvious reason. And besides, since we're talking x86, we're not looking at high-reliability servers here (if you're looking to build high-reliabiity servers with x86 equipment, you NEED completely redundant systems anyway), so I'm not real sure how big of an advantage this sort of thing would give you anyway.
Both the P4 and the Athlon are good chips in their own right, and while I do applaud Intel's improved thermal protection circuitry, this isn't exactly what I'd call a really important feature. At best it's good idiot protection in my mind. It's strongest point is to prevent people from frying a chip by improperly installing a heatsink in the first place, which is NOT hard to do unless you're a.) REALLY lazy or b.) not very smart.
Good article, but I think there's one extra important things that you didn't touch on, and that's the Waterloo/Microsoft work mentality.
Waterloo has a certain mentatiliy that they push towards their students (or at least those in Comp Sci and Eng. programs), and that is basically to get the top grades regardless of all else. It's a very competative school, and it's the only school I know of (in Canada at least) where students are publicly rated according to their grades from highest to lowest.
To put it bluntly, Waterloo is definitely NOT known as a party school. Sure, I've gone out drinking with more then a few Waterloo engineers, but by and large, those people a.) hate the school and generally don't enjoy the competative atmosphere, and b.) are usually ranked amoung the lowest in the class:>
From what I understand Microsoft really likes the attitude that the top Waterloo grads tend to bring to the table, ie work for 60-80 hours a week without every enjoying yourself just to get some sort of reputation.
FWIW I actually just finished an engineering degree at the University of Guelph (only about 30 min away from UofW). We also have a co-op program at Guelph, as do pretty much all Canadian engineering schools these days (I suspect that other countries have a lot of similar programs as well). I can tell you that co-op is a great idea, but it has it's share of flaws, regardless of what school you go to (I know of and have worked with a number of people from Waterloo co-op as well as Guelph co-op and a number of other schools). Good idea, but bad implementation.
People always talk about how spam is popular because it's so cheap.
I've always wonder, just HOW cheap is spam? The article mentions 1 million e-mail addresses for $60. Figure that this is probably a high estimate, but you would still have to then pay for at least a few employees in the company to send the spam as well as management and support personel. Then add in the cost of an ISP to send it through. All this probably adds up to around $100 for every million addresses.
Now figure that most of these spam messages are for cheesy-ass products that probably generate only about $5-$10 for every sale once all expenses are taken care of.
So, for the spammer to make back their money, if we assume $10 profit per sale and $100 per 1 million spam messages, they need to get at least 1 out of every 100,000 messages sent to turn into a sale.
Now, I figure that at least 50% of spam goes to non-existant or unused accounts, so that cuts out half of their potential sales. Next figure that 10% of the remaining e-mails probably gets filtered without anyone even seeing it. Finally a lot of other recipients get messages in languages that they can't understand.
So I figure that spamers need at least 1 our of every 25,000 people out there to be DUMB enough to actually buy whatever product is being sold, and that's just to cover the expense of spaming! Ok, that's a really rough estimate, but the general idea remains the same.
Ok, I'm sure some will say that there really are at least 1 in 25,000 people who are dumb enough to honestly believe that a $20 pill will add 3 inches to their penis, but I've got to wonder.
The simple questing that I'm wondering though is this: Is spam actually profitable to anyone? I get the feeling that it's a lot like a pyramid scheme/MLM where a few people at the top make lots of money while the millions of people at the bottom that are actually sending out the spam get screwed over. Hmm.. maybe that helps to explain why so much spam involves pyrimd schemes/MLMs in the first place...
I think that The "we make shit up" Register started the G5 64bit rumour.
Actually, I think it's Motorola that made up this rumor, seeing as their publiclly available roadmap shows the G5 with 64-bit support. See this page on their website. Seems like a fairly natural progression of the Gx line, with the next version being the G5 with "32 and 64 bit products, backwards compatibility".
The Register may have blown this out of proportion (they've been known to do that on MANY occasions), but it looks like this time they at least had a tiny smattering of facts to back them up.
Umm.. I don't know where you got those numbers from, but fabs have been REALLY expensive for quite some time. Obviously a big fab is more expensive then a small fab, but their cost hasn't really changed relative to inflation by all that much.
FWIW AMD's Fab 35 in Germany opened in 1999 at a cost of over $2.0B. Now in 2003, this new fab is up to $2.5B, but it's using 300mm wafers vs. AMD's fab which uses 200mm wafers.
I did a few quick calculations about this sort of thing a while back.
The first thing I figured was that your typical microprocess fab plant had depreciation measured at over a million dollars per day (figure that the plants cost 2 - 3 billion USD to build, and then need to be majorly overhauled about every 3 years or so). So if you figure that there are 1,000 people working there, at a average salary of $50,000, you're paying out only about $136,000 a day in pay.
