Well, there are already heat sinks that will keep even a powerful CPU cool with *VERY* low noise. It's not hard, because you have lots of room, which allows you to use heat sinks with tons of surface area, and large-diameter fans that move good amounts of air without much noise.
Even power supplies can come awfully quiet in actively-cooled models, and completely passively-cooled models are available.
Video cards, on the other hand, are to the point where they are drawing as much - and sometimes MORE - power than the host CPUs. (100+ watts in some cases). However, if the heat sink extends more than about 1/2" from the PCB (which makes it what, 3/16" above the GPU?), then people whine and moan because they lose an adjacent PCI(-E) slot. There's vastly less room for heat sink surface and fan size, so higher RPMs need to be employed.
Overall, I think that video cards are a perfect place for this to be introduced. Maybe we'll even see it make it's way to the CPU heat sinks eventually. Personally, I think that Intel would *love* to be able to suck 200+ watts out of a CPU.
It won't be used for heat sinks because heat sinks are made for idiots - stick it on with a phase-change pad, power up machines, and be done with it.
The integral heat spreaders are a good example of how things are engineered for the lowest common denominator. You can achieve significantly better cooling by prying off the heat spreader, lapping your heat sink (and in some cases, the CPU) and properly applying a thermal compound, because the thermal transfer between the die and the heat spreader isn't that good. However, in the *worst* scenario, where some idiot has screwed up the interface between the CPU and the heat sink, the greater surface area of the heat spreader makes it a significantly better design. Fewer returns = higher profits.
Buy an Eden-based machine if that's what you like, but I don't see how it's "wrong" for a computer to have liquid cooling.
For most of the history of computing, liquid cooling has been associated with the ultra-high-end computing. I once saw a picture of the pipes that carry the liquid to cool one of the NSA's supercomputers, they were 8" or 10" pipes, if I recall.
Besides, cars get faster and more powerful, televisions get larger, stereos get more powerful, houses get bigger, and cometic surgeons use more and more implants. I don't see anything atypical about computers competing for insanity, either. =)
With a decent (133 MHz, 64-bit) PCI bus providing 600-800 MB/sec bandwidth, it's going to take a good number of fast drives in RAID 0 before you even come close to filling one bus. With any other RAID level, you won't even have to worry about filling that bus.
Wal-Mart didn't shoulder those businesses, the local townspeople did. Really. Wal-Mart didn't walk in and raid their coffers, your neighbors (and perhaps YOU) made a decision to stop giving your money to the local businesses.
Myself, I don't shop at Wal-Mart if there's any way I can avoid it. I've bought something there about twice in the last year - I nearly always spend my money somewhere else.
Yes, the torque is great in diesels, but unfortunately, the soot and particulate that they put out is particularly worse to sufferers of asthma and allergies, as well as causing lung cancer. They also have their fair share of other toxic chemicals, as well.
Hydrogen burns cleanly, but there are the associated problems with efficiency in creation and delivery. There just aren't any easily-right answers.
30% efficient? Your numbers are hugely off. That might have been true waaaaaaay back in the day before switching power supplies, but it's not now. If that were true, a power supply delivering 300 watts to the computer would have to pull a kilowatt from the wall, and two computers would be enough to trip a 15-amp circuit that is so prevalent in newer construction, three computers would be much more than enough to trip a 20-amp circuit.
At normal load, most power supplies are around or above 70% efficient, primarily because the ATX 12V v2.0 specs explicitly call for a minimum efficiency of 70% at full load.
Now, note that even ultra-high-efficiency power supplies, which cost more than just double what a normal power supply costs, only specify 85% efficiency (an increase of 21%), and are reputed to save $17 per year per PC. For realistic usage of 3 hours per day, 10 cents per KW/h, 200 watt draw (which for AVERAGE usage is probably high), you would actually end up saving (365 days * 0.6 kw/h/day *.21 efficiency *.1 dollars/kilowatt/hour)=$4.60 per *year*.
Maybe it's just me, but spending an extra $100-$150 on a power supply that will save just $4.60 per year seems a bit silly.
