Posted by
ryuzaki0
on from the they-will-unveil-it dept.
Magorak writes "ZDNet is running an article on the upcoming PC Expo show in New York which is going to feature new notebooks from NEC and IBM which will have Transmeta's TM5400 processor in it. There's mentions of Transmeta's background, future company partnerships, and other goodies."
And what shall be unveiled.....
by
Anonymous Coward
·
· Score: 5
Deep within the bowels of Transmeta headquarters, Linus slaves away at his workstation, having been bound to his desk for the past month. One thing he was sure of, as they had promised Transmeta were sticklers for deadlines. But they fed him, and they only beat him four times a day, which is better than alot of engineers got these days. However, he thought, sweat pouring down his brow, this time is different.
The skies had grown black around the Transmeta towers. Everyone knew they were up to no good, so to speak, but no one knew the unspeakable horrors they intended to release upon the world. The only man who understood, and who knew the full ramifications of their actions was Linus himself, and he was in no position to do anything. But, he thought, I know who can.
He muttered the incantation quietly, so as not to trigger the tape recorder set to record whenever someone spoke in the room. Such precautions seemed unnecessary at first, but Linus had not anticipated the true nature of Transmeta when he had taken the job. Now he spent days translating arcane texts that would drive most men mad, and coding the software to drive the most diabolical machine man had ever dared to construct. It was by only the vaguest definitions a computer, driven by processors in a beowulf-style cluster. But there were no disks, nor any traditional interfaces. The processors drove masses of flesh and neurons, specifically designed with one purpose in mind. To summon onto the earth the most concentrated forms of evil known in the universe, and to attempt to bind them to Transmeta's will. However, Linus, having read most of the occult texts, knew this could never be, and the summoning alone would be enough to end mankind's existance as they know it.
The air crackled with the energy from Linus' spell. A warm, muggy feeling took over the sterile cold basement chambers. He was coming.
Suddenly, a vague shape took form in the corner of the room. Two eyes formed in the center of the wavering mass. In a low gutteral voice, it intoned "why have you brought me, Linesus of Linuux", using Linus' true name.
"you must stop Transmeta before they unveil their plan" Linus pleaded, knowing the great being's power would prevent the tape recorder from kicking on.
"I understand your problem, and I fully agree that this evil must not be unleashed. However, I require two things."
"ANYTHING" Linus cried.
"for my colleague, I require a certain young Portman," gnarphlager slowly stated, "and for myself, I require cheese . .."
Perhaps not reading the article is why you haven't seen this question answered?
Those frustrations are weight and battery life, he said. "Frustration studies" done by IBM have shown that users want a notebook weighing about 3 pounds, with eight hours or more battery life, he said.
"If we can do that, we'll bring it out in the fall," Suarez said. "We're pretty confident that we can get close to that eight-hour mark."
Re:Do we only care because it's Linus?
by
dtr21
·
· Score: 5
I agree that a lot of the fuss here has come about because Linus is involved in Transmeta. But there's a lot more to their technology than Intel's speedstep.
Processors with multiple execution units have become very common. They work by looking for instruction level parallelism - in short, if 2 instructions use totally different registers, whose operands have already been computed, then you can execute them in parallel.
Modern processors tend to support out of order execution and use branch prediction to avoid wasting clock cycles. Both of these, combined with several execution units, make your processor very fast - but use a huge amount of silicon area (read: lots of power, space that could otherwise be used for cache, etc). Their other big problem is that they look at the program from scratch each time it is executed - which means they can't avoid old mistakes.
The idea that Transmeta had is as revolutionary as the early RISC philosophy. The original observation that lead to RISC was that compilers don't tend to make use of the more complicated CISC instructions (string operations, polynomial evaluation on VAX, etc) and that these operations require a lot of hardware to implement, and as such slow down the processor. The RISC goal was to use a small number of relatively simple instructions. This means that you need fewer transistors on the processor, which means that it uses less power. Having fewer transistors generally makes the processors critical path shorter as well, which allows it to be clocked faster. In addition, the simple RISC instructions allow easy pipelining - which in itself leads to a huge speed benefit.
However, modern RISC processors are much more complex than their ancestors. They support many more instructions, and implement things like dynamic instruction scheduling. This has lead to RISC chips using more and more silicon.
Transmeta's idea is to go through the same process of reducing the complexity of the processor once more. They save a lot of silicon by shifting work onto the compiler.
If you think about it, this makes a lot of sense. The compiler is responsible for allocating registers, so it already knows about instruction level parallelism, and therefore it knows which instructions can be executed in parallel. If you let the compiler tell you this, rather than working it out in hardware every time you run your program, you can potentially save a lot of wasted silicon.
Transmeta went for code morphing because it saves people having to re-compile code to run on their processor. They gain the benefits describes above, without the huge cost of trying to replace the x86 instruction set. Their code-morphing engine effectively translates x86 (ar any other architecture they want to implement) into native instructions. It finds instruction level parallelism, and exploits this in native code (the Transmeta can execute 4 32-bit instructions simultaneously - although you are limited in what operations can be caried out at the same time).
