Slashdot Mirror
Transmeta Testing Mass Production
jackstaley sent us an article about Transmeta testing mass production of its Crusoe processor. They talk about IBM (which can make copper chips, but interestingly enough, has licensing deals with Intel that should protect Transmeta from potential lawsuits) as well as exporting the production to Taiwan.
It was. Or at least the use of a normally reserved opcode to tell the morpher to switch instruction sets seemd, and the rest of the explination of the demo (like them saying there was a picoJava morpher) seemed more consistant with two morphers at once then running a x86 JVM.
There is (as far as I know) only the one production morpher. Which makes sense. Don't dilute your manpower until you have some mesure of success....
The lower power part, as I understand it, is the ability to dynamically increase and decrease power requirements as the need arises. *NO* commercial chip offers that kind of feature. That is a truly revolutionary feature. Not even restamping other procs would result in the kind of power consumption that you should be able to get with the Transmeta chips. Currently chips either run at a continuous power level, have multiple steps (Speed Step) or simply go into sleep mode. The ability to throttle back to near nothing is a great energy saver, especially when you consider that most of the time your proc is doing absolutely nothing.
I wholy agree with your point about it basically being a x86 emulator. I'll get excited when it can emulate the PPC chipset effectively. I'm more interested in the Transmeta chips from a 'what's next' point of view. i don't think this first generation of chips will be all that stunning in a price/perfomance area but the next generation or two should prove the technology to be solid, useful, and practical. And in the long run, when Intel and other manufacturers begin reverse engineering these features it'll benefit us all (I hope).
Fsck cluebie moderators. I'll say what I want, offtopic or not. And fsck having to qualify every bloody statement just
2) I believe that the "code morphing" software is loaded at boot-time, but I could be wrong here. At the very least, all the little caches and everything that cache translated code (and provide a large part of the performance) would have to be flushed every time you switched from user-space to kernel-space in your example. This would (presumably) create a large performance hit.
Anyway, I think the web site has more details.
---
I'm very surprised that this post got moderated up at all. No offense, but it doesn't provide any information, or ask any question that hasn't been asked and answered before.
The Transmeta website, cleverly located at www.transmeta.com provides temperature differences, performance numbers, technical background on the chip (It's not RISC, quite the opposite)... Further, there are white-papers for those seeking enlightenment.
If that is not enough, ars-technica (www.arstechnica.com) has done a phenomenal review of Crusoe's underlying technology.
Really, rather than rehashing all of this, once again, why not just go to the authoritative source, Transmeta, and get answers to the questions. Or did I just swallow a big fat piece of bait?
-- What you do today will cost you a day of your life.
Interesting thoughts, but they seem a bit irrelevant to me. They demoed a laptop playing Q3A on a transmeta chip at the launch. The chip is definitely out there. THere just aren't products available. So total system power consumption is still unknown, but it will of course vary system-to-system. Chip power usage is publicized, and presumably very well understood within the company.
---
From the article:
Big Blue also provides legal cover from Intel. Because of extensive cross-license agreements, only IBM and ST Microelectronics can manufacture Intel-clone chips for other designers and provide these designers complete legal immunity, said sources close to Intel.
However, from the crusoe tech page on the transmeta web site
The hardware component is a very simple, high-performance, low-power VLIW (Very Long Instruction Word) engine with an instruction set that bears no resemblance to that of x86 processors. Instead, it is the surrounding software layer that gives programs the impression that they are running on x86 hardware. This innovative software layer is called the Code Morphing software because it dynamically "morphs" (that is, translates) x86 instructions into the hardware engine's native instruction set.
So is it really a clone, or an emulator? I'm guessing Crusoe is just playing it safe
This is supposed to be great art. So why does it look like a bunch of decapitated naked people? -- Calvin
One of the demo's they had at unveling was a copy of DOOM, and x86 executable with part of the inner loop replaced by picoJava. It seemed to run that pretty quick, so it can switch instruction sets dyanmically. There is undoubtly some penality, but it apparently isn't huge.