On the other hand, purchasing a more efficient platform to begin with will save VASTLY more electricity. Replacing that with one of the efficient designs from Via would end up with a total real-world consumption closer to 40-50 watts. Switch from a CRT to an LCD, and you've dropped that from ~100 watts to ~35 watts.
Of course, if we simply increased the CAFE (Corporate Average Fuel Economy) by just five MPH, we would likely do far, FAR more good not just for the environment, but for world stability as well.
... the "dish" he was holding up looks like the microwave antennas that oh-so-many wireless companies use. Not that there's anything inherantly wrong with the technology, but so far I have yet to find a single company of that sort which actually delivered a product worth having.
Really. I've been approached by many of them, and had them pitched to be the greatest thing in the world. However, I've worked on systems connected by every one of them, and without fail, every one of their networks sucked. Low bandwidth, high latency, and poor availability was the name of the game. I even saw some of their customers with >1000ms latency... to the first-hop.
One of my friends (although how good of a friend is open for debate) used to work for one of them. He called me over and over, telling me how he got over 10 megabits/second with 10ms latency over the connection, blah, blah, blah. However, other customers of theirs had told me vastly different stories. A year later, after he no longer worked for the company, he admitted that he got such great numbers because he was sitting less than 100 feet from the transmitter.
So, maybe they'll pull it off, but if I lived in Seattle, I wouldn't hold my breath.
Yes, it's screwy. And yes, they're effectively saying "We can't make a secure OS." But at least they ARE admitting that, and mitigating the extent of the damage that can be done...
It's like an alcoholic finally admitting that he has a problem, and telling the guy at the liquor store not to sell him anything again. In an odd sort of way...
You get to live on a cruise ship in international waters, and work "below the radar", so to speak. What a great way to lie low until the heat cools off. Shoot, forget running there AFTER the feds are looking for you, it seems like a great place from which to RUN all kinds of criminal activity.
For download speed, cable is tough to beat - ComCast currently offers 3 mbits, and I think they're moving up to 4 mbits - and even 6 mbits if you pay a little extra.
The downside of Comcast is the upload speed - 384 kbits. That's more than plenty for surfing, email, gaming, etc., but if you do large uploads (I regularly sync up large file repositories between home and an office server), then a DSL offering can get the nod - IF you get a service level with a reasonably higher upload speed.
All in all, nearly everyone I know (including gamers) prefer ComCast - but those with DSL from a company with decent uptime and network connectivity are rarely saddened that they don't have cable. Those with DSL from a shoddy company quickly end up switching.
When you show me an LED lamp that will thread into a regular socket, *truly* produce as much light as a 60-watt bulb while consuming 60 watts of electricity (or less), and costs a few dollars. Until then, I'm not holding my breath.
Despite all of the hype, LEDs really aren't as efficient (in terms of lumens per watt) as a lot of people think. Their "higher efficiency" is due to the fact that it is very easy to tightly concentrate the light beam from them. For one of those keychain lights or a reading lamp, that's fine: But when you're trying to fully illuminate a 600 square foot room, that just doesn't cut it.
Also, they tend to be very efficient at very low power consumptions, but as you increase the power, the efficiency drops VERY rapidly. If you need a one-watt bulb, that's alright, but when you're replacing 40-, 60-, or 100-watt bulbs, that will bite you in the butt.
So, let's say that you get the efficiency and coverage things figured out, and you *truly* come out ahead of not only incandescent lamps, but fluorescent lamps as well (don't hold your breath). You're still stuck needing something like 300-600 LEDs to equal a 60-watt lamp. You'd better be making those whiz-bang LEDs at a twentieth of a cent to be anywhere near economical. Besides, who wants a 500-LED array hanging off of their ceiling? Yes, it may appeal to the college-dorm geek, but a foot-wide protrusion isn't going to appeal to the rest of society.
Now, some potential benefits: LEDs can be easily and efficiently dimmed, unlike fluorescents, and that is a larger obstacle than many people realize. The color can also be adjusted, and poor color quality is a HUGE obstacle to the adoption of fluorescents.