Long story short, the price of workers is an order of magnitude less expensive then the cost of the fab, so it really doesn't make that much difference how much it costs to pay the workers.
So where do you put the fab? Easy, wherever you get a sweet package from the local government for cheap land and good financing/tax breaks. That's how NY managed to win this fab, that's how Taiwan got started, and that's how China is hoping to build up their fab industry.
Interesting comment about Nero and applications "requiring" reboots in general. Usually you don't actually have to reboot.
With Nero, if you tell it not to reboot, it works perfectly under Win2K. In fact, rebooting doesn't change a single thing, though that doesn't stop them from saying that you need to reboot after you install it. I've run into the same thing with a number of other applications in Win2K. For example, my ISP (Sympatico) has a POS bit of software to do PPPoE for their DSL connections. Now, this software just flat out doesn't work on my PC, but that didn't stop the tech support people from telling me to install and uninstall it about 500 times. Each time for both install and uninstall it requires a reboot to the system, while other PPPoE clients I've used manage to do the same without ever rebooting. What's worse, the Sympatico "Access Manager" software doesn't even give you the choice of whether to reboot or not, it just pops up a window with an "Ok" button that automatically reboots your system, no "Cancel" button and closing the application with Alt-F4 also seems to cause a reboot.
Of course, not to let Microslop off the hook, they do require a reboot for EVERY security patch I've ever downloaded from them I think. I couldn't possibly imagine running an MS server! You would have to reboot once a week just to keep the system secure!
Ada is a language which is very good for certain things, but not that great of a choice for most application code.
I just finished a Systems Engineering degree at my school, and Ada was one of MANY languages we did (aside: I don't quite get some of the other schools out there that use only ONE language, we did Java, C, C++, Perl, Tcl/Tk, Ada, assembly, and there were also courses offered in VB and Fortran, though I didn't take either). I only used Ada for one course, a real-time systems course (more for real-time applications rather then an RTOS course, though we did discuss real-time OSes). For this sort of course, Ada made a reasonably good choice because of the way it handled timing and simultaneous tasks. Sure, that sort of thing could also be done in C or most other languages, but it would have been a fair bit trickier.
However, the real strength of Ada though is that it forces reasonably good code at compile time. It tends to pick up a LOT more potential problems at compile time then a language like Java would, and it certainly picks up a hell of a lot more then C would! It also lets your trap just about any exception and at least try to recover for that sort of problem, or at the very least stop and then restart everything, where with most other languages you would tend to be booted out to the operating system altogether.
Basically, the language was designed by the military for military purposes, and for that use, it works well. It is NOT designed to run just any old application. Writing a text editor in Ada might be possible, but it's a fairly bad idea. There are good reasons why there are dozens of different languages out there. All languages have their set of strengths and weaknesses.
Re:They used �Intel-like� approach to design?!? HA
on
ATI R300 and R250V
·
· Score: 2, Insightful
Just a bit of food for thought...
Intel had to REALLY stretch to get the PIII core up to 1.1GHz on a 180nm fab process, including having to recall their first attempt entirely. With the P4, they had little trouble releasing a 2.0GHz core on the exact same 180nm fab process.
Which do you think is faster, a 1.1GHz PIII or a 2.0GHz P4? Intel's design strategy for the P4 wasn't all about marketing, being a little bit less efficient but clocking a LOT higher isn't entirely a bad thing.
Of course, I don't know just how well this correlates to ATI's newest video card, we'll just have to wait and see.
Just think, how many bragging rights does 60 f/s get you these days?! NONE! But if your card gets over 200 f/s then you get ALL the bragging rights in the chat rooms! You then become really 'l33t!!!
Ohh, other then bragging rights though, you're mostly correct. The human eye tends to consider motion to be fluid at somewhere between 16 and 25 f/s, depending on who you ask. The eye also stops noticing any difference between frame rates at around 60 f/s, and it stops noticing even complete picture changes slightly higher then that (hence the reason why 60Hz refresh rates are a bit too low for most people, but 75 or 85Hz is fine).
Another thing to remember is that most games play at a refresh rate of 75 or 85Hz, so beyond that, the difference is totally none-existant (the extra frames are never even displayed), so all those people who say "I can tell the difference between 75 f/s and 120 f/s" are completely full of shit and are seeing things that very litterally are NOT there.
I know that a lot of people all talk about how these are only "average" frame rates and that when things get really complicated in games your frame rate drops, however most of these people have never bothered to qualify this. If they did, they would find that in most typical games, the worst-case frame rate is more then 50% of your average frame rate quoted in most of these tests. In other words, anything beyond ~150 f/s makes absolutely NO difference at all because both cards ALWAYS display exactly 75 f/s (or 85 f/s, depending on the refresh rate). What's more, most of the frame-rate tests used these days are designed as an absolute worst-case scenario to begin with, so in typical play the numbers shown in this tests are closer to your minimum frame rate then an average.
