Why is it that this inventor is not included in "society" and must be prevented from benefitting from it?
Esperandi The man who is told he must sacrifice himself for the good of others always wonders who these "others" are and why its not them who should be sacrificed since they seem to be so fond of it.
He's not hoarding the technique, and he's not keeping it from "humanity." He's patenting it, and then he'll license it out to people who want to use it.
Since you obviously hate people profiting things and think that simply being born human entitles you to the products of this mans mind, I'll approach this from a viewpoint you can grok.
Okay, say you've got Megacorp X. Megacorp X wants to have the next best video compression technology (which utilitizes LOTS of FFTs). Now, if this man didn't have a patent, they'd hire a programmer to work at a meager salary, and get 98% of the research and devlopment for free. This man, however, has to get a job at McDonalds or something because he let other people convince him that the products of his mind were "the right of humanity." On the other hand, if this man patents it, he can license it to Megacorp X for a hefty fee, and he'll be well funded for whatever project he wishes to approach next, and, most importantly, justice will have been preserved and not perverted. (I'm talking natural justice here, not legal justice... the kind where if you're stranded on an island your "right as a human" isn't going to get the damn coconut out of the tree, you do it yourself and you profit from it)
If you discover a significant mathematical principle, rest assured ever single person that uses it because you released it so freely is just saying behind your back "What a tremendous dumbass, doesn't he know that he's worth something?" while you'll sit smugly trying to convince them that you're not.
I know most of the replies about the DMCA are going to revolve around DeCSS, but you could probably help your chances in getting your voice heard if you approached it form a less-iffy state. DeCSS really doesn't scream legitimacy since its first release WAS a program to copy DVDs and not simply a DVD player. Theres also the issue of them reverse engineering Xings player and all.... its just not a very solid issue to stand behind.
However, there is a very solid issue pending in courts now - Streambox VCR. they are being charged with the exact same thing because they circumvented RealNetwork's copyrighting "switch" that they put in real streams served from RealServers. Now, Streambox VCR is _directly_ analogous to using a VCR to record a television program. This application, and this legal battle, could more easily be used to persuade the government that this sort of leniency is a good thing.
And listen to the sane posts, please don't flame, it will not help our situation at all.
Progress costs. I can't realy think of any scientific advance, especially in medicine, that didn't involve multiple deaths and mistakes. The guy who invented refrigeration died doing it, would you rather he have lived and we never got it? To produce things takes many errors and corrections.
Hopefully in the future we'll have computers powerful enough to simulate what a certain drug will do if it is ever combined with other drugs and things like that, but for right now we've got to live with risks or we'll end up standing still...
Like you, I'm not a doctor, but your idea sparked one of my own... sorta like a genetic-ly modifying douche.... Like they could bathe a womans eggs in it and fix any possible defects that might exist in the eggs... it'd be harder from guys since they're constantly re-producing sperm, but women start with a finite number of eggs and produce no more...
The marketing guy explained that Transmeta has no hype, its all buzz;) Buzz is when other people speculate about your company and products, hype is when you speculate about your own company and products... I'm not sure which is better, or which leads to less let-downs...
If you can sign up to receive Transmeta's developers kit thingee they have on their website if you're not in a company and will probably never really produce anything? I'd just love to read about it and toy with ideas, but I doubt I'd ever produce anything of real value... I don't want to sign up and have them smack me on the ass...
Here's the info: DNA Computing : New Computing Paradigms (Texts in Theoretical Computer Science) by Gheorghe Paun, Grzegorz Rozenberg, Arto Salomaa, W. Brauer (Editor)
A few people emailed me asking about this book, I figure this is a good time to post it for all to see... its a very good book and after reading it you'll realize this is a really nifty thing for DNA-based computers.
Oh, and NO DNA knowledge is needed to read this book, I knew bothing going in, now I feel like I could do the stuff in my basement if I knew how to get DNA out of a living cell or how to synthesize it...
Thank God he's finally out, when do you think he'll steal my credit card number, hijack my online account, and reprogram a cell phone to get his calls billed to me?
I don't see why anyone who wants to defend REAL hackers (read Steven Levy's book "Hackers" for a slew of them, real stories too, not kiddie crap) would stand behind Kevin, he's just a boldfaced criminal.
Esperandi Do these same people stand behind store clerks who steal credit card numbers?
