Perhaps. I guess we won't know until someone actually defines the theory and puts it into practice.
At one extreme, you'll have the strictly serial code, a sequence of instructions that all depend on the preceding instructions/results. Splitting this into separate threads would be horrendous for performance, because you'd have to be continuously shuttling data between the registers of one core to the other. At the opposite extreme will be code with very few sequential dependencies, something that could easily have been written as explicitly parallel if the programmer wanted to. I suspect many applications will fall somewhere in between, but the majority of code will be serial.
That's an observation based on common practice today, of course. The basic definition of a "program" is a *sequence* of instructions, so that's how we're taught to think about computing. That's also how we relate to life though: in terms of timelines. Even if we can train ourselves to think of programs as collections of independent events, in order to interface them to the real world, we have to serialize them.
That assumes the program has to interface to the real world while it runs. E.g., a video codec that is decoding a bitstream for display on a framebuffer in realtime, probably ought to do the decoding in scanline order. But a batch-mode codec that's just transcoding from one file format to another can process the data in an arbitrary order. If the memory hardware had a quirk that made some other order most optimal, you could break up the processing to accomodate that... I think this illustrates the divide between "theory" and "reality" pretty well - programs that must operate in real time, or provide real time response to user input, probably won't be able to take much advantage of the theoretical merge/split optimizations.
You have to look deeper to understand that statement. Games inherently involve multiple tasks running simultaneously. Look at every arcade game made from the 1980s onward and you'll see a cluster of processors working in concert. In a modern game every enemy has its own independent AI controlling it. The reality is that it *simplifies* game development to make those AIs run in their own thread. It simplifies game development to have a smart dedicated audio processor that you just periodically tell "play this list of sounds (i.e., an audio program) at time t" because your main CPU can keep focused on processing the game action. It simplifies game development to have a smart dedicated graphics object processor that you tell "draw this list of objects/textures (i.e., a video program) and move it there, and tell me about any collisions."
Look at MAME and look at how much effort it takes to make a single Pentium running at however-many gigahertz play a 1990s era arcade game smoothly. The fact is that making a single CPU do everything *sucks* for responsiveness. The secret to the playability of those games is multiprocessing, multiple CPUs splitting up the tasks amongst themselves. Jumping through hoops to make all of that stuff play nicely on a PC is ridiculous. (And yes, I was in the arcade game business in the 80s and 90s, I know a bit about how game development on PCs and on arcade consoles works.)
The fact is that good, responsive games inherently need to be multi-threaded. Running them on multiple cores is a natural path to improving their responsiveness.
None of this will mean anything to desktop PCs, this is something that only the HPC crowd would need. First you have to start from the basic assumption that you have an algorithm that you only know how to code in serial fashion. And, you need it to run as fast as possible. You would like to run it on a single 6 GHz CPU but all you have on hand is a pair of 3 GHZ CPUs, and this algorithm of yours can't be partitioned for parallel execution.
Intuition tells me that if a competent programmer with complete access to information about and understanding of a particular algorithm can't figure out how to effectively parallelize it, there's no way a hardware state machine will do a better job of it.
Again, only the HPC crowd would ever run this way. E.g., you could boot DOS on your 3 GHz Opteron and it would be several orders of magnitude faster than Windows, but it would only run one thing at a time. Very few desktop PC users can live with single-tasking today, otherwise more people would still be using DOS. And multiple cores are better for running multiple independent tasks than a single core, so this whole pursuit is only useful if you want to use a a monster machine to solve a single problem at a time (e.g. Deep Thought).
If they're spitting out linear PCM then the audio can of course go to.02ms resolution. Probably there's no reason not to use PCM for such small sound clips.
If they were using MP3, then there would be a minimum of 4.3ms resolution for a short block, or 13ms for a long block. It's not clear to me whether an encoder would be smart enough to switch to a short block for these sounds.
As a fiddle player I regularly play ornaments with sequences of grace notes of about 20ms each (a standard Irish 5-note roll or a fast bow triplet played at about 130bpm); they tend to blur on MP3 recordings. One of many reasons why I avoid MP3s...
That seems pretty unlikely, such a wide variance would make movies unwatchable on many machines. A variance of even a few milliseconds would completely destroy lipsyncing. I think you meant "hundreds of microseconds."