However, that still doesn't change the bragging-rights requirement..
Re:How does this compare to a 'workstation'?
on
nForce2 Preview
·
· Score: 1
Sweeping generalization mode = On
Generally speaking, a modern PC workstation will beat the pants off any of the modern Unix workstations unless you spend BIG-$$$. IBM just brought out some new workstations that use their Power4 chips, which would be a lot faster then PCs, and some of the Alpha workstation type systems that are hanging around are still pretty fast.
However, in the under $25,000 workstation market, PCs are the fastest platforms out there, usually by a reasonably large margin. Sweeping generalization mode = Off
Now, that being said, the real key to many workstations is not the hardware but the software, and here the Unix workstations sometimes still have an advantage.
When enabled, the integrated graphics WILL use some of your system RAM, and a fixed amount (set in the BIOS) of it at that (unlike Intel's integrated chipsets which dynamically take what they need).
That being, I don't really miss the missing 32MB out of my 512MB of system RAM too often. These days RAM is so cheap, using 32MB for graphics isn't that big of a deal.
Of course, as soon as you plug an external AGP card into the board it automagically disables the integrated graphics and doesn't use any system memory.
c) it's integrated, meaning hard to replace if something goes out and not customizable
Err, sticking a replacement card in the AGP slot is tough? Integrated graphics or not, there's still a standard AGP slot that you can use for an add-in graphics card.
Besides, if you REALLY don't want the integrated graphics, there is always the option of buying a board using the chipset that DOESN'T USE IT!
Re:Now if Nvidia fully support this....
on
nForce2 Preview
·
· Score: 1
FWIW I run an nForce 420 based system, and the integrated sound, ethernet and video all work in Linux. The sound support is somewhat weak (it treats it like a plain old AC97 codec, at least with the current drivers), but it works. This is all with drivers written by and semi-supported by nVidia (err, the sound driver is a slightly modified, open source i810 sound driver). Ohh, and USB works as well, using the standard drivers (can't remember off the top of my head if it's OHCI or UHCI).
It's real simple actually. AMD let's nVidia (and VIA, SiS, ALi and ATI, and just about anyone else who would come along) to produce and sell chipsets for their motherboards for little to no licensing fees. Intel has more expensive licensing fees and has flat out refused to license several companies, including both nVidia and VIA. VIA is sorta skirting the issue, claiming that their acquisition of part of S3 gave their a cross-licensing agreement with Intel, though they're still in court. nVidia doesn't have that option.
At the start of '97, Intel sold over 95% of all the x86 motherboard chipsets. They had a larger market share of motherboard chipsets then they ever had of x86 processors. By the end of '99 their market share had dropped down to about 40%. They've since being doing their best to claw back up.
Have you looked at the spec for ATI's chipsets? They're not at all interesting for the desktop. They have a couple chipsets that look GREAT for laptops (acceptable performance and low power consumption), but for the desktop there are TONS of better chipsets out there.
First off they're only using DDR266 (single channel) for their integrated graphics, which means performance is likely going to stink (integrated graphics performance is almost entirely determined by the amount of memory bandwidth and bandwidth reduction techniques used). Secondly their other integrated components look totally standard (virtually identical to what SiS, VIA and Intel have), and what's worse, right now they're still using PCI to connect their north and south bridges together, just after the very last of the companies (ALi) have moved away from such a setup. This will somewhat limit the bandwidth and performance for I/O on such a system.
To top it all off, it's not likely that ATI will improve the (rather pityful) driver situation with most current motherboards. If nothing else, nVidia has everyone else on the market beat in terms of the quality of their motherboard drivers (even Intel has caused me no end of troubles with their buggy drivers.. SiS was worse and VIA and ALi are worse still). Combine that with the significantly better integrated graphics performance (way more bandwidth due to the dual-channel DDR controller), top-notch audio and a handful of other neat tricks that the nForce brings to the table, and it's no wonder that it's quickly becoming the most popular Athlon chipset (all of the largest OEM customers for the Athlon use the nForce for many of their systems).
Re:Why two ethernet controllers?
on
nForce2 Preview
·
· Score: 1
One little extra reason on top of those already mentioned might have something to do with the history of the nForce's integrated NIC. At least some motherboard manufacturers (Asus is the first that jumps to mind) didn't use this integrated NIC on their first nForce board and instead used an external ethernet controller on their board. The reason, apparently, was because nVidia did not (and I believe still doesn't) have any NT 4.0 drivers for their NIC (they do have Win98/Me and Win2K/XP, as well as Linux drivers available).
Using the extra 3Com ethernet controller would let at least one of the two controllers work under just about any operating system.