Don't get me wrong, I'd love to see LEDs become what people say they will. But when you get through the hype, BS, and semi-dishonestly measured ratings ("I'll take this 100 milliwatt LED and rate it's efficiency at 1 milliwatt to boost the score...."), they're just not nearly as close to mainstream adoption as a general lighting source. Spot-lighting, sure. General illumination, no.
...getting what you pay for. Most people that I know that use cheap registrars end up with hastles like these.
I get domains for $10 per year with great service, but that's from BulkRegister - and handle a lot of domains.
That is not to say (of course) that no cheap registrar will give you good service, neither do I imply that an expensive registrar will give you better service. However, when you pick a cheap, relatively unheard of registrar, you ARE taking your chances.
Why can't they dip the keyboards in alcohol, let them dry, and plug them back in?
My experience with cleaning "dirty" keyboards started waaaay back in the day when I spilled a Big Gulp of Coca-Cola into my Commodore-64, the screen instantly went black, and I had one hour until my mother got home. The Commodore got disassembled, washed, cleaned, dried, and reassembled before she got home - there's nothing like adrenaline to make you work quickly!
I've had my card's service "interrupted" twice because of suspicious activity. The first time, it prevented someone from charging thousands of dollars to my account, so the second time, I wasn't at all upset that it had been interrupted even though the activity was perfectly normal.
Domain names are cheap, and it's not hard at all to employ some CGI and HTTP tricks to conjure up 1,000 domains with (seemingly) unique content on thousands of pages each - while appearing to the search engine as static HTML, not dynamic... each of those linking to your page.
Did you read the article? Basically he says that extra cache gives everybody breathing room, but for the foreseeable future it's better to invest in making maximum use out of what will be the norm...multiple processors (cores), and multiple threads is the way to do it, as all modern operating systems have some provision for kernel level threads.
The future is multithreaded; single threaded apps will simply plateau at some point.
Like I said, just because someone wrote an article doesn't make it true. To use one of the examples (video games), look at the development of Doom 3 - which was (at the time of release) the most advanced video game from arguably the best video game coder in the world. It was originally intended to develop as a multithreaded application to take advantage of multiple CPUs. In the end, John Carmack decided that it was far much more trouble than it was worth.
There you go: Real world evidence that (at least for the proximal future) it's just not worth it to parallelize video games.
What games are you playing that are dependent on clock speed, but don't use multiple threads now?
None. But watch how much CPU time each of those threads get. It's *very* rare for one of my dual-CPU machines to get more than about 55% CPU utilization, because none of the current games actually *use* two CPUs. With Q3 there was a way to get certain video drivers to run on the second CPU, which resulted in net speed increases of only about 10%.
At this point, the obvious rebuttle would be... "but people will want more in the future."
You're right, they will - and that won't be the first time we've been in that boat. Back in the days of software rendering, *no* CPU/video card combination was quick enough to do much. So, companied developped video cards to offload the work, and not just the FPS but also the quality jumped AMAZINGLY - much, much, MUCH more than could have been achieved by dual CPUs.
As time has gone on, video card performance has increased astoundingly - and the overall graphical display of the video game has arguably been influenced much more by that than by the CPU. Even something like an Athlon 1600 with a good video card will give a better overall playing experience at Doom3 than an Athlon64 with a lesser card.
Besides, the video game market is always getting more and more competitive, with costs going up and up - and buggy code being the norm. Adding the increased code complexity and cost of truly supporting multiprocessing is something that video game makers really don't want to deal with.
So, right now, what's the big bottleneck? According to the game industry, physics. According to the recent buzz, a CPU can only handle about 200 models. So, you want to parrallelize the task? Sure, write your code (which will NOT scale linearly), and spend the bucks on a 4-way machine. Even assuming you don't run into bandwidth bottlenecks (which you likely *will* if you aren't running something like a 4-way opteron with seperate memory subsystems), then you're looking at a total of about 600-700 models. On the other hand, the add-in card that has been announced can handle something like 20,000 models, and probably won't cost more than $100-$200. That's roughly 30 times better, for what will likely be about 1/10th of the price. So much for scaling with CPUs.