Just because you see two guys fighting doesn't automatically mean one of them is a good guy and the other's a bad guy. They can very easily both be bad guys.
Easily. Of course, the gear running on wall-warts probably all run at different voltages. I suppose if you knew it was all running at 5, 7 or 12VDC you could just split off some lines from your ATX power supply to everything else.
I used to do a lot of (2) but nowadays I try not to do so in public. On the OpenLDAP project we pretty much have to do (1) for all contributions; it's rare to get something contributed that conforms to the documented coding style.
Funny you should mention insurance. Personally, I think you *should* only have to pay them once. I pay for a year's worth of insurance, nothing happens, I never file a claim, I've just paid them for absolutely nothing. They should automatically carry me for the next year, ad infinitum, until I actually have to file a claim, and then we can talk about renewal fees.
As for "make musicians work for a living" - most of them *do*. It's only the few "megastars" backed by Large Greedy Record Labels that get to spin out crap and get it gold plated. Yes, cut copyright back to the original 14 years. But get the damn labels out of the way. As posted earlier, disallow the assignment of copyrights, don't allow corporations to own copyrights or patents. (I'll yield the point on trademarks, they really have nothing to do with individuals. By their nature they only make sense being owned by businesses.)
Apparently you've never heard of the word "loyalty." Here's a better term to remember though: "enlightened self interest." If you screw the company you're working for, they can go under, and then you'll be out of a job. Or not. Or you can simply make such a bad reputation for yourself in your industry that you make yourself unemployable.
The fact is it's stupid to play games like that, and employers who understand that people are not dispensable will profit while employers who don't will eventually disappear.
As an artist, I want music to be *relevant to a listener*. When you have 15,000 songs on your iPod, you're not my audience any more, you really are Just a Consumer.
There's what, 6 billion people on this planet now? Nobody will ever write a song that all 6 billion people love. There will always be a market for good music of multiple genres, because people are different and have different tastes. The fact that contemporary music today hits Platinum sales records is an extreme anomaly, directly resulting from too much big business marketing. The reality is that no song is so good and so basic that it's deeply relevant to billions of people, because all those people's lives are all so different from each other.
That's some pretty narrow-minded thinking. Movies like Star Wreck prove you false. The only thing that will change in the way large projects are completed is that only dedicated artists who actually love their craft will work in a given industry. People will collaborate on movies solely because they believe in the story and the storytelling process, and the sharks, posers, and wannabes will no longer exist. They'll be off pursuing other non-parasitic careers of their own.
You're thinking in isolation, instead of globally. If the entire system worked that way, with all companies playing by the same rules, then life would go on, and companies would pay employees *precisely* because their employees create value.
As I've always said, Intellect is more important than Intellectual Property. When you produce good work, you should be paid for your effort invested, that's what makes you uniquely valuable. The product you create is just a commodity, meaningless.
I'm following you OK on using alcohol, but solar panels on a mainstream automobile are pretty futile. A decent amorphous silicon cell can generate about 4 watts per square foot in full sunlight. Just as a reminder, there's 760 watts in 1 horsepower. Even with the car completely covered in panels there's not enough surface area to generate a meaningful amount of electricity for propulsion (and of course, with the car completely covered, on average 2/3 to 3/4 of the panels will be in shadow at any given time). At best it will let you run your electric accessories for free but even a lot of car stereos will draw more than the panels can produce.
Kind of echoing bi_boy. Great post. I drive a Probe GTS, the last of that model line. I wish Ford had not killed it off, but clearly the only reason it's such a fun car is due to the Mazda chassis and powertrain. Mazda's "Zoom Zoom" ad campaign sounds kinda hokey, but the reality is that they really understand what it takes to design a car that is satisfying to drive, in ways Ford has never understood.
Mazda's Renesis engine in the RX-8 is a decent improvement over the previous versions. I must say they've got the coolest designs around (in the "affordable" range; a Ford GT would be pretty cool too) and I'd be all over a new RX-7 if they bring it back.
No, a revoked cert *is* the solution. There are many possible reasons why a cert may need to be revoked (e.g., issuer got hacked, policy change, name change, etc.) so just locking down the process of issuing certs is not a solution. The big problem here (re: slowness of CRL checking) is that X.509 certificates are intended to be part of a *distributed* authentication system. You're not supposed to have the entire world dependent on a couple of root-level CAs, they're supposed to delegate authority to widely distributed sites. Just like DNS would fall over dead if everyone always had to start their queries at the root nameservers before getting anywhere. The fact that most companies are ignorant and just buy a couple server certs from Verisign is part of the problem here. If there were more intermediate CAs distributed around the net that could handle the CRL distribution load, then verifying cert validity in realtime would be a lot faster.