Just a little addition to the above. An awful lot of corn grown these days is BT corn (genetically engineered corn with an organic pesticide gene added). I don't have any exact numbers, but I'm fairly certain that it's somewhere in the 50% range. This further reduces the amount of pesticide beyond the techniques mentioned above.
Farming is very much a high-tech business these days, at least in North America and most of Europe (and at least for those farmers that wish to stay in business:> ).
You're actually quite right in general, 32-bits is enough for most applications and, all else being equal, it IS faster then 64-bits (half as much data to read from memory).
Now, that being said, there are a few reasons for going to 64-bits. The real main reason is the native support for more then 4GB of memory. You can hack support for more then 4GB of memory into a 32-bit processor (see Intel's PSE), but it's not pretty. This was a bad kludge when it was done back in the 16-bit days, and it's still a bad kludge.
The second reason for going to x86-64 in particular instead of standard x86 is that it doubles the number of registers you have available. This is not a 32-bit vs. 64-bit issue in general, but rather an x86 vs. x86-64 in specific issue. AMD estimates that this will boost performance by 5 to 15% on applications, simply by recomiling them for x86-64 (same hardware and same compiler, just x86 vs. x86-64).
As for the people who talk about "why don't we have 256-bit processors yet" though, you're dead-on as to the reason, they would quite simply be SLOW. The "128-bit" processors that you see in modern gaming consoles like the Playstation 2 are talking about VERY different bitness then in the general computing world (what they talk about is more analogous to the 128-bit or 256-bit graphics cards that most PCs use these days).
Slashdot Mirror
Poor bandwidth? How, exactly, is the P4's 3.2GB/s of bandwidth poor as compared to the Athlon's 2.1GB/s or the PIII's 1.06GB/s of bandwidth? Of all things that is actually one of the real strong suits for the P4! (at least until AMD brings out their Hammer chips, which have boatloads of bandwidth as far as x86 chips are concerned).
As for the thermal throttling issue, that is as much of a non-issue as the issue of Athlons melting when their heatsink is ripped off. Heatsinks don't fall off, and P4's don't throttle back to lower speed when operating properly (though I would be more then a bit worried about throttling on some of those "laptops" that use desktop P4 processors).
The main reasons why the P4 is sometimes slower then the PIII is mainly because it's L1 data cache is quite small (8K, vs. 16K for the PIII or 64K for the Athlon) and it has a 20-stage pipeline which causes long delays any time it stalls.
The best car-comparison I've been able to find with MHz is the displacement of the engine.
Will a 3.0L engine give more HP then a 2.0L engine? All else being equal, usually yes, but that's not always the case. Most 3.0L engines only produce ~200HP these days, while Honda's S2000 produces 240HP out of a 2.0L engine, and that's without even taking into account things like turbochargers.
ie all else being equal, higher MHz does tend to give a faster processor, but this is not always the case, particularly when looking different processor designs.
Even once you get the HP figure though, will that tell you which car is faster? Not really, an 18-wheeler will have WAY more HP then the above-mentioned S2000, but which is faster? Things like aerodynamics, suspension, transmission, brakes, etc. etc. all play into determining which vehicle is "faster".
ie the processor is only one small part of the system, other components like hard drives
Besides all that, just how do you define "faster". A 1/4 mile time or 0-60mph time will tell you which car has the best acceleration, but will it show you which will be fastest on a rally course? Which will handle best on really rough terrain? Which will be fastest carrying a 2000kg load of cargo?
ie only looking at a benchmark for a single application will ONLY tell you which computer is faster for that one application, and you should really focus on benchmarks that are representative of what you want to do with your computer.
A bit of a side-note to the above, I have tried an nForce, and I HIGHLY recomend it to anyone looking for a reliable chipset with drivers that don't suck.
I've had problems with drivers from virtually EVERY chipset manufacturer out there. Intel had previously been the best in my mind, but they were FAR from fault-less. The early PIIX4 drivers (ie when the 430TX and 440LX chipsets first came out) were TERRIBLE! If you installed them in Win95 with the wrong order of patches your only recourse was to format and reinstall everything. They also OFTEN did not end up working, and when they did they often decreased performance over the stock Windows drivers. Similarly the early i8xx series of drivers were piss-poor as well, though at least these just caused the system to fail to boot in most cases.
The upside for Intel though is that they did eventually get their drivers right after 6 months or so. VIA's drivers usually aren't quite as bad right from the get-go, but they just never quite get the bugs worked out of them, even 3+ years later. SiS has been better, though a bit up and down, and besides, they mostly just stick to Microsoft drivers for most things (it's kinda scary when Microsoft writes the most reliable software for other companies hardware).
nVidia's nForce drivers aren't 100% perfect, I did have one slight problem with them when I first installed my motherboard, but that was using the pre-release drivers shipped with the motherboard (yeah, cutting myself on the bleeding edge again). However I was able to download the release drivers which were available at that time from nVidia, and they worked flawlessly (as have the 1.05 drivers, as well as several versions of their video drivers for the integrated video). Their Linux drivers have had a somewhat bumpier road to stability, and the audio driver still leaves a lot to be desired, but at least they HAVE Linux drivers, which is a good start (though it would be nice if they were open-source).