Look at video encoding/decoding. Rather than try and scale with CPUs, inexpensive hardware improvements blow away anything you could do on a multi-CPU system.
So, once the physics cards become more mainstream (or, as is likely the case, become integrated into video cards or motherboard chipsets), then in time, something else will become the bottleneck. And, without my trusty crystal ball, I can't *guarantee* that it won't be handled by the CPU, but I think it's pretty likely that it won't.
After all, a CPU is actually a pretty crude tool. They're designed to be able to do a wide va
Extra-hot baths will cook your boys worse than most hot tubs, and still aren't all *that* bad. If you look at all of the things that affect your fertility (even just wearing "tighty-whitey" underwear), you'd think that the human race was doomed, but experience proves otherwise...
Just because someone wrote an article doesn't mean it's true.
Even in the P4-vs-A64 area, where the A64 *usually* wins, there are still applications where the P4 wins just because of the higher clock speed.
Getting more specific to your example, take a dual 1GHz machine and a single 2 GHz machine, and go recompress a DVD or play a video game. See which is faster.
Now I know what my problem was! Stupid me! I spent as much money as possible!
As obvious as that sounds, you'd be amazed at how many companies did exactly that. As an example, the chief competitor of the company I work for spent over three million dollars on high-end Sun infrastructure, and did all of their coding in Java. We spent $25,000 on white boxes running Linux, and coded in Perl. Their site was continually bogged down and slow, our site could handle MUCH more traffic very speedily.
When we were entertaining potential investors, every one of them would come in, and immediately say that we had to switch everything to Oracle on Sun or IBM hardware - usually getting into the millions of dollars in costs. Here's the kicker: The reason was not because of any actual advantage, but just because investment banks like to see those big names.
Slashdot Mirror
Well, there are already heat sinks that will keep even a powerful CPU cool with *VERY* low noise. It's not hard, because you have lots of room, which allows you to use heat sinks with tons of surface area, and large-diameter fans that move good amounts of air without much noise.
Even power supplies can come awfully quiet in actively-cooled models, and completely passively-cooled models are available.
Video cards, on the other hand, are to the point where they are drawing as much - and sometimes MORE - power than the host CPUs. (100+ watts in some cases). However, if the heat sink extends more than about 1/2" from the PCB (which makes it what, 3/16" above the GPU?), then people whine and moan because they lose an adjacent PCI(-E) slot. There's vastly less room for heat sink surface and fan size, so higher RPMs need to be employed.
Overall, I think that video cards are a perfect place for this to be introduced. Maybe we'll even see it make it's way to the CPU heat sinks eventually. Personally, I think that Intel would *love* to be able to suck 200+ watts out of a CPU.
steve
It won't be used for heat sinks because heat sinks are made for idiots - stick it on with a phase-change pad, power up machines, and be done with it.
The integral heat spreaders are a good example of how things are engineered for the lowest common denominator. You can achieve significantly better cooling by prying off the heat spreader, lapping your heat sink (and in some cases, the CPU) and properly applying a thermal compound, because the thermal transfer between the die and the heat spreader isn't that good. However, in the *worst* scenario, where some idiot has screwed up the interface between the CPU and the heat sink, the greater surface area of the heat spreader makes it a significantly better design. Fewer returns = higher profits.
steve
Buy an Eden-based machine if that's what you like, but I don't see how it's "wrong" for a computer to have liquid cooling.
For most of the history of computing, liquid cooling has been associated with the ultra-high-end computing. I once saw a picture of the pipes that carry the liquid to cool one of the NSA's supercomputers, they were 8" or 10" pipes, if I recall.
Besides, cars get faster and more powerful, televisions get larger, stereos get more powerful, houses get bigger, and cometic surgeons use more and more implants. I don't see anything atypical about computers competing for insanity, either. =)
steve
With a decent (133 MHz, 64-bit) PCI bus providing 600-800 MB/sec bandwidth, it's going to take a good number of fast drives in RAID 0 before you even come close to filling one bus. With any other RAID level, you won't even have to worry about filling that bus.
steve
Wal-Mart didn't shoulder those businesses, the local townspeople did. Really. Wal-Mart didn't walk in and raid their coffers, your neighbors (and perhaps YOU) made a decision to stop giving your money to the local businesses.