"If I captured jpegs from my TV capture card and posted them on my homepage (as I've seen many people do), is that illegal? I'm paying for the signal, the TV card, the internet connection... aren't I paying for the content, too? "
It would probably not be illegal because what you posted was such a small amount (and amount matters in copyright issues)... but if you posted a TV show, it would be very illegal. And no, you're not paying for the content. Advertisers are paying for the content. You are paying for access to the things which the advertisers fund. The advertisers fund the things because you pay to watch them and show with big ratings that their commercials will get watched by a bunch of people.
You can tape shows for your own personal use. You can tape shows in order to "time-shift" programs and watch them at your convenience. You CANNOT re-broadcast those tapes, even if you do it for free. Its against the law. Just like I can't run a book through a Xerox machine, OCR it, and put it on the Internet because I paid a buck fifty and got a library card. You're not buying the content. The fact that you THINK you're paying for the content truly is wholly irrelevant. Ignorance of the law has never been an excuse.
The line between fair use and theft is pretty gray, but its basically this: If you use something in such a way that it would result in the original author losing money, its theft. For example, let's say you record, encode, and serve out, the television show The Simpsons. Theoretically for free. Now, if people can get The Simpsons online, there is a good chance people will view it online and not watch the re-runs on TV. That means the advertisers won't be filling the ad spots during the re-runs (or at least not paying as much) and the loss of revenue flows uphill to the creator who stands on top and just got his legs cut off by someone who thought they owned The Simpsons because they paid $25/mo to get 50 channels.
"and they are thus allocated to television stations to serve the public"
Correct me if I'm wrong, but these airwaves are auctioned off and PAID FOR by the networks, not "allocated to television stations to serve the public". If they are allocated "to serve the public" then I'll never watch network television ever again.
"RealNetworks Wins Preliminary Injunction in Landmark Case Protecting Copyright Holders in the Digital Age"
This is absolutely true.... and happened WEEKS ago! _Preliminary_ injuntion. This story is about the court case AFTER this injunction succeeded. In the court case which happened yesterday (the 18th), Streambox won. They are not 2 competing press releases, they are 2 press releases about 2 different events.
Esperandi I can't fathom how you got moderated up for this.
Good question... you don't really need this though. The way the processor works, the code morphing layer has a cache. Once it translates a certain OpenGL instructions fallout (a sinlge OpenGL instruction boils down to hundreds of little math instructions through the API that deal with matrices), it caches that. The next time that comes around, those instructions aren't re-translated, they're simply re-translated. If this happens a few times (and in real world applications it does constantly... unfortunately benchmark programs specifically do the opposite of this, trying hundreds of different things in a few seconds) the code is also optimized.
So you don't need to add these things into the code morphing layer, they will automatically be folded into the cache which is sort of like a "superfast bag o tricks" the Crusoe uses. Its sorta like how they emulated the N64 on a PC. They simply do things differently, which results in much more speed out of much less hardware.
"Can you imagine the performance increase you'd see if you compiled directly into the native Crusoe instruction set, so it could bypass all that code-morphing (emulation) software? "
I think the advantage would be very small. In fact, I'm not sure if there would be much at all. A native optimizing compiler would work just like the code morphing layer.... hmm... this is working my brain;) On one hand, you bypass the cacheing structure, there would be a very small increase. On the other hand, if its not optimized for the pipeline as well as the code morphing layer is, there would be a sizeable performance hit. Where do you believe the advantage would come from?
Esperandi Emulators aren't always slower. Look at UltraHLE. Look at UltraHLE under Wine which runs faster than UltraHLE does under Windows...
While I would really like to start writing assembly code for the VLIW processor and all that (hey, i think it would be fun, don't look at me like that), i do see one advantage to this view...
Theoretically I could be sitting using my brand year-old 700MHz Transmeta-based notebook, and use it to go to the Transmeta website. Wow, an upgrade that increases my speed by 50%!
This probably will not happen, I'm assuming they pretty much squeezed as much optimization into their code morphing module as possible before release to have an impressive product, but in the future it could happen when they emulate other platforms.
For instance, they might release a sort of dual-boot configuration for the code morphing module that lets you run Alpha-based apps. They might release this early, without many optimizations, and then it gets better over time. If someone swoops in and write native VLIW code for it, well, that code is going to get broken. Also, with separate code morphing engines, the VLIW code would become a very strange type of kinda-platform-specific beast. There'd be releases of software "compatible with revision 9.02.3 of the Crusoe TM5400 which emulates the Alpha platform" and such things.
