I'm not sure if you got the parents point (apologies if you did). By trivial string handling he's talking about recursive structures, and the erroneous strings he's mentioning are probably programs as input to a compiler. The 'non-trivial' strings are the class of strings that you would need a full grammar in order to parse, rather than a reg-exp. But yeah, not every time - horses for courses and all that.
There seems to flaw in both of the analogies that the physicist presents in the article. The example that you quote of the water circling has a basic flaw. The water molecules are reasonably uniform and so there doesn't appear to be any difference in the water after it has swirled around the stone. But imagine that each molecule is a slightly different colour, and that overall they make up a recognisable pattern. After they have swirled around the stone the pattern is destroyed - which would certainly be visible.
The other example that he uses is invisibility to radar - but radar assumes that the observer is actively sending out rays. Preventing them from being reflected makes the object invisible to radar. But moving from radar to light the observer becomes passive and sees the reflections of the ambient light in scene. If the object stops reflecting these then it is still visible - just very dark. We already have objects with this kind of "invisibility", the main character in the Matrix wears one most of the time...
Neither approach seems to be practical when applied to light and visibility.
Not quite. If there are four separate memory subsystems then the fillrate of 80 *billion pixels* could be divided into 20 Gp/s per subsystem. But to get 20GB/s that would imply one byte per pixel. Four bytes per pixel is the absolute minimum, but 80 GB/s is way above DDR3. The claim was actually in the Register article rather than the press release so it possible that somebody copied down the wrong number and/or unit in a briefing...
The other weird anomaly is the fillrate. 80 Gp/s is a long way ahead of the 7800-GTX. A long way. Currently the 16 ROPs on the 7800-GTX limit the fillrate to 25GB/s or 1.6Gp/s assuming 16 bytes per pixel. The claim of a 148 megapixel framebuffer in 2GB of memory implies 16 bytes per pixel but this would produce a whopping memory bandwidth of about 1.2 Terrabytes per second. Even if it were a single byte per pixel there is no way they will get 80GB/s of fillrate out of just two cards. DDR3 is maxed out at 25.6GB/s. Somebody in the marketing department has added a large side serving of bullshit to this press release.
Well sure we could be. The actual supercomputer in the article isn't a general purpose machine although it does run the Linpack. They managed to get such a high performance by limiting the operations, much like a GPU. In more general terms a processor capable of 8Gflop/s can be had for about $100 - so general purpose flops would be about 120Gflop/s for the $1500. Not quite as impressive but still quite high...
JIT on a separate core would be sweet, but an even better use would be pushing the garbage collection onto a second core. Current GC methods aren't as accurate as they could be, and leave certain blocks uncollected because the time taken to find them would cripple response times for software on a single core. A second core would allow much more aggressive GC and reclaim more memory.
The big distinction is how much parallelism exists in the workloads. The classic quote in the literature for graphics is that 'It is an embarassingly parallel task'. Most general workloads are not. I think the main dead with the multi-core -> single-core work was that it was targetted at multi-threaded code. You can't really fake a single core that well for a sequential application because of the latency, but you can hide latency really well across multi-threaded code. The problem with building a single sequential core out of cheap vector parts is the same.
I think someone further down the page has the right idea - imagine you are building a 4-core processor (like AMD). Throw one of the cores away. Use those transistors to build a mini-vector processor. It's not that graphics specific. You get more raw Gflop/s for the 'right' applications from that one core, and a lot more processing for the 'wrong' out of the other three than out of any hybrid architecture. Suddenly its a new tradeoff - how much vector performance, how much scalar?
Ok, time to blow off the moderations that I've made so far as you're missing something fairly obvious. Take a look at the processor on the 7800GTX, with 300 million transistors it is currently the most complex chip being shipped (I don't how big Cell is). In exchange the peak processing power is 320Gflop/s (40 in the vertex processors, 280 in the fragment shaders). For comparison the floating point performance in a CPU is ~ 8Gflop/s. That's a whopping 40 cores to break even on graphics processing.
Once you can fab a processor large enough to contain 40 functional cores - how big a GPU do you think you could fab on the same process? The simple fact is that a GPU is completely crippled compared to a CPU. There are huge tradeoffs in the design to get that kind of performance. Stream processing is very limited compared to a von Neumann architecture if you care about latency in the slightest. But for graphics - it's perfect. Throwing completely independent parallel chunks of data through an array of vector processors is a much simpler challenge than attempting to extract parallelism from sequential code. The sequential code has pesky things like control-flow that is missing in the gfx shaders, and I don't mean the rubbish that ATi/Nvidia are selling as control-flow in their current designs. That is sheer marketing given the size of the shader batches and the depth of the pipelines.