According to this article http://www.rpi.edu/~danony/Research/Pyro.htm the 200,000eV was when configured to produce X-rays. The neutron generator yielded neutrons at 2.5MeV. Still a tiny amount of energy of course...
But yeah, you can get a lot of mileage from having dedicated coprocessors (including dedicated I/O processors). Something that everyone except Apple and commodity PC makers seems to already understand. Of course, there are still limits to that, and it's always nice to have a faster central processor.
If it's true that these other electronics companies' products aren't fully conforming to the spec, then it should be simple to objectively prove their non-conformance and get them to change it. Someone like CableLabs ought to be certifying each company's implementation before allowing them to slap on a "CableCard compatible" logo on their products and literature. The fact that you're seeing compatibility problems means either the spec is broken, or the certification process is broken. The basic idea, freeing viewers from the inconvenience of set top boxes, was valid.
re: recovering costs - why should the majority of cable subscribers pay the costs incurred by a few stupid customers? If I subscribe and never need to call for service or troubleshooting, why should I be paying for the cost to repair some idiot's equipment down the street? From my perspective, you're just milking customers, because I never see any benefit from the charges I'm paying for.
Personally, I've already downloaded all of the CableCard 2.0 specs. Unfortunately the 1.0 specs are no longer freely available. But anyway, assuming CableCard 2.0 goes anywhere, it will be a simple matter for a 3rd party hacker to implement the same protocols purely in software, certification be damned. At that point, the cards become become irrelevant...
Slashdot Mirror
Perhaps. I guess we won't know until someone actually defines the theory and puts it into practice.
At one extreme, you'll have the strictly serial code, a sequence of instructions that all depend on the preceding instructions/results. Splitting this into separate threads would be horrendous for performance, because you'd have to be continuously shuttling data between the registers of one core to the other. At the opposite extreme will be code with very few sequential dependencies, something that could easily have been written as explicitly parallel if the programmer wanted to. I suspect many applications will fall somewhere in between, but the majority of code will be serial.
That's an observation based on common practice today, of course. The basic definition of a "program" is a *sequence* of instructions, so that's how we're taught to think about computing. That's also how we relate to life though: in terms of timelines. Even if we can train ourselves to think of programs as collections of independent events, in order to interface them to the real world, we have to serialize them.
That assumes the program has to interface to the real world while it runs. E.g., a video codec that is decoding a bitstream for display on a framebuffer in realtime, probably ought to do the decoding in scanline order. But a batch-mode codec that's just transcoding from one file format to another can process the data in an arbitrary order. If the memory hardware had a quirk that made some other order most optimal, you could break up the processing to accomodate that... I think this illustrates the divide between "theory" and "reality" pretty well - programs that must operate in real time, or provide real time response to user input, probably won't be able to take much advantage of the theoretical merge/split optimizations.
You have to look deeper to understand that statement. Games inherently involve multiple tasks running simultaneously. Look at every arcade game made from the 1980s onward and you'll see a cluster of processors working in concert. In a modern game every enemy has its own independent AI controlling it. The reality is that it *simplifies* game development to make those AIs run in their own thread. It simplifies game development to have a smart dedicated audio processor that you just periodically tell "play this list of sounds (i.e., an audio program) at time t" because your main CPU can keep focused on processing the game action. It simplifies game development to have a smart dedicated graphics object processor that you tell "draw this list of objects/textures (i.e., a video program) and move it there, and tell me about any collisions."
Look at MAME and look at how much effort it takes to make a single Pentium running at however-many gigahertz play a 1990s era arcade game smoothly. The fact is that making a single CPU do everything *sucks* for responsiveness. The secret to the playability of those games is multiprocessing, multiple CPUs splitting up the tasks amongst themselves. Jumping through hoops to make all of that stuff play nicely on a PC is ridiculous. (And yes, I was in the arcade game business in the 80s and 90s, I know a bit about how game development on PCs and on arcade consoles works.)