But in that case, recompilers would come along... it would be a fairly sizable mess, but it will happen. You think no hacker is going to get in there and get curious about the code-morphing module? Its stored in DRAM, theres gotta be a way to get at it;)
I didn't mention this in my first post because I wasn't sure how right I am, but I assume that if Transmeta was to produce an architecture to emulate RISC, their pipelining would be re-done and such.
But, excepting that, when taking highly optimized RISC code, assuming the Crusoe couldn't optimize it any more, its advantages wane but still exist. The 128 bit instruction words pack in 4 instructions in a clock cycle, that's an advantage there. Also, I'm not sure how many RISC instructions can go into a single Crusoe atom, but if its more than 2, that's the equivalent of doing 8 times as many RISC instructions right there.
I can see your points about RISC optimizied code being re-optimized... its like compressing compressed data when you first look at it, but the Crusoe processor will "understand" or will "learn" the given optimizations and re-do them ONCE for its own pipeline architecture... I think the cahcheing will give it a great advantage, but I guess it will have to wait for a real world trial since the Crusoe is designed for real world applications where repetitive code that is continually re-scheduled for normal processors is re-scheduled once and saved really shows an advantage.
Esperandi I'm glad you understand emulation CAN be faster though, lots of people refuse to admit that;)
Heh, you didn't read the whitepaper, right? Didn't think so. Where to start?
First, the lying. You expose your ignorance (that means lack of knowledge, not stupidity) of how a computer works here. You see, you got the processor, then above the processor you have the microcode, then you have the stuff above it. The microcode is where you send an encoded opcode like ADD AX, 0004h or something similar. The microcode then translates this to something the processor can use. In the case of an ADD that involves memory locations, it translates that into code that moves the stuff from memory into a register, does the add, etc. So pretty much all instructions in a CPU are translated down. Even moreso with things like MMX and Katmai SIMD instructions. So they're not lying, you just don't get their meaning.
Now, as to cacheing. In that cache, the code is also optimized. Also, that cache is not a cache of x86instructions. It is a cache of Crusoe "molecules". Molecules consist of 4 atoms which are executed in parallel in one clock cycle. During the optimization, the instructions are reordered so that things both execute faster and execute lesser code. The simple example they gave in their whitepaper (which I urge you to read) was when there were 4 x86 instructions. Those instructions would get translated to many micro-ops and sent thru a rescheduler every time they are executed. In a Crusoe, they are translated and optimized once, then every time they get executed they execute less code and execute it faster without re-scheduling. It ends up a lot faster.
Think about this for performance: DVD playing (another example from the whitepaper). The P3 700 works at 700MHz decoding the DVD keeping a constant framerate. Once the first frame of video has displayed on a Crusoe, the instructions for the DVD decode process are cached and optimized. The Crusoe can clock down to 400MHz and drop its voltage consumption and keep exactly on line with the performance of the 700MHz P3. So no, the performance isn't that of a P120 in realworld use.
Why does it fail at benchmarks then? Easy, benchmakrs don't repeat code. They do different things very quickly and do not allow any cacheing or optimization to be done. This _never_ happens in real life. Right now I'm typing in this textbox, and the code executing every single time I type a character would be caches and optimized and would be executing many times faster on a Crusoe than on an equivalent x86 machine. It would only slow down if I typed one character, and then clicked used several hundred different features in my browser in the space of a minute. That is never going to happen because I can't type or move the mouse that fast.
Esperandi The Crusoe is the worlds first Real World Processor. Oh, and a little PS here... trust me, I am never one to jump when Linus says so, I haven't used Linux in a couple years and I don't plan on switching to it any time soon. I like hardware compatibility and the idea programmers get paid for their work.
"RISC emulation is vastly different. Each instruction doesn't do a whole lot, but runs extremely fast. So, basically, to emulate a RISC processor, you have to translate each instruction into one or more Curose instructions that don't benefit from RISC hardware's pipelining and other efficiencies. You're going to end up with a vastly slower PC"
The way a Crusoe would handle RISC "emulation" would be very very efficient when applied to RISC processors, much more so than for CISC processors. How? Well, in a RISC chip, since the instructions are smaller, there is more repetition of them. What would be four or five CISC operations gets translated into one Crusoe molecule and is then cached and optimized. When it repeats, it doesn't have to re-translate, it simply executes the optimized version that is cached.