For Photoshop, the dual-1GHz Macs are pretty much always beaten out by fast single processor AthlonXP or P4 processors, and when you throw a dual-processor AthlonMP or Xeon chip into the mix, it just starts to look messy for Apple.
There's a VERY good reason why Apple/Motorola have NEVER published any cross-platform benchmarks of their chips, and that is because they are SLOW! The fastest G4's are now slower then the slowest PC chips being sold today, and it's only getting worse as time goes by. Even in Apple's last hold-out, Photoshop, the P4 and the Athlon are very solidly in the lead in terms of performance if you look at ANY independant tests.
Note: you have to always be VERY careful with Photoshop benchmark comparisons. It is BY FAR the easiest program to skew results in favor of one system or another, by VERY large margins. By simply chosing different filters, different images and different weightings to put on it all, you can pretty much get a test that will show that ANY chip is faster then any other chip.
Actually, the hard drive manufacturers are about the only people who got it right.
1MB really IS 1,000,000 bytes! exactly as the hard drive manufacturers advertised them. However with computer's we tend to use the term "megabyte" to refer to what is really a "mibibyte" (MiB), ie 2^20.
It's become considerably more complicated then even just that.
For example, teh 486 had something like 8K of L1 cache and usually no L2 cache. The AthlonXP has 128K of L1 and 256K of L2 cache, while the P4 has about 30K of L1 (it's cache is measured somewhat differently) and 512K of L2 cache. These are VERY important figures in determining overall system performance.
Also, we've now got all kinds of DDR and QDR buses. For example, the PIII used to run off of a 133MHz bus, but until recently the P4 only used a 100MHz bus (now up to 133MHz). However the P4's bus is a QDR bus, so it's effectively 400MHz (or 533MHz with the new P4s). On the other hand, while DDR memory runs at only 133MHz, RDRAM runs at 400MHz, but while the RDRAM would seem 3 times faster, you have to also take into account that it only offers 1/4 of the width of pipe, so DDR actually offers more total bandwidth.
Long story short, there is definitely not any single number that you can use to compare the performance of a system these days (though I don't really think that there ever was). Best bet, read the reviews and try to find some benchmarks that as closely as possible match what YOU are planning on using the computer for. Anandtech.com usually does a good job of benchmarking systems, while Tom's Hardware usually does a piss-poor job of it and the trade rags are even worse, but I suppose that at least they're a start, and certainly better then just looking at "MHz".
I find comments like this rather interesting.
Absolutely NO ONE would have ever considered the consequences of ripping a heatsink off the processor if it hadn't been for that ridiculous Tom's Hardware video where they did just that (hint to the general readership: Tom measured the P4 at a constant 29C, the overheat protection circuitry on a P4 kicks in somewhere south of 60C... do you see a problem with his results?).
Heatsinks do NOT fall off, or if they do, your processor burning up is probably the least of your worries as compared to, for example, the severe physical thrashing that your computer case is receiving to cause that heatsink to fall off.
If the fan fails, any processor will shut the system down, either by using the RPM detection circuitry built into all modern motherboards, or by using the temperature sensors in the motherboards. Sure, the P4 could keep running without a fan, though if it's not throwing an obvious error message, this might actually be worse then completely shutting down. Why? Because then you've got a server running at 50% (or less) of it's normal performance for no immediately obvious reason. And besides, since we're talking x86, we're not looking at high-reliability servers here (if you're looking to build high-reliabiity servers with x86 equipment, you NEED completely redundant systems anyway), so I'm not real sure how big of an advantage this sort of thing would give you anyway.
Both the P4 and the Athlon are good chips in their own right, and while I do applaud Intel's improved thermal protection circuitry, this isn't exactly what I'd call a really important feature. At best it's good idiot protection in my mind. It's strongest point is to prevent people from frying a chip by improperly installing a heatsink in the first place, which is NOT hard to do unless you're a.) REALLY lazy or b.) not very smart.
Good article, but I think there's one extra important things that you didn't touch on, and that's the Waterloo/Microsoft work mentality.
:>
Waterloo has a certain mentatiliy that they push towards their students (or at least those in Comp Sci and Eng. programs), and that is basically to get the top grades regardless of all else. It's a very competative school, and it's the only school I know of (in Canada at least) where students are publicly rated according to their grades from highest to lowest.