The fact is that good, responsive games inherently need to be multi-threaded. Running them on multiple cores is a natural path to improving their responsiveness.
None of this will mean anything to desktop PCs, this is something that only the HPC crowd would need. First you have to start from the basic assumption that you have an algorithm that you only know how to code in serial fashion. And, you need it to run as fast as possible. You would like to run it on a single 6 GHz CPU but all you have on hand is a pair of 3 GHZ CPUs, and this algorithm of yours can't be partitioned for parallel execution.
Intuition tells me that if a competent programmer with complete access to information about and understanding of a particular algorithm can't figure out how to effectively parallelize it, there's no way a hardware state machine will do a better job of it.
Again, only the HPC crowd would ever run this way. E.g., you could boot DOS on your 3 GHz Opteron and it would be several orders of magnitude faster than Windows, but it would only run one thing at a time. Very few desktop PC users can live with single-tasking today, otherwise more people would still be using DOS. And multiple cores are better for running multiple independent tasks than a single core, so this whole pursuit is only useful if you want to use a a monster machine to solve a single problem at a time (e.g. Deep Thought).
If they're spitting out linear PCM then the audio can of course go to .02ms resolution. Probably there's no reason not to use PCM for such small sound clips.
If they were using MP3, then there would be a minimum of 4.3ms resolution for a short block, or 13ms for a long block. It's not clear to me whether an encoder would be smart enough to switch to a short block for these sounds.
As a fiddle player I regularly play ornaments with sequences of grace notes of about 20ms each (a standard Irish 5-note roll or a fast bow triplet played at about 130bpm); they tend to blur on MP3 recordings. One of many reasons why I avoid MP3s...
That seems pretty unlikely, such a wide variance would make movies unwatchable on many machines. A variance of even a few milliseconds would completely destroy lipsyncing. I think you meant "hundreds of microseconds."
What a sobering thought. "We've invented a space heater that produces computations as an operational byproduct."
Of course, IBM had multicores years ago, so AMD wasn't really the innovator on that front.
Just because you see two guys fighting doesn't automatically mean one of them is a good guy and the other's a bad guy. They can very easily both be bad guys.
Easily. Of course, the gear running on wall-warts probably all run at different voltages. I suppose if you knew it was all running at 5, 7 or 12VDC you could just split off some lines from your ATX power supply to everything else.
I used to do a lot of (2) but nowadays I try not to do so in public. On the OpenLDAP project we pretty much have to do (1) for all contributions; it's rare to get something contributed that conforms to the documented coding style.
Funny you should mention insurance. Personally, I think you *should* only have to pay them once. I pay for a year's worth of insurance, nothing happens, I never file a claim, I've just paid them for absolutely nothing. They should automatically carry me for the next year, ad infinitum, until I actually have to file a claim, and then we can talk about renewal fees.
As for "make musicians work for a living" - most of them *do*. It's only the few "megastars" backed by Large Greedy Record Labels that get to spin out crap and get it gold plated. Yes, cut copyright back to the original 14 years. But get the damn labels out of the way. As posted earlier, disallow the assignment of copyrights, don't allow corporations to own copyrights or patents. (I'll yield the point on trademarks, they really have nothing to do with individuals. By their nature they only make sense being owned by businesses.)
Apparently you've never heard of the word "loyalty." Here's a better term to remember though: "enlightened self interest." If you screw the company you're working for, they can go under, and then you'll be out of a job. Or not. Or you can simply make such a bad reputation for yourself in your industry that you make yourself unemployable.
The fact is it's stupid to play games like that, and employers who understand that people are not dispensable will profit while employers who don't will eventually disappear.
As an artist, I want music to be *relevant to a listener*. When you have 15,000 songs on your iPod, you're not my audience any more, you really are Just a Consumer.
There's what, 6 billion people on this planet now? Nobody will ever write a song that all 6 billion people love. There will always be a market for good music of multiple genres, because people are different and have different tastes. The fact that contemporary music today hits Platinum sales records is an extreme anomaly, directly resulting from too much big business marketing. The reality is that no song is so good and so basic that it's deeply relevant to billions of people, because all those people's lives are all so different from each other.
That's some pretty narrow-minded thinking. Movies like Star Wreck prove you false. The only thing that will change in the way large projects are completed is that only dedicated artists who actually love their craft will work in a given industry. People will collaborate on movies solely because they believe in the story and the storytelling process, and the sharks, posers, and wannabes will no longer exist. They'll be off pursuing other non-parasitic careers of their own.