They've got lots of catches in the hardware of the chip that let it execute things out of order with no penalties, thereby allowing a lot more optimization. Now when they decide to re-write the code morphine module to handle a RISC instruction set it will probably take 8 or 16 RISC instructions and translate them into a single molecule which is executed in paralell.
The same applies to 3DNow and such. When a 3DNow instruction goes to the processor it has to be translated into micro-ops and executed, and then re-scheduled. Every single time. The Crusoe would perform slower the first time this instruction is executed, but afterwards would perform much faster...
Transmeta doesn't really have to participate in the clock speed race (BTW, if you factor price into it, Intel isn't even close right now)... A 700MHz P3 and a 700MHz TM5400 CPU do not perform the same. In a repetitive section of code (which is most of the code being executed in real world applications), the P3 just executes the code over and over re-doing the instruction translation, the scheduling, and all those things. The TM5400 "learns" what the code is doing and optimizes it down to the point where it executes that code at the same speed when stepped down to 400MHz as the P3 700 does running at full speed and full power consumption.
The example they give is a DVD player, by the time the first video frame has loaded the entire DVD-decoding process has been translated, stored in a translation cache (so it is never translated again), and optimized. If it continued running at 700MHz it would be performing like a 1GHz P3, but instead it steps down intelligently. If stepping down would lose frames or slow execution speed, it wouldn't do it... so until code really really needs over 700MHz in real world applications, Intel will not have outpaced Transmeta. Ithink they'll be around for awhile. And I don't expect this to be the only thing they do. They say they're going to be a professional R&D lab, i doubt they're just goin to keep adapting these processors to different things, they'll move on...
These facts eliminate a few of the bad things you pointed out... like what is the point of integrating 3D if all of the 3D instructions are cached and optimized? There wouldn't be much of a speed improvement if any....
Seriously, DNA computers are more advanced right now than quantum computers or molecular-level computers. I've got the book on them (theres only one) and they seem really promising...
I just got done reading the technical whitepaper on how the new Transmeta Crusoe CPUs really work, and it sounds like mixing these fuel cells with the Crusoe processors would produce one of the best mobile computing platforms in the world. Super speed with amazing abilities to conserve power added to a more plentiful power source, hurrah!
BTW, to understand why Transmeta is saying their CPUs won't perform well on existing benchmarks and will produce much better results in real life, read their whitepaper. In short, its because the processor excel at re-executing stuff, and benchmarks do things like using every single feature of a word processor once. When was the last time you did that?
Like the other 900 people responding to this/. article and the last, you need to learn that trademark laws are codified and real and are available at the USPTO website. Take a look at this quote from you for instance: "Considering how much money Coca-Cola makes off of selling T-shirts, refrigerator magnets, windbreakers and the like with their "Coke" trademark on them, they aren't going to let this happen. It means a lot of money to them. "
Now, if you knew that trademarks are assigned to specific goods and services listed as a trademark is registered, you would know 2 things: #1, Coke has trademarks on softdrinks, T-shirts, magnets, windbreakers, ande verything else. The products are listed by name in a very very long list that Coke's trademarks cover. And they go no further. #2, The Linux trademark lists ONE ITEM under its goods and services: computer operating systems. In other words, what Linus and his lawyers are doing is extra-legal at best, downright illegal at worst. His trademark does not apply to domain names, T-Shirts, hell, not even company names if they don't sell operating systems!
So no, Linus is not doing what he is legally responsible to do. He IS being a bully and trying to enforce this trademark that applies only in the situation of naming an operating system.
Linus is getting tricky... he must know that his trademark doesn't extend to anything except to the names of computer operating systems. He is legally bound to uphold THAT trademark, but when he goes and tries to "uphold his trademark" in things like domain names and such, he's simply going outside the law.
My read of the new Transmeta patent confirmed that yes, the processor will "emulate" other instruction sets... but that is not the important heart of the patent. The patent is not on a chip that translates instructions from one set to another (watch for that one, it should be coming soon), the patent is very specifically on a very important issue to operating system designers. Basically, their chip sounds like it will handle multiprocessing (not multithreading since the processes will not share their environment) and prevent process deadlock. I've never seen a study on deadlock avoidance in a situation where the instruction sets are multivarious and being cross-translated but I imagine it could get quite hairy!
Slashdot Mirror
Why is it that this inventor is not included in "society" and must be prevented from benefitting from it?
Esperandi
The man who is told he must sacrifice himself for the good of others always wonders who these "others" are and why its not them who should be sacrificed since they seem to be so fond of it.