So I don't think the big 'ol wheel of reincarnation is going to move rendering back into software anytime soon. But what people forget is that AMD is not really a processor company. They are a fab company that just happens to design some kick-ass processors. Their main business is silicon, and buying ATi is the biggest chunk of vertical integration you can imagine...
No that was stupidity. Arrogance was making me pay a ridiculous amount of money for a 2Gb memory stick for the psp because they use a propriatory format. Then expecting me to fork out again because they use a *different* propriatory format on their phones. Bastards.
Although putting the semantics in the whitespace sounds like a good idea there is a subtle problem - you can see the whitespace. If the tab setting between your editor and the interpreter is different then it can really screw you up. I found this whilst learning Haskell for the first time - made the language a real bitch to learn...
What you are after isn't RFID. There are indoor ultrasonic positioning systems that do what you are asking for. The transmitters are about RFID tag size. The batteries don't last forever but the time between recharges is getting better all the time and for an application like this they could have a really long sleep cycle between pulses.
Really? On a site with such a large amount of unix experience you think it's strange for people to want a device that performs a single function well rather than a jack-of-all-trades?
Furthermore they claim that in this 'internet' of computers there will be some sort of central database that holds IP addresses. So there must be some sort of key for each record in the database. Hey, what if we made that key a name! Then the IP addresses could change but computers could look up the name and get an IP address for direct communication!! A Dynamic Naming System! Dude, add me as a claimant, we're gonna march this one all the way to the bank
If you drive your 85MPH Toyota on a crowded highway and it never gets above 45 would you blame the highway, or would you blame Toyota? The end-to-end link between you and the ISP is sold rated at a certain speed - but you contend for access to the internet across their network with the other customers. This is why a DSL is shit cheap compared to a dedicated line and why millions have DSL rather than one or two organisations.
The other reply above has already made the point that quoting is unnecessary in a dialogue - although most people (including myself) are too lazy to make the effort to avoid needing quotes.
When the material is not a dialogue, such as your example of discussing an essay, then large pieces of material may be neccesary - but not in contiguous blocks. In order to make a point the smallest surrounding context is used. Much like penix1's point this is an art in choosing what to quote, and how to phrase comments on it, and what not to quote.
The basic problem with blogs is that they are not works in any sense. They are splashes of thought, designed to capture the normal tumble of useless thoughts that most people choose to forget. Because of this property they are home to the worst excesses of bad quoting.
I replied in haste previously, you are correct that it is not plagarism as long as the attribution is clearly marked. I've read through the lower replies that discuss this in some depth. However, it is *something*, and something that is clearly wrong. While it is not plagarism, and it is technically copyright infringement (as I stated somewhat unclearly), it is actually a different problem. It is diluting the original work in some way, although I can't think of a clear term for what this is.
Actually it is. Citation of source text is only allowed under fair use provisions which require the passge being quoted to be of a reasonably small size. If I copy an entire book, then add a citation at the bottom and some fascile quote like 'this rocks' then even with the citation it is clearly plagarism. This is the depth that some blogs sink to.
While 3d acceleration works fine under linux how are you getting it working under VMware? If VMware supported 3d hardware I'd think about using for games that cedega doesn't support but currently they don't touch it with a 30ft pole. Even being able to switch control of the video card between the host/guest OS for full-screen support would be a step forward.
Thanks for the tip. I figured that a failed pedant clutching at straws by complaining about a failed pendant clutching at straws was fairly ironic. Wow, guess I was so wrong about that one. Care to comment about my knowledge of sarcasm next?
Wow, do you have this much difficulty with communication in real life, or are you just a troll online? It scans wrongly to me and I was commenting to the GP about the use of the word. My post also quite clearly pointed out that my original post was wrong and that the OP was correct. Which all makes your final line all the more ironic...
I looked this up as well and it seems suprising that "right" is an adverb - because none of the adverb uses of the word scan correctly as english. Looking in an English dictionary I see that this isn't an American/English difference but is the same in both cases. The OP's language did strike me as incorrect as well, but I guess that it is just an uncommon usage.
Someone can be right, as opposed to wrong. But you cannot do something right - you would do it correctly. Right is not a word that has that kind of active tense (I don't the technical term).