Myself, I don't shop at Wal-Mart if there's any way I can avoid it. I've bought something there about twice in the last year - I nearly always spend my money somewhere else.
steve
Yes, the torque is great in diesels, but unfortunately, the soot and particulate that they put out is particularly worse to sufferers of asthma and allergies, as well as causing lung cancer. They also have their fair share of other toxic chemicals, as well.
Hydrogen burns cleanly, but there are the associated problems with efficiency in creation and delivery. There just aren't any easily-right answers.
steve
30% efficient? Your numbers are hugely off. That might have been true waaaaaaay back in the day before switching power supplies, but it's not now. If that were true, a power supply delivering 300 watts to the computer would have to pull a kilowatt from the wall, and two computers would be enough to trip a 15-amp circuit that is so prevalent in newer construction, three computers would be much more than enough to trip a 20-amp circuit.
At normal load, most power supplies are around or above 70% efficient, primarily because the ATX 12V v2.0 specs explicitly call for a minimum efficiency of 70% at full load.
Now, note that even ultra-high-efficiency power supplies, which cost more than just double what a normal power supply costs, only specify 85% efficiency (an increase of 21%), and are reputed to save $17 per year per PC. For realistic usage of 3 hours per day, 10 cents per KW/h, 200 watt draw (which for AVERAGE usage is probably high), you would actually end up saving (365 days * 0.6 kw/h/day *
Maybe it's just me, but spending an extra $100-$150 on a power supply that will save just $4.60 per year seems a bit silly.
On the other hand, purchasing a more efficient platform to begin with will save VASTLY more electricity. Replacing that with one of the efficient designs from Via would end up with a total real-world consumption closer to 40-50 watts. Switch from a CRT to an LCD, and you've dropped that from ~100 watts to ~35 watts.
Of course, if we simply increased the CAFE (Corporate Average Fuel Economy) by just five MPH, we would likely do far, FAR more good not just for the environment, but for world stability as well.
steve
... the "dish" he was holding up looks like the microwave antennas that oh-so-many wireless companies use. Not that there's anything inherantly wrong with the technology, but so far I have yet to find a single company of that sort which actually delivered a product worth having.
Really. I've been approached by many of them, and had them pitched to be the greatest thing in the world. However, I've worked on systems connected by every one of them, and without fail, every one of their networks sucked. Low bandwidth, high latency, and poor availability was the name of the game. I even saw some of their customers with >1000ms latency... to the first-hop.
One of my friends (although how good of a friend is open for debate) used to work for one of them. He called me over and over, telling me how he got over 10 megabits/second with 10ms latency over the connection, blah, blah, blah. However, other customers of theirs had told me vastly different stories. A year later, after he no longer worked for the company, he admitted that he got such great numbers because he was sitting less than 100 feet from the transmitter.
So, maybe they'll pull it off, but if I lived in Seattle, I wouldn't hold my breath.
steve
Actors swing sticks around, and special effects groups add in the light later. :)
steve
Why in the world does Apple think that if I want to watch a Batman trailer, that I need iTunes?
This is exactly the sort of thing that Micro$oft would get raked across the coals for, but just because Apple beat them to it, it seems to be A-OK.
steve
Yes, it's screwy. And yes, they're effectively saying "We can't make a secure OS." But at least they ARE admitting that, and mitigating the extent of the damage that can be done...
It's like an alcoholic finally admitting that he has a problem, and telling the guy at the liquor store not to sell him anything again. In an odd sort of way...
That reminds me of the age-old joke, did you hear about the constipated mathematician? He worked it out with a pencil an paper. (ba-ding!)
Why, are you that anxious to see "Meet the Binxes?"
steve
You get to live on a cruise ship in international waters, and work "below the radar", so to speak. What a great way to lie low until the heat cools off. Shoot, forget running there AFTER the feds are looking for you, it seems like a great place from which to RUN all kinds of criminal activity.
steve
For download speed, cable is tough to beat - ComCast currently offers 3 mbits, and I think they're moving up to 4 mbits - and even 6 mbits if you pay a little extra.