He's not hoarding the technique, and he's not keeping it from "humanity." He's patenting it, and then he'll license it out to people who want to use it.
Since you obviously hate people profiting things and think that simply being born human entitles you to the products of this mans mind, I'll approach this from a viewpoint you can grok.
Okay, say you've got Megacorp X. Megacorp X wants to have the next best video compression technology (which utilitizes LOTS of FFTs). Now, if this man didn't have a patent, they'd hire a programmer to work at a meager salary, and get 98% of the research and devlopment for free. This man, however, has to get a job at McDonalds or something because he let other people convince him that the products of his mind were "the right of humanity." On the other hand, if this man patents it, he can license it to Megacorp X for a hefty fee, and he'll be well funded for whatever project he wishes to approach next, and, most importantly, justice will have been preserved and not perverted. (I'm talking natural justice here, not legal justice... the kind where if you're stranded on an island your "right as a human" isn't going to get the damn coconut out of the tree, you do it yourself and you profit from it)
If you discover a significant mathematical principle, rest assured ever single person that uses it because you released it so freely is just saying behind your back "What a tremendous dumbass, doesn't he know that he's worth something?" while you'll sit smugly trying to convince them that you're not.
Esperandi
I know most of the replies about the DMCA are going to revolve around DeCSS, but you could probably help your chances in getting your voice heard if you approached it form a less-iffy state. DeCSS really doesn't scream legitimacy since its first release WAS a program to copy DVDs and not simply a DVD player. Theres also the issue of them reverse engineering Xings player and all.... its just not a very solid issue to stand behind.
However, there is a very solid issue pending in courts now - Streambox VCR. they are being charged with the exact same thing because they circumvented RealNetwork's copyrighting "switch" that they put in real streams served from RealServers. Now, Streambox VCR is _directly_ analogous to using a VCR to record a television program. This application, and this legal battle, could more easily be used to persuade the government that this sort of leniency is a good thing.
And listen to the sane posts, please don't flame, it will not help our situation at all.
Esperandi
Progress costs. I can't realy think of any scientific advance, especially in medicine, that didn't involve multiple deaths and mistakes. The guy who invented refrigeration died doing it, would you rather he have lived and we never got it? To produce things takes many errors and corrections.
Hopefully in the future we'll have computers powerful enough to simulate what a certain drug will do if it is ever combined with other drugs and things like that, but for right now we've got to live with risks or we'll end up standing still...
Esperandi
Like you, I'm not a doctor, but your idea sparked one of my own... sorta like a genetic-ly modifying douche.... Like they could bathe a womans eggs in it and fix any possible defects that might exist in the eggs... it'd be harder from guys since they're constantly re-producing sperm, but women start with a finite number of eggs and produce no more...
Esperandi
The marketing guy explained that Transmeta has no hype, its all buzz ;) Buzz is when other people speculate about your company and products, hype is when you speculate about your own company and products... I'm not sure which is better, or which leads to less let-downs...
Esperandi
If you can sign up to receive Transmeta's developers kit thingee they have on their website if you're not in a company and will probably never really produce anything? I'd just love to read about it and toy with ideas, but I doubt I'd ever produce anything of real value... I don't want to sign up and have them smack me on the ass...
Esperandi
Here's the info:
DNA Computing : New Computing Paradigms (Texts in Theoretical Computer Science)
by Gheorghe Paun, Grzegorz Rozenberg, Arto Salomaa, W. Brauer (Editor)
A few people emailed me asking about this book, I figure this is a good time to post it for all to see... its a very good book and after reading it you'll realize this is a really nifty thing for DNA-based computers.
Oh, and NO DNA knowledge is needed to read this book, I knew bothing going in, now I feel like I could do the stuff in my basement if I knew how to get DNA out of a living cell or how to synthesize it...
Esperandi
You mean its like taking the Apache source code, putting it on your FTP, and making people view ads to make you money before they can get in...
Esperandi
What would the GPL think of that? I need to read it again, but I think they'd disallow it...
Thank God he's finally out, when do you think he'll steal my credit card number, hijack my online account, and reprogram a cell phone to get his calls billed to me?
I don't see why anyone who wants to defend REAL hackers (read Steven Levy's book "Hackers" for a slew of them, real stories too, not kiddie crap) would stand behind Kevin, he's just a boldfaced criminal.
Esperandi
Do these same people stand behind store clerks who steal credit card numbers?