Slashdot Mirror
I'm not sure if you got the parents point (apologies if you did). By trivial string handling he's talking about recursive structures, and the erroneous strings he's mentioning are probably programs as input to a compiler. The 'non-trivial' strings are the class of strings that you would need a full grammar in order to parse, rather than a reg-exp. But yeah, not every time - horses for courses and all that.
There seems to flaw in both of the analogies that the physicist presents in the article. The example that you quote of the water circling has a basic flaw. The water molecules are reasonably uniform and so there doesn't appear to be any difference in the water after it has swirled around the stone. But imagine that each molecule is a slightly different colour, and that overall they make up a recognisable pattern. After they have swirled around the stone the pattern is destroyed - which would certainly be visible.
The other example that he uses is invisibility to radar - but radar assumes that the observer is actively sending out rays. Preventing them from being reflected makes the object invisible to radar. But moving from radar to light the observer becomes passive and sees the reflections of the ambient light in scene. If the object stops reflecting these then it is still visible - just very dark. We already have objects with this kind of "invisibility", the main character in the Matrix wears one most of the time...
Neither approach seems to be practical when applied to light and visibility.
Not quite. If there are four separate memory subsystems then the fillrate of 80 *billion pixels* could be divided into 20 Gp/s per subsystem. But to get 20GB/s that would imply one byte per pixel. Four bytes per pixel is the absolute minimum, but 80 GB/s is way above DDR3. The claim was actually in the Register article rather than the press release so it possible that somebody copied down the wrong number and/or unit in a briefing...
The other weird anomaly is the fillrate. 80 Gp/s is a long way ahead of the 7800-GTX. A long way. Currently the 16 ROPs on the 7800-GTX limit the fillrate to 25GB/s or 1.6Gp/s assuming 16 bytes per pixel. The claim of a 148 megapixel framebuffer in 2GB of memory implies 16 bytes per pixel but this would produce a whopping memory bandwidth of about 1.2 Terrabytes per second. Even if it were a single byte per pixel there is no way they will get 80GB/s of fillrate out of just two cards. DDR3 is maxed out at 25.6GB/s. Somebody in the marketing department has added a large side serving of bullshit to this press release.
Well sure we could be. The actual supercomputer in the article isn't a general purpose machine although it does run the Linpack. They managed to get such a high performance by limiting the operations, much like a GPU. In more general terms a processor capable of 8Gflop/s can be had for about $100 - so general purpose flops would be about 120Gflop/s for the $1500. Not quite as impressive but still quite high...
No? 'cos a GTX-7800 does 320Gflop/s and you could buy a few of those for $1500...
JIT on a separate core would be sweet, but an even better use would be pushing the garbage collection onto a second core. Current GC methods aren't as accurate as they could be, and leave certain blocks uncollected because the time taken to find them would cripple response times for software on a single core. A second core would allow much more aggressive GC and reclaim more memory.
The big distinction is how much parallelism exists in the workloads. The classic quote in the literature for graphics is that 'It is an embarassingly parallel task'. Most general workloads are not. I think the main dead with the multi-core -> single-core work was that it was targetted at multi-threaded code. You can't really fake a single core that well for a sequential application because of the latency, but you can hide latency really well across multi-threaded code. The problem with building a single sequential core out of cheap vector parts is the same.
I think someone further down the page has the right idea - imagine you are building a 4-core processor (like AMD). Throw one of the cores away. Use those transistors to build a mini-vector processor. It's not that graphics specific. You get more raw Gflop/s for the 'right' applications from that one core, and a lot more processing for the 'wrong' out of the other three than out of any hybrid architecture. Suddenly its a new tradeoff - how much vector performance, how much scalar?
Ok, time to blow off the moderations that I've made so far as you're missing something fairly obvious. Take a look at the processor on the 7800GTX, with 300 million transistors it is currently the most complex chip being shipped (I don't how big Cell is). In exchange the peak processing power is 320Gflop/s (40 in the vertex processors, 280 in the fragment shaders). For comparison the floating point performance in a CPU is ~ 8Gflop/s. That's a whopping 40 cores to break even on graphics processing.
Once you can fab a processor large enough to contain 40 functional cores - how big a GPU do you think you could fab on the same process? The simple fact is that a GPU is completely crippled compared to a CPU. There are huge tradeoffs in the design to get that kind of performance. Stream processing is very limited compared to a von Neumann architecture if you care about latency in the slightest. But for graphics - it's perfect. Throwing completely independent parallel chunks of data through an array of vector processors is a much simpler challenge than attempting to extract parallelism from sequential code. The sequential code has pesky things like control-flow that is missing in the gfx shaders, and I don't mean the rubbish that ATi/Nvidia are selling as control-flow in their current designs. That is sheer marketing given the size of the shader batches and the depth of the pipelines.