To put it bluntly, Waterloo is definitely NOT known as a party school. Sure, I've gone out drinking with more then a few Waterloo engineers, but by and large, those people a.) hate the school and generally don't enjoy the competative atmosphere, and b.) are usually ranked amoung the lowest in the class
From what I understand Microsoft really likes the attitude that the top Waterloo grads tend to bring to the table, ie work for 60-80 hours a week without every enjoying yourself just to get some sort of reputation.
FWIW I actually just finished an engineering degree at the University of Guelph (only about 30 min away from UofW). We also have a co-op program at Guelph, as do pretty much all Canadian engineering schools these days (I suspect that other countries have a lot of similar programs as well). I can tell you that co-op is a great idea, but it has it's share of flaws, regardless of what school you go to (I know of and have worked with a number of people from Waterloo co-op as well as Guelph co-op and a number of other schools). Good idea, but bad implementation.
People always talk about how spam is popular because it's so cheap.
I've always wonder, just HOW cheap is spam? The article mentions 1 million e-mail addresses for $60. Figure that this is probably a high estimate, but you would still have to then pay for at least a few employees in the company to send the spam as well as management and support personel. Then add in the cost of an ISP to send it through. All this probably adds up to around $100 for every million addresses.
Now figure that most of these spam messages are for cheesy-ass products that probably generate only about $5-$10 for every sale once all expenses are taken care of.
So, for the spammer to make back their money, if we assume $10 profit per sale and $100 per 1 million spam messages, they need to get at least 1 out of every 100,000 messages sent to turn into a sale.
Now, I figure that at least 50% of spam goes to non-existant or unused accounts, so that cuts out half of their potential sales. Next figure that 10% of the remaining e-mails probably gets filtered without anyone even seeing it. Finally a lot of other recipients get messages in languages that they can't understand.
So I figure that spamers need at least 1 our of every 25,000 people out there to be DUMB enough to actually buy whatever product is being sold, and that's just to cover the expense of spaming! Ok, that's a really rough estimate, but the general idea remains the same.
Ok, I'm sure some will say that there really are at least 1 in 25,000 people who are dumb enough to honestly believe that a $20 pill will add 3 inches to their penis, but I've got to wonder.
The simple questing that I'm wondering though is this: Is spam actually profitable to anyone? I get the feeling that it's a lot like a pyramid scheme/MLM where a few people at the top make lots of money while the millions of people at the bottom that are actually sending out the spam get screwed over. Hmm.. maybe that helps to explain why so much spam involves pyrimd schemes/MLMs in the first place...
I think that The "we make shit up" Register started the G5 64bit rumour.
Actually, I think it's Motorola that made up this rumor, seeing as their publiclly available roadmap shows the G5 with 64-bit support. See this page on their website. Seems like a fairly natural progression of the Gx line, with the next version being the G5 with "32 and 64 bit products, backwards compatibility".
The Register may have blown this out of proportion (they've been known to do that on MANY occasions), but it looks like this time they at least had a tiny smattering of facts to back them up.
Umm.. I don't know where you got those numbers from, but fabs have been REALLY expensive for quite some time. Obviously a big fab is more expensive then a small fab, but their cost hasn't really changed relative to inflation by all that much.
FWIW AMD's Fab 35 in Germany opened in 1999 at a cost of over $2.0B. Now in 2003, this new fab is up to $2.5B, but it's using 300mm wafers vs. AMD's fab which uses 200mm wafers.
I did a few quick calculations about this sort of thing a while back.
The first thing I figured was that your typical microprocess fab plant had depreciation measured at over a million dollars per day (figure that the plants cost 2 - 3 billion USD to build, and then need to be majorly overhauled about every 3 years or so). So if you figure that there are 1,000 people working there, at a average salary of $50,000, you're paying out only about $136,000 a day in pay.
Long story short, the price of workers is an order of magnitude less expensive then the cost of the fab, so it really doesn't make that much difference how much it costs to pay the workers.
So where do you put the fab? Easy, wherever you get a sweet package from the local government for cheap land and good financing/tax breaks. That's how NY managed to win this fab, that's how Taiwan got started, and that's how China is hoping to build up their fab industry.
Interesting comment about Nero and applications "requiring" reboots in general. Usually you don't actually have to reboot.
With Nero, if you tell it not to reboot, it works perfectly under Win2K. In fact, rebooting doesn't change a single thing, though that doesn't stop them from saying that you need to reboot after you install it. I've run into the same thing with a number of other applications in Win2K. For example, my ISP (Sympatico) has a POS bit of software to do PPPoE for their DSL connections. Now, this software just flat out doesn't work on my PC, but that didn't stop the tech support people from telling me to install and uninstall it about 500 times. Each time for both install and uninstall it requires a reboot to the system, while other PPPoE clients I've used manage to do the same without ever rebooting. What's worse, the Sympatico "Access Manager" software doesn't even give you the choice of whether to reboot or not, it just pops up a window with an "Ok" button that automatically reboots your system, no "Cancel" button and closing the application with Alt-F4 also seems to cause a reboot.