"If I captured jpegs from my TV capture card and posted them on my homepage (as I've seen many people do), is that illegal? I'm paying for the signal, the TV card, the internet connection... aren't I paying for the content, too? "
It would probably not be illegal because what you posted was such a small amount (and amount matters in copyright issues)... but if you posted a TV show, it would be very illegal. And no, you're not paying for the content. Advertisers are paying for the content. You are paying for access to the things which the advertisers fund. The advertisers fund the things because you pay to watch them and show with big ratings that their commercials will get watched by a bunch of people.
You can tape shows for your own personal use. You can tape shows in order to "time-shift" programs and watch them at your convenience. You CANNOT re-broadcast those tapes, even if you do it for free. Its against the law. Just like I can't run a book through a Xerox machine, OCR it, and put it on the Internet because I paid a buck fifty and got a library card. You're not buying the content. The fact that you THINK you're paying for the content truly is wholly irrelevant.
Ignorance of the law has never been an excuse.
The line between fair use and theft is pretty gray, but its basically this: If you use something in such a way that it would result in the original author losing money, its theft.
For example, let's say you record, encode, and serve out, the television show The Simpsons. Theoretically for free. Now, if people can get The Simpsons online, there is a good chance people will view it online and not watch the re-runs on TV. That means the advertisers won't be filling the ad spots during the re-runs (or at least not paying as much) and the loss of revenue flows uphill to the creator who stands on top and just got his legs cut off by someone who thought they owned The Simpsons because they paid $25/mo to get 50 channels.
Esperandi
"and they are thus allocated to television stations to serve the public"
Correct me if I'm wrong, but these airwaves are auctioned off and PAID FOR by the networks, not "allocated to television stations to serve the public". If they are allocated "to serve the public" then I'll never watch network television ever again.
Esperandi
Did you even read what you pasted?
"RealNetworks Wins Preliminary Injunction in Landmark Case Protecting Copyright Holders in the Digital Age"
This is absolutely true.... and happened WEEKS ago! _Preliminary_ injuntion. This story is about the court case AFTER this injunction succeeded. In the court case which happened yesterday (the 18th), Streambox won. They are not 2 competing press releases, they are 2 press releases about 2 different events.
Esperandi
I can't fathom how you got moderated up for this.
Good question... you don't really need this though. The way the processor works, the code morphing layer has a cache. Once it translates a certain OpenGL instructions fallout (a sinlge OpenGL instruction boils down to hundreds of little math instructions through the API that deal with matrices), it caches that. The next time that comes around, those instructions aren't re-translated, they're simply re-translated. If this happens a few times (and in real world applications it does constantly... unfortunately benchmark programs specifically do the opposite of this, trying hundreds of different things in a few seconds) the code is also optimized.
So you don't need to add these things into the code morphing layer, they will automatically be folded into the cache which is sort of like a "superfast bag o tricks" the Crusoe uses. Its sorta like how they emulated the N64 on a PC. They simply do things differently, which results in much more speed out of much less hardware.
Esperandi
"Can you imagine the performance increase you'd see if you compiled directly into the native Crusoe instruction set, so it could bypass all that code-morphing (emulation) software? "
;) On one hand, you bypass the cacheing structure, there would be a very small increase. On the other hand, if its not optimized for the pipeline as well as the code morphing layer is, there would be a sizeable performance hit. Where do you believe the advantage would come from?
I think the advantage would be very small. In fact, I'm not sure if there would be much at all. A native optimizing compiler would work just like the code morphing layer.... hmm... this is working my brain
Esperandi
Emulators aren't always slower. Look at UltraHLE. Look at UltraHLE under Wine which runs faster than UltraHLE does under Windows...
While I would really like to start writing assembly code for the VLIW processor and all that (hey, i think it would be fun, don't look at me like that), i do see one advantage to this view...
;)
Theoretically I could be sitting using my brand year-old 700MHz Transmeta-based notebook, and use it to go to the Transmeta website. Wow, an upgrade that increases my speed by 50%!
This probably will not happen, I'm assuming they pretty much squeezed as much optimization into their code morphing module as possible before release to have an impressive product, but in the future it could happen when they emulate other platforms.
For instance, they might release a sort of dual-boot configuration for the code morphing module that lets you run Alpha-based apps. They might release this early, without many optimizations, and then it gets better over time. If someone swoops in and write native VLIW code for it, well, that code is going to get broken. Also, with separate code morphing engines, the VLIW code would become a very strange type of kinda-platform-specific beast. There'd be releases of software "compatible with revision 9.02.3 of the Crusoe TM5400 which emulates the Alpha platform" and such things.