The downside of Comcast is the upload speed - 384 kbits. That's more than plenty for surfing, email, gaming, etc., but if you do large uploads (I regularly sync up large file repositories between home and an office server), then a DSL offering can get the nod - IF you get a service level with a reasonably higher upload speed.
All in all, nearly everyone I know (including gamers) prefer ComCast - but those with DSL from a company with decent uptime and network connectivity are rarely saddened that they don't have cable. Those with DSL from a shoddy company quickly end up switching.
steve
When you show me an LED lamp that will thread into a regular socket, *truly* produce as much light as a 60-watt bulb while consuming 60 watts of electricity (or less), and costs a few dollars. Until then, I'm not holding my breath.
Despite all of the hype, LEDs really aren't as efficient (in terms of lumens per watt) as a lot of people think. Their "higher efficiency" is due to the fact that it is very easy to tightly concentrate the light beam from them. For one of those keychain lights or a reading lamp, that's fine: But when you're trying to fully illuminate a 600 square foot room, that just doesn't cut it.
Also, they tend to be very efficient at very low power consumptions, but as you increase the power, the efficiency drops VERY rapidly. If you need a one-watt bulb, that's alright, but when you're replacing 40-, 60-, or 100-watt bulbs, that will bite you in the butt.
So, let's say that you get the efficiency and coverage things figured out, and you *truly* come out ahead of not only incandescent lamps, but fluorescent lamps as well (don't hold your breath). You're still stuck needing something like 300-600 LEDs to equal a 60-watt lamp. You'd better be making those whiz-bang LEDs at a twentieth of a cent to be anywhere near economical. Besides, who wants a 500-LED array hanging off of their ceiling? Yes, it may appeal to the college-dorm geek, but a foot-wide protrusion isn't going to appeal to the rest of society.
Now, some potential benefits: LEDs can be easily and efficiently dimmed, unlike fluorescents, and that is a larger obstacle than many people realize. The color can also be adjusted, and poor color quality is a HUGE obstacle to the adoption of fluorescents.
Don't get me wrong, I'd love to see LEDs become what people say they will. But when you get through the hype, BS, and semi-dishonestly measured ratings ("I'll take this 100 milliwatt LED and rate it's efficiency at 1 milliwatt to boost the score...."), they're just not nearly as close to mainstream adoption as a general lighting source. Spot-lighting, sure. General illumination, no.
steve
I get domains for $10 per year with great service, but that's from BulkRegister - and handle a lot of domains.
That is not to say (of course) that no cheap registrar will give you good service, neither do I imply that an expensive registrar will give you better service. However, when you pick a cheap, relatively unheard of registrar, you ARE taking your chances.
steve
Why can't they dip the keyboards in alcohol, let them dry, and plug them back in?
My experience with cleaning "dirty" keyboards started waaaay back in the day when I spilled a Big Gulp of Coca-Cola into my Commodore-64, the screen instantly went black, and I had one hour until my mother got home. The Commodore got disassembled, washed, cleaned, dried, and reassembled before she got home - there's nothing like adrenaline to make you work quickly!
steve
What heat sink are you using, and what voltage are you giving the chip?
steve
I've had my card's service "interrupted" twice because of suspicious activity. The first time, it prevented someone from charging thousands of dollars to my account, so the second time, I wasn't at all upset that it had been interrupted even though the activity was perfectly normal.
steve
Domain names are cheap, and it's not hard at all to employ some CGI and HTTP tricks to conjure up 1,000 domains with (seemingly) unique content on thousands of pages each - while appearing to the search engine as static HTML, not dynamic... each of those linking to your page.
steve
Did you read the article? Basically he says that extra cache gives everybody breathing room, but for the foreseeable future it's better to invest in making maximum use out of what will be the norm...multiple processors (cores), and multiple threads is the way to do it, as all modern operating systems have some provision for kernel level threads.