Of course, not to let Microslop off the hook, they do require a reboot for EVERY security patch I've ever downloaded from them I think. I couldn't possibly imagine running an MS server! You would have to reboot once a week just to keep the system secure!
Ada is a language which is very good for certain things, but not that great of a choice for most application code.
I just finished a Systems Engineering degree at my school, and Ada was one of MANY languages we did (aside: I don't quite get some of the other schools out there that use only ONE language, we did Java, C, C++, Perl, Tcl/Tk, Ada, assembly, and there were also courses offered in VB and Fortran, though I didn't take either). I only used Ada for one course, a real-time systems course (more for real-time applications rather then an RTOS course, though we did discuss real-time OSes). For this sort of course, Ada made a reasonably good choice because of the way it handled timing and simultaneous tasks. Sure, that sort of thing could also be done in C or most other languages, but it would have been a fair bit trickier.
However, the real strength of Ada though is that it forces reasonably good code at compile time. It tends to pick up a LOT more potential problems at compile time then a language like Java would, and it certainly picks up a hell of a lot more then C would! It also lets your trap just about any exception and at least try to recover for that sort of problem, or at the very least stop and then restart everything, where with most other languages you would tend to be booted out to the operating system altogether.
Basically, the language was designed by the military for military purposes, and for that use, it works well. It is NOT designed to run just any old application. Writing a text editor in Ada might be possible, but it's a fairly bad idea. There are good reasons why there are dozens of different languages out there. All languages have their set of strengths and weaknesses.
Just a bit of food for thought...
Intel had to REALLY stretch to get the PIII core up to 1.1GHz on a 180nm fab process, including having to recall their first attempt entirely. With the P4, they had little trouble releasing a 2.0GHz core on the exact same 180nm fab process.
Which do you think is faster, a 1.1GHz PIII or a 2.0GHz P4? Intel's design strategy for the P4 wasn't all about marketing, being a little bit less efficient but clocking a LOT higher isn't entirely a bad thing.
Of course, I don't know just how well this correlates to ATI's newest video card, we'll just have to wait and see.
Of course it matters!
Just think, how many bragging rights does 60 f/s get you these days?! NONE! But if your card gets over 200 f/s then you get ALL the bragging rights in the chat rooms! You then become really 'l33t!!!
Ohh, other then bragging rights though, you're mostly correct. The human eye tends to consider motion to be fluid at somewhere between 16 and 25 f/s, depending on who you ask. The eye also stops noticing any difference between frame rates at around 60 f/s, and it stops noticing even complete picture changes slightly higher then that (hence the reason why 60Hz refresh rates are a bit too low for most people, but 75 or 85Hz is fine).
Another thing to remember is that most games play at a refresh rate of 75 or 85Hz, so beyond that, the difference is totally none-existant (the extra frames are never even displayed), so all those people who say "I can tell the difference between 75 f/s and 120 f/s" are completely full of shit and are seeing things that very litterally are NOT there.
I know that a lot of people all talk about how these are only "average" frame rates and that when things get really complicated in games your frame rate drops, however most of these people have never bothered to qualify this. If they did, they would find that in most typical games, the worst-case frame rate is more then 50% of your average frame rate quoted in most of these tests. In other words, anything beyond ~150 f/s makes absolutely NO difference at all because both cards ALWAYS display exactly 75 f/s (or 85 f/s, depending on the refresh rate). What's more, most of the frame-rate tests used these days are designed as an absolute worst-case scenario to begin with, so in typical play the numbers shown in this tests are closer to your minimum frame rate then an average.
However, that still doesn't change the bragging-rights requirement..
Sweeping generalization mode = On
Generally speaking, a modern PC workstation will beat the pants off any of the modern Unix workstations unless you spend BIG-$$$. IBM just brought out some new workstations that use their Power4 chips, which would be a lot faster then PCs, and some of the Alpha workstation type systems that are hanging around are still pretty fast.
However, in the under $25,000 workstation market, PCs are the fastest platforms out there, usually by a reasonably large margin.
Sweeping generalization mode = Off
Now, that being said, the real key to many workstations is not the hardware but the software, and here the Unix workstations sometimes still have an advantage.
When enabled, the integrated graphics WILL use some of your system RAM, and a fixed amount (set in the BIOS) of it at that (unlike Intel's integrated chipsets which dynamically take what they need).
That being, I don't really miss the missing 32MB out of my 512MB of system RAM too often. These days RAM is so cheap, using 32MB for graphics isn't that big of a deal.
Of course, as soon as you plug an external AGP card into the board it automagically disables the integrated graphics and doesn't use any system memory.
c) it's integrated, meaning hard to replace if something goes out and not customizable
Err, sticking a replacement card in the AGP slot is tough? Integrated graphics or not, there's still a standard AGP slot that you can use for an add-in graphics card.