But in that case, recompilers would come along... it would be a fairly sizable mess, but it will happen. You think no hacker is going to get in there and get curious about the code-morphing module? Its stored in DRAM, theres gotta be a way to get at it
Esperandi
I didn't mention this in my first post because I wasn't sure how right I am, but I assume that if Transmeta was to produce an architecture to emulate RISC, their pipelining would be re-done and such.
;)
But, excepting that, when taking highly optimized RISC code, assuming the Crusoe couldn't optimize it any more, its advantages wane but still exist. The 128 bit instruction words pack in 4 instructions in a clock cycle, that's an advantage there. Also, I'm not sure how many RISC instructions can go into a single Crusoe atom, but if its more than 2, that's the equivalent of doing 8 times as many RISC instructions right there.
I can see your points about RISC optimizied code being re-optimized... its like compressing compressed data when you first look at it, but the Crusoe processor will "understand" or will "learn" the given optimizations and re-do them ONCE for its own pipeline architecture... I think the cahcheing will give it a great advantage, but I guess it will have to wait for a real world trial since the Crusoe is designed for real world applications where repetitive code that is continually re-scheduled for normal processors is re-scheduled once and saved really shows an advantage.
Esperandi
I'm glad you understand emulation CAN be faster though, lots of people refuse to admit that
Heh, you didn't read the whitepaper, right? Didn't think so. Where to start?
First, the lying. You expose your ignorance (that means lack of knowledge, not stupidity) of how a computer works here. You see, you got the processor, then above the processor you have the microcode, then you have the stuff above it. The microcode is where you send an encoded opcode like ADD AX, 0004h or something similar. The microcode then translates this to something the processor can use. In the case of an ADD that involves memory locations, it translates that into code that moves the stuff from memory into a register, does the add, etc. So pretty much all instructions in a CPU are translated down. Even moreso with things like MMX and Katmai SIMD instructions. So they're not lying, you just don't get their meaning.
Now, as to cacheing. In that cache, the code is also optimized. Also, that cache is not a cache of x86instructions. It is a cache of Crusoe "molecules". Molecules consist of 4 atoms which are executed in parallel in one clock cycle. During the optimization, the instructions are reordered so that things both execute faster and execute lesser code. The simple example they gave in their whitepaper (which I urge you to read) was when there were 4 x86 instructions. Those instructions would get translated to many micro-ops and sent thru a rescheduler every time they are executed. In a Crusoe, they are translated and optimized once, then every time they get executed they execute less code and execute it faster without re-scheduling. It ends up a lot faster.
Think about this for performance: DVD playing (another example from the whitepaper). The P3 700 works at 700MHz decoding the DVD keeping a constant framerate. Once the first frame of video has displayed on a Crusoe, the instructions for the DVD decode process are cached and optimized. The Crusoe can clock down to 400MHz and drop its voltage consumption and keep exactly on line with the performance of the 700MHz P3. So no, the performance isn't that of a P120 in realworld use.
Why does it fail at benchmarks then? Easy, benchmakrs don't repeat code. They do different things very quickly and do not allow any cacheing or optimization to be done. This _never_ happens in real life. Right now I'm typing in this textbox, and the code executing every single time I type a character would be caches and optimized and would be executing many times faster on a Crusoe than on an equivalent x86 machine. It would only slow down if I typed one character, and then clicked used several hundred different features in my browser in the space of a minute. That is never going to happen because I can't type or move the mouse that fast.
Esperandi
The Crusoe is the worlds first Real World Processor.
Oh, and a little PS here... trust me, I am never one to jump when Linus says so, I haven't used Linux in a couple years and I don't plan on switching to it any time soon. I like hardware compatibility and the idea programmers get paid for their work.
"RISC emulation is vastly different. Each instruction doesn't do a whole lot, but runs extremely fast. So, basically, to emulate a RISC processor, you have to translate each instruction into one or more Curose instructions that don't benefit from RISC hardware's pipelining and other efficiencies. You're going to end up with a vastly slower PC"
The way a Crusoe would handle RISC "emulation" would be very very efficient when applied to RISC processors, much more so than for CISC processors. How? Well, in a RISC chip, since the instructions are smaller, there is more repetition of them. What would be four or five CISC operations gets translated into one Crusoe molecule and is then cached and optimized. When it repeats, it doesn't have to re-translate, it simply executes the optimized version that is cached.