So I don't think the big 'ol wheel of reincarnation is going to move rendering back into software anytime soon. But what people forget is that AMD is not really a processor company. They are a fab company that just happens to design some kick-ass processors. Their main business is silicon, and buying ATi is the biggest chunk of vertical integration you can imagine...
No that was stupidity. Arrogance was making me pay a ridiculous amount of money for a 2Gb memory stick for the psp because they use a propriatory format. Then expecting me to fork out again because they use a *different* propriatory format on their phones. Bastards.
Really? How many children have died in Africa alone because of the Catholic church's views on contraception?
Although putting the semantics in the whitespace sounds like a good idea there is a subtle problem - you can see the whitespace. If the tab setting between your editor and the interpreter is different then it can really screw you up. I found this whilst learning Haskell for the first time - made the language a real bitch to learn...
What you are after isn't RFID. There are indoor ultrasonic positioning systems that do what you are asking for. The transmitters are about RFID tag size. The batteries don't last forever but the time between recharges is getting better all the time and for an application like this they could have a really long sleep cycle between pulses.
Really? On a site with such a large amount of unix experience you think it's strange for people to want a device that performs a single function well rather than a jack-of-all-trades?
Furthermore they claim that in this 'internet' of computers there will be some sort of central database that holds IP addresses. So there must be some sort of key for each record in the database. Hey, what if we made that key a name! Then the IP addresses could change but computers could look up the name and get an IP address for direct communication!! A Dynamic Naming System! Dude, add me as a claimant, we're gonna march this one all the way to the bank
If you drive your 85MPH Toyota on a crowded highway and it never gets above 45 would you blame the highway, or would you blame Toyota? The end-to-end link between you and the ISP is sold rated at a certain speed - but you contend for access to the internet across their network with the other customers. This is why a DSL is shit cheap compared to a dedicated line and why millions have DSL rather than one or two organisations.
Nice pun dude. I guarantee that nobody on /. will get it...
The other reply above has already made the point that quoting is unnecessary in a dialogue - although most people (including myself) are too lazy to make the effort to avoid needing quotes.
When the material is not a dialogue, such as your example of discussing an essay, then large pieces of material may be neccesary - but not in contiguous blocks. In order to make a point the smallest surrounding context is used. Much like penix1's point this is an art in choosing what to quote, and how to phrase comments on it, and what not to quote.
The basic problem with blogs is that they are not works in any sense. They are splashes of thought, designed to capture the normal tumble of useless thoughts that most people choose to forget. Because of this property they are home to the worst excesses of bad quoting.
I replied in haste previously, you are correct that it is not plagarism as long as the attribution is clearly marked. I've read through the lower replies that discuss this in some depth. However, it is *something*, and something that is clearly wrong. While it is not plagarism, and it is technically copyright infringement (as I stated somewhat unclearly), it is actually a different problem. It is diluting the original work in some way, although I can't think of a clear term for what this is.
Actually it is. Citation of source text is only allowed under fair use provisions which require the passge being quoted to be of a reasonably small size. If I copy an entire book, then add a citation at the bottom and some fascile quote like 'this rocks' then even with the citation it is clearly plagarism. This is the depth that some blogs sink to.
While 3d acceleration works fine under linux how are you getting it working under VMware? If VMware supported 3d hardware I'd think about using for games that cedega doesn't support but currently they don't touch it with a 30ft pole. Even being able to switch control of the video card between the host/guest OS for full-screen support would be a step forward.
Thanks for the tip. I figured that a failed pedant clutching at straws by complaining about a failed pendant clutching at straws was fairly ironic. Wow, guess I was so wrong about that one. Care to comment about my knowledge of sarcasm next?
Wow, do you have this much difficulty with communication in real life, or are you just a troll online? It scans wrongly to me and I was commenting to the GP about the use of the word. My post also quite clearly pointed out that my original post was wrong and that the OP was correct. Which all makes your final line all the more ironic...
I looked this up as well and it seems suprising that "right" is an adverb - because none of the adverb uses of the word scan correctly as english. Looking in an English dictionary I see that this isn't an American/English difference but is the same in both cases. The OP's language did strike me as incorrect as well, but I guess that it is just an uncommon usage.
Someone can be right, as opposed to wrong. But you cannot do something right - you would do it correctly. Right is not a word that has that kind of active tense (I don't the technical term).