By having the compiler do all of the work in scheduling instructions, they allow a huge amount of silicon to be saved - therefore reducing power consumption and allowing faster clock speeds.
There are more benefits. The code morphing engine can heavily optimise frequently used blocks of code. Potentially, they could store information about the behaviour of a jump instruction in their code, to allow them to do much better branch prediction. If you've seen the program execute 500 times, you've got a pretty good idea how it's going to behave next time in a lot of cases.
The point is that the x86 architecture has been a problem for a long time. It's old, slow, and difficult to optimise. What Transmeta have done is to find a nice, neat way to eliminate the problems caused by old ISAs. Effectively, they use a 2 stage compilation. Sun do this with Java - compile code to Java Binaries (platform independant) and then use a native Virtual Machine to run these. Transmeta are applying a similar concept, in a radical new way to their processor. The x86 instruction set is being used as the platform independant middle stage in their model. It gets compiled and optimised to native code by their code morphing engine. They've also been able to save a lot of silicon in the process.
And, since their code morphing engine knows a lot more about the machine it's running on (cache size / associativity, memory size, etc) it can probably do a better job of compiling code than most x86 compilers can. Gcc and the like are good - but if they have to support machines with many memory sizes, configurations, and know nothing about caches and the like, then there's only so much they can do.
Make no mistake, I believe that Transmeta has revolutionary ideas that will change the face of computing. No longer will legacy hardware be a problem for chip designers. Using the hybrid approach developed by Transmeta will allow faster and faster processor designs, taking full advantage of modern ideas, whilst still supporting legacy ISAs. It's a fantastic concept, and I hope they go all the way with it.
If you want a long-running laptop and don't want to wait for Transmetta, I would suggest looking into the iBook. (Anti-Apple folks, please read on before you turn on the flame-throwers...)
Sure, it's butt-ugly (white with your choice of blue or orange), big (11.6" x 13.5" x 2.1"), and weighs a lot (6.6 pounds with the battery), but it's...
fairly cheap ($1599, last time I checked), fairly fast (G3 chip w/ full cache... much faster than most PIII notebooks, which are crippled to reduce heat), has a really nice-looking LCD screen, has the cheapest wireless networking option of anybody out there, and has a battery that lasts 5-6 hours (ymmv).
Best of all, thanks to LinuxPPC, they are great Linux portables.
No PCMCIA, but most of the stuff that you would use cards for (Ethernet, wireless networking, modem, USB) are already built in, so that's no great loss for most users.
Of course, if you can't get past how goofy it looks, maybe the Transmetta laptops are worth the wait.
Either way, I would hate to pay much for a Pentium-based laptop. They are slow, hot, and suck power too quickly.
Slashdot Mirror
Deep within the bowels of Transmeta headquarters, Linus slaves away at his workstation, having been bound to his desk for the past month. One thing he was sure of, as they had promised Transmeta were sticklers for deadlines. But they fed him, and they only beat him four times a day, which is better than alot of engineers got these days. However, he thought, sweat pouring down his brow, this time is different.
."
The skies had grown black around the Transmeta towers. Everyone knew they were up to no good, so to speak, but no one knew the unspeakable horrors they intended to release upon the world. The only man who understood, and who knew the full ramifications of their actions was Linus himself, and he was in no position to do anything. But, he thought, I know who can.
He muttered the incantation quietly, so as not to trigger the tape recorder set to record whenever someone spoke in the room. Such precautions seemed unnecessary at first, but Linus had not anticipated the true nature of Transmeta when he had taken the job. Now he spent days translating arcane texts that would drive most men mad, and coding the software to drive the most diabolical machine man had ever dared to construct. It was by only the vaguest definitions a computer, driven by processors in a beowulf-style cluster. But there were no disks, nor any traditional interfaces. The processors drove masses of flesh and neurons, specifically designed with one purpose in mind. To summon onto the earth the most concentrated forms of evil known in the universe, and to attempt to bind them to Transmeta's will. However, Linus, having read most of the occult texts, knew this could never be, and the summoning alone would be enough to end mankind's existance as they know it.
The air crackled with the energy from Linus' spell. A warm, muggy feeling took over the sterile cold basement chambers. He was coming.
Suddenly, a vague shape took form in the corner of the room. Two eyes formed in the center of the wavering mass. In a low gutteral voice, it intoned "why have you brought me, Linesus of Linuux", using Linus' true name.
"you must stop Transmeta before they unveil their plan" Linus pleaded, knowing the great being's power would prevent the tape recorder from kicking on.
"I understand your problem, and I fully agree that this evil must not be unleashed. However, I require two things."
"ANYTHING" Linus cried.
"for my colleague, I require a certain young Portman," gnarphlager slowly stated, "and for myself, I require cheese . .
Those frustrations are weight and battery life, he said. "Frustration studies" done by IBM have shown that users want a notebook weighing about 3 pounds, with eight hours or more battery life, he said.