They've got lots of catches in the hardware of the chip that let it execute things out of order with no penalties, thereby allowing a lot more optimization. Now when they decide to re-write the code morphine module to handle a RISC instruction set it will probably take 8 or 16 RISC instructions and translate them into a single molecule which is executed in paralell.
The same applies to 3DNow and such. When a 3DNow instruction goes to the processor it has to be translated into micro-ops and executed, and then re-scheduled. Every single time. The Crusoe would perform slower the first time this instruction is executed, but afterwards would perform much faster...
Esperandi
Transmeta doesn't really have to participate in the clock speed race (BTW, if you factor price into it, Intel isn't even close right now)... A 700MHz P3 and a 700MHz TM5400 CPU do not perform the same. In a repetitive section of code (which is most of the code being executed in real world applications), the P3 just executes the code over and over re-doing the instruction translation, the scheduling, and all those things. The TM5400 "learns" what the code is doing and optimizes it down to the point where it executes that code at the same speed when stepped down to 400MHz as the P3 700 does running at full speed and full power consumption.
The example they give is a DVD player, by the time the first video frame has loaded the entire DVD-decoding process has been translated, stored in a translation cache (so it is never translated again), and optimized. If it continued running at 700MHz it would be performing like a 1GHz P3, but instead it steps down intelligently. If stepping down would lose frames or slow execution speed, it wouldn't do it... so until code really really needs over 700MHz in real world applications, Intel will not have outpaced Transmeta. Ithink they'll be around for awhile. And I don't expect this to be the only thing they do. They say they're going to be a professional R&D lab, i doubt they're just goin to keep adapting these processors to different things, they'll move on...
These facts eliminate a few of the bad things you pointed out... like what is the point of integrating 3D if all of the 3D instructions are cached and optimized? There wouldn't be much of a speed improvement if any....
Esperandi
Seriously, DNA computers are more advanced right now than quantum computers or molecular-level computers. I've got the book on them (theres only one) and they seem really promising...
Esperandi
Get ready to feed your PC.
I just got done reading the technical whitepaper on how the new Transmeta Crusoe CPUs really work, and it sounds like mixing these fuel cells with the Crusoe processors would produce one of the best mobile computing platforms in the world. Super speed with amazing abilities to conserve power added to a more plentiful power source, hurrah!
BTW, to understand why Transmeta is saying their CPUs won't perform well on existing benchmarks and will produce much better results in real life, read their whitepaper. In short, its because the processor excel at re-executing stuff, and benchmarks do things like using every single feature of a word processor once. When was the last time you did that?
Esperandi
Like the other 900 people responding to this /. article and the last, you need to learn that trademark laws are codified and real and are available at the USPTO website. Take a look at this quote from you for instance:
"Considering how much money Coca-Cola makes off of selling T-shirts, refrigerator magnets, windbreakers and the like with their "Coke" trademark on them, they aren't going to let this happen. It means a lot of money to them. "
Now, if you knew that trademarks are assigned to specific goods and services listed as a trademark is registered, you would know 2 things: #1, Coke has trademarks on softdrinks, T-shirts, magnets, windbreakers, ande verything else. The products are listed by name in a very very long list that Coke's trademarks cover. And they go no further.
#2, The Linux trademark lists ONE ITEM under its goods and services: computer operating systems. In other words, what Linus and his lawyers are doing is extra-legal at best, downright illegal at worst. His trademark does not apply to domain names, T-Shirts, hell, not even company names if they don't sell operating systems!
So no, Linus is not doing what he is legally responsible to do. He IS being a bully and trying to enforce this trademark that applies only in the situation of naming an operating system.
Esperandi
Linus is getting tricky... he must know that his trademark doesn't extend to anything except to the names of computer operating systems. He is legally bound to uphold THAT trademark, but when he goes and tries to "uphold his trademark" in things like domain names and such, he's simply going outside the law.
Esperandi
My read of the new Transmeta patent confirmed that yes, the processor will "emulate" other instruction sets... but that is not the important heart of the patent. The patent is not on a chip that translates instructions from one set to another (watch for that one, it should be coming soon), the patent is very specifically on a very important issue to operating system designers. Basically, their chip sounds like it will handle multiprocessing (not multithreading since the processes will not share their environment) and prevent process deadlock. I've never seen a study on deadlock avoidance in a situation where the instruction sets are multivarious and being cross-translated but I imagine it could get quite hairy!
Esperandi