You're thinking in isolation, instead of globally. If the entire system worked that way, with all companies playing by the same rules, then life would go on, and companies would pay employees *precisely* because their employees create value.
As I've always said, Intellect is more important than Intellectual Property. When you produce good work, you should be paid for your effort invested, that's what makes you uniquely valuable. The product you create is just a commodity, meaningless.
One can only hope. The entire auto market got really boring when compact sports cars succumbed to the SUV craze. I think I'd get in line for a Kabura.
I'm following you OK on using alcohol, but solar panels on a mainstream automobile are pretty futile. A decent amorphous silicon cell can generate about 4 watts per square foot in full sunlight. Just as a reminder, there's 760 watts in 1 horsepower. Even with the car completely covered in panels there's not enough surface area to generate a meaningful amount of electricity for propulsion (and of course, with the car completely covered, on average 2/3 to 3/4 of the panels will be in shadow at any given time). At best it will let you run your electric accessories for free but even a lot of car stereos will draw more than the panels can produce.
Kind of echoing bi_boy. Great post. I drive a Probe GTS, the last of that model line. I wish Ford had not killed it off, but clearly the only reason it's such a fun car is due to the Mazda chassis and powertrain. Mazda's "Zoom Zoom" ad campaign sounds kinda hokey, but the reality is that they really understand what it takes to design a car that is satisfying to drive, in ways Ford has never understood.
Not all rotary engines suffer from these problems.
http://www.freedom-motors.com/techno.html
Mazda's Renesis engine in the RX-8 is a decent improvement over the previous versions. I must say they've got the coolest designs around (in the "affordable" range; a Ford GT would be pretty cool too) and I'd be all over a new RX-7 if they bring it back.
No, a revoked cert *is* the solution. There are many possible reasons why a cert may need to be revoked (e.g., issuer got hacked, policy change, name change, etc.) so just locking down the process of issuing certs is not a solution. The big problem here (re: slowness of CRL checking) is that X.509 certificates are intended to be part of a *distributed* authentication system. You're not supposed to have the entire world dependent on a couple of root-level CAs, they're supposed to delegate authority to widely distributed sites. Just like DNS would fall over dead if everyone always had to start their queries at the root nameservers before getting anywhere. The fact that most companies are ignorant and just buy a couple server certs from Verisign is part of the problem here. If there were more intermediate CAs distributed around the net that could handle the CRL distribution load, then verifying cert validity in realtime would be a lot faster.
According to this article http://www.rpi.edu/~danony/Research/Pyro.htm the 200,000eV was when configured to produce X-rays. The neutron generator yielded neutrons at 2.5MeV. Still a tiny amount of energy of course...
Yes... but unfortunately, as a society, most of us are asleep/drugged and far from the "ever-vigilant" citizens that we need to be.
Sounds like an Atari or an Amiga. Or a mainframe.
But yeah, you can get a lot of mileage from having dedicated coprocessors (including dedicated I/O processors). Something that everyone except Apple and commodity PC makers seems to already understand. Of course, there are still limits to that, and it's always nice to have a faster central processor.
The fat lady isn't singing yet. She's not even backstage waiting for her cue.
If it's true that these other electronics companies' products aren't fully conforming to the spec, then it should be simple to objectively prove their non-conformance and get them to change it. Someone like CableLabs ought to be certifying each company's implementation before allowing them to slap on a "CableCard compatible" logo on their products and literature. The fact that you're seeing compatibility problems means either the spec is broken, or the certification process is broken. The basic idea, freeing viewers from the inconvenience of set top boxes, was valid.
re: recovering costs - why should the majority of cable subscribers pay the costs incurred by a few stupid customers? If I subscribe and never need to call for service or troubleshooting, why should I be paying for the cost to repair some idiot's equipment down the street? From my perspective, you're just milking customers, because I never see any benefit from the charges I'm paying for.
Personally, I've already downloaded all of the CableCard 2.0 specs. Unfortunately the 1.0 specs are no longer freely available. But anyway, assuming CableCard 2.0 goes anywhere, it will be a simple matter for a 3rd party hacker to implement the same protocols purely in software, certification be damned. At that point, the cards become become irrelevant...