"If we can do that, we'll bring it out in the fall," Suarez said. "We're pretty confident that we can get close to that eight-hour mark."
I agree that a lot of the fuss here has come about because Linus is involved in Transmeta. But there's a lot more to their technology than Intel's speedstep.
Processors with multiple execution units have become very common. They work by looking for instruction level parallelism - in short, if 2 instructions use totally different registers, whose operands have already been computed, then you can execute them in parallel.
Modern processors tend to support out of order execution and use branch prediction to avoid wasting clock cycles. Both of these, combined with several execution units, make your processor very fast - but use a huge amount of silicon area (read: lots of power, space that could otherwise be used for cache, etc). Their other big problem is that they look at the program from scratch each time it is executed - which means they can't avoid old mistakes.
The idea that Transmeta had is as revolutionary as the early RISC philosophy. The original observation that lead to RISC was that compilers don't tend to make use of the more complicated CISC instructions (string operations, polynomial evaluation on VAX, etc) and that these operations require a lot of hardware to implement, and as such slow down the processor. The RISC goal was to use a small number of relatively simple instructions. This means that you need fewer transistors on the processor, which means that it uses less power. Having fewer transistors generally makes the processors critical path shorter as well, which allows it to be clocked faster. In addition, the simple RISC instructions allow easy pipelining - which in itself leads to a huge speed benefit.
However, modern RISC processors are much more complex than their ancestors. They support many more instructions, and implement things like dynamic instruction scheduling. This has lead to RISC chips using more and more silicon.
Transmeta's idea is to go through the same process of reducing the complexity of the processor once more. They save a lot of silicon by shifting work onto the compiler.
If you think about it, this makes a lot of sense. The compiler is responsible for allocating registers, so it already knows about instruction level parallelism, and therefore it knows which instructions can be executed in parallel. If you let the compiler tell you this, rather than working it out in hardware every time you run your program, you can potentially save a lot of wasted silicon.
Transmeta went for code morphing because it saves people having to re-compile code to run on their processor. They gain the benefits describes above, without the huge cost of trying to replace the x86 instruction set. Their code-morphing engine effectively translates x86 (ar any other architecture they want to implement) into native instructions. It finds instruction level parallelism, and exploits this in native code (the Transmeta can execute 4 32-bit instructions simultaneously - although you are limited in what operations can be caried out at the same time).
By having the compiler do all of the work in scheduling instructions, they allow a huge amount of silicon to be saved - therefore reducing power consumption and allowing faster clock speeds.
There are more benefits. The code morphing engine can heavily optimise frequently used blocks of code. Potentially, they could store information about the behaviour of a jump instruction in their code, to allow them to do much better branch prediction. If you've seen the program execute 500 times, you've got a pretty good idea how it's going to behave next time in a lot of cases.
The point is that the x86 architecture has been a problem for a long time. It's old, slow, and difficult to optimise. What Transmeta have done is to find a nice, neat way to eliminate the problems caused by old ISAs. Effectively, they use a 2 stage compilation. Sun do this with Java - compile code to Java Binaries (platform independant) and then use a native Virtual Machine to run these. Transmeta are applying a similar concept, in a radical new way to their processor. The x86 instruction set is being used as the platform independant middle stage in their model. It gets compiled and optimised to native code by their code morphing engine. They've also been able to save a lot of silicon in the process.
And, since their code morphing engine knows a lot more about the machine it's running on (cache size / associativity, memory size, etc) it can probably do a better job of compiling code than most x86 compilers can. Gcc and the like are good - but if they have to support machines with many memory sizes, configurations, and know nothing about caches and the like, then there's only so much they can do.
Make no mistake, I believe that Transmeta has revolutionary ideas that will change the face of computing. No longer will legacy hardware be a problem for chip designers. Using the hybrid approach developed by Transmeta will allow faster and faster processor designs, taking full advantage of modern ideas, whilst still supporting legacy ISAs. It's a fantastic concept, and I hope they go all the way with it.
Sure, it's butt-ugly (white with your choice of blue or orange), big (11.6" x 13.5" x 2.1"), and weighs a lot (6.6 pounds with the battery), but it's...
fairly cheap ($1599, last time I checked),
fairly fast (G3 chip w/ full cache... much faster than most PIII notebooks, which are crippled to reduce heat),
has a really nice-looking LCD screen,
has the cheapest wireless networking option of anybody out there,
and has a battery that lasts 5-6 hours (ymmv).
Best of all, thanks to LinuxPPC, they are great Linux portables.
No PCMCIA, but most of the stuff that you would use cards for (Ethernet, wireless networking, modem, USB) are already built in, so that's no great loss for most users.
Of course, if you can't get past how goofy it looks, maybe the Transmetta laptops are worth the wait.
Either way, I would hate to pay much for a Pentium-based laptop. They are slow, hot, and suck power too quickly.
Information wants to be anthropomorphized.