The future is multithreaded; single threaded apps will simply plateau at some point.
Like I said, just because someone wrote an article doesn't make it true. To use one of the examples (video games), look at the development of Doom 3 - which was (at the time of release) the most advanced video game from arguably the best video game coder in the world. It was originally intended to develop as a multithreaded application to take advantage of multiple CPUs. In the end, John Carmack decided that it was far much more trouble than it was worth.
There you go: Real world evidence that (at least for the proximal future) it's just not worth it to parallelize video games.
What games are you playing that are dependent on clock speed, but don't use multiple threads now?
None. But watch how much CPU time each of those threads get. It's *very* rare for one of my dual-CPU machines to get more than about 55% CPU utilization, because none of the current games actually *use* two CPUs. With Q3 there was a way to get certain video drivers to run on the second CPU, which resulted in net speed increases of only about 10%.
At this point, the obvious rebuttle would be... "but people will want more in the future."
You're right, they will - and that won't be the first time we've been in that boat. Back in the days of software rendering, *no* CPU/video card combination was quick enough to do much. So, companied developped video cards to offload the work, and not just the FPS but also the quality jumped AMAZINGLY - much, much, MUCH more than could have been achieved by dual CPUs.
As time has gone on, video card performance has increased astoundingly - and the overall graphical display of the video game has arguably been influenced much more by that than by the CPU. Even something like an Athlon 1600 with a good video card will give a better overall playing experience at Doom3 than an Athlon64 with a lesser card.
Besides, the video game market is always getting more and more competitive, with costs going up and up - and buggy code being the norm. Adding the increased code complexity and cost of truly supporting multiprocessing is something that video game makers really don't want to deal with.
So, right now, what's the big bottleneck? According to the game industry, physics. According to the recent buzz, a CPU can only handle about 200 models. So, you want to parrallelize the task? Sure, write your code (which will NOT scale linearly), and spend the bucks on a 4-way machine. Even assuming you don't run into bandwidth bottlenecks (which you likely *will* if you aren't running something like a 4-way opteron with seperate memory subsystems), then you're looking at a total of about 600-700 models. On the other hand, the add-in card that has been announced can handle something like 20,000 models, and probably won't cost more than $100-$200. That's roughly 30 times better, for what will likely be about 1/10th of the price. So much for scaling with CPUs.
Look at video encoding/decoding. Rather than try and scale with CPUs, inexpensive hardware improvements blow away anything you could do on a multi-CPU system.
So, once the physics cards become more mainstream (or, as is likely the case, become integrated into video cards or motherboard chipsets), then in time, something else will become the bottleneck. And, without my trusty crystal ball, I can't *guarantee* that it won't be handled by the CPU, but I think it's pretty likely that it won't.
After all, a CPU is actually a pretty crude tool. They're designed to be able to do a wide va
Extra-hot baths will cook your boys worse than most hot tubs, and still aren't all *that* bad. If you look at all of the things that affect your fertility (even just wearing "tighty-whitey" underwear), you'd think that the human race was doomed, but experience proves otherwise...
steve
Just because someone wrote an article doesn't mean it's true.
Even in the P4-vs-A64 area, where the A64 *usually* wins, there are still applications where the P4 wins just because of the higher clock speed.
Getting more specific to your example, take a dual 1GHz machine and a single 2 GHz machine, and go recompress a DVD or play a video game. See which is faster.
steve
Now I know what my problem was! Stupid me! I spent as much money as possible!
As obvious as that sounds, you'd be amazed at how many companies did exactly that. As an example, the chief competitor of the company I work for spent over three million dollars on high-end Sun infrastructure, and did all of their coding in Java. We spent $25,000 on white boxes running Linux, and coded in Perl. Their site was continually bogged down and slow, our site could handle MUCH more traffic very speedily.
When we were entertaining potential investors, every one of them would come in, and immediately say that we had to switch everything to Oracle on Sun or IBM hardware - usually getting into the millions of dollars in costs. Here's the kicker: The reason was not because of any actual advantage, but just because investment banks like to see those big names.
steve