Besides, if you REALLY don't want the integrated graphics, there is always the option of buying a board using the chipset that DOESN'T USE IT!
FWIW I run an nForce 420 based system, and the integrated sound, ethernet and video all work in Linux. The sound support is somewhat weak (it treats it like a plain old AC97 codec, at least with the current drivers), but it works. This is all with drivers written by and semi-supported by nVidia (err, the sound driver is a slightly modified, open source i810 sound driver). Ohh, and USB works as well, using the standard drivers (can't remember off the top of my head if it's OHCI or UHCI).
It's real simple actually. AMD let's nVidia (and VIA, SiS, ALi and ATI, and just about anyone else who would come along) to produce and sell chipsets for their motherboards for little to no licensing fees. Intel has more expensive licensing fees and has flat out refused to license several companies, including both nVidia and VIA. VIA is sorta skirting the issue, claiming that their acquisition of part of S3 gave their a cross-licensing agreement with Intel, though they're still in court. nVidia doesn't have that option.
At the start of '97, Intel sold over 95% of all the x86 motherboard chipsets. They had a larger market share of motherboard chipsets then they ever had of x86 processors. By the end of '99 their market share had dropped down to about 40%. They've since being doing their best to claw back up.
ATI's chipset? Bleah!
Have you looked at the spec for ATI's chipsets? They're not at all interesting for the desktop. They have a couple chipsets that look GREAT for laptops (acceptable performance and low power consumption), but for the desktop there are TONS of better chipsets out there.
First off they're only using DDR266 (single channel) for their integrated graphics, which means performance is likely going to stink (integrated graphics performance is almost entirely determined by the amount of memory bandwidth and bandwidth reduction techniques used). Secondly their other integrated components look totally standard (virtually identical to what SiS, VIA and Intel have), and what's worse, right now they're still using PCI to connect their north and south bridges together, just after the very last of the companies (ALi) have moved away from such a setup. This will somewhat limit the bandwidth and performance for I/O on such a system.
To top it all off, it's not likely that ATI will improve the (rather pityful) driver situation with most current motherboards. If nothing else, nVidia has everyone else on the market beat in terms of the quality of their motherboard drivers (even Intel has caused me no end of troubles with their buggy drivers.. SiS was worse and VIA and ALi are worse still). Combine that with the significantly better integrated graphics performance (way more bandwidth due to the dual-channel DDR controller), top-notch audio and a handful of other neat tricks that the nForce brings to the table, and it's no wonder that it's quickly becoming the most popular Athlon chipset (all of the largest OEM customers for the Athlon use the nForce for many of their systems).
One little extra reason on top of those already mentioned might have something to do with the history of the nForce's integrated NIC. At least some motherboard manufacturers (Asus is the first that jumps to mind) didn't use this integrated NIC on their first nForce board and instead used an external ethernet controller on their board. The reason, apparently, was because nVidia did not (and I believe still doesn't) have any NT 4.0 drivers for their NIC (they do have Win98/Me and Win2K/XP, as well as Linux drivers available).
Using the extra 3Com ethernet controller would let at least one of the two controllers work under just about any operating system.
Just a little addition to the above. An awful lot of corn grown these days is BT corn (genetically engineered corn with an organic pesticide gene added). I don't have any exact numbers, but I'm fairly certain that it's somewhere in the 50% range. This further reduces the amount of pesticide beyond the techniques mentioned above.
:> ).
Farming is very much a high-tech business these days, at least in North America and most of Europe (and at least for those farmers that wish to stay in business
You're actually quite right in general, 32-bits is enough for most applications and, all else being equal, it IS faster then 64-bits (half as much data to read from memory).
Now, that being said, there are a few reasons for going to 64-bits. The real main reason is the native support for more then 4GB of memory. You can hack support for more then 4GB of memory into a 32-bit processor (see Intel's PSE), but it's not pretty. This was a bad kludge when it was done back in the 16-bit days, and it's still a bad kludge.
The second reason for going to x86-64 in particular instead of standard x86 is that it doubles the number of registers you have available. This is not a 32-bit vs. 64-bit issue in general, but rather an x86 vs. x86-64 in specific issue. AMD estimates that this will boost performance by 5 to 15% on applications, simply by recomiling them for x86-64 (same hardware and same compiler, just x86 vs. x86-64).
As for the people who talk about "why don't we have 256-bit processors yet" though, you're dead-on as to the reason, they would quite simply be SLOW. The "128-bit" processors that you see in modern gaming consoles like the Playstation 2 are talking about VERY different bitness then in the general computing world (what they talk about is more analogous to the 128-bit or 256-bit graphics cards that most PCs use these days).