There is no evidence that it can run Alpha code particurally fast. It has 48 (or 40?) registers internally. More then enough to do software registr renaming, and have operatinog regesters left when emulating a 4 register (10 if you count segment registes, the PC and flags) machine. The Alpha has 32 user visable registers (or is it 32int plus 32FP?). The 21264 has 80 renaming registers (invisable to the programmer) in two banks. The Crosue would have far less, so could be expected not to be able to hide as much latency.
Oh, and since the Crosue is designed to emulate the 32bit x86, it is not likely to have 64 bit registers, or a 64bit ALU. And it's FPU certonally is geared towards the x86 quirks (80 bits, almost IEEE, but not quite).
I wouldn't want to try to emulate the Alpha on a Crosue. It could probbably do a great 68000 (or the CPU32 in the palm, which is almost a 68000). Maybe even a good ARM, or V8 SPARC (these are pushing it though). Could definitly do a good JVM. Would do a great 6502, or Z80. :-)
But not a Alpha. Not a V9 SPARC. Not the 64bit MIPS. Not the IA64.
In my opinion the two biggest selling points of the Transmeta chip are it's code morphing and low power consumption. The difference in power is supposed to be on the order of 40 times less than an equivalent Mobile PIII. There was an excellent article in one of the IEEE magazine around May or so. As far as code morphing/architecture support I've only seen support for x86 mentioned.
Many 'experts' believe that Transmeta's future will not be in notebooks but in subnotebooks and the 'web pad' devices. Also consider PDAs. The last WinCE device (Cassiopeia?) I looked at had a nice 200Mhz CPU but also had to be recharged every 14 hours. If Transmeta can offer twice the maximum cpu power and 2-3+ times the operational length and do it cheaply, manufacturers will jump all over them.
Everyone's playing wait and see though. Well, IBM announced a Transmeta laptop a couple months ago but I haven't seen anything about it since.
Fsck cluebie moderators. I'll say what I want, offtopic or not. And fsck having to qualify every bloody statement just
Does this mean that the Crusoe-based products that we were made to think are currently in the latter stages of production are in fact not, since only now are the details of production being arranged?
Or does it mean that the Crusoe-based products that are in the latter stages of production are not based on mass-produced Crusoe processes, but on (expensive) non-mass-produced ones instead?
Neither alternative sounds very promising. Is there another?
"The question of whether machines can think is no more interesting than [] whether submarines can swim" - Dijkstra
There's a talk by Dave Ditzel on the Crusoe here./ ee380/000209-ee380-100.asx
htt p://stanford-online.stanford.edu/courses/seminars
Great talk. Learn the facts.
That the technology behind Transmeta's Crusoe chip is great is not really open to doubt, but what is IMHO open to doubt is the speed at which Transmeta is moving at. It's been close to a year now since the chips were launched, and yet we still haven't seen the Crusoe in anything and Transmeta are still engaged in production negotiations.
At this rate it appears as though by the time that Transmeta get something finally made and used its competitors will have moved beyond it. Time to market is a significant factor in the computing industry, and it looks to me like Transmeta are just taking too long to deliver.
If they don't get their act together soon, the Crusoe will become vapourware because nobody will be using it.
Now, that's all fine and dandy, but isn't this just a little out of sorts in that you can get a run-of-the-puppymill wIntel chip that does the same thing, but runs cheaper? Of course this is not the only factor, there are a couple of questions, the answers to which would completely change the tides in my personal opinion of the usefulness of the Crusoe chips:
1, is there a significant heat/power consumption difference from an x86 chip (that works at a speed comparable to the Crusoe running x86 code, we'll give Crusoe a chance here)
2, can the Crusoe chip execute code written in difference instruciton sets at the same time (well, as much as any other popular single-pipe processor executes multiple threads), eg can it run Crusoe-optimized x86 Debian running an application compiled for Alpha Linux?
I haven't seen hard facts on either of these. Opinions and facts would be appreciated - I can't find them. Honestly, though, I don't think the Crusoe is going to be the end-all of portable chips that the article's author seems to indicate it will be. I think we still have a couple of (chip) generations before everyone has their own personal wireless web pad (tm).
funny munging
funny munging
First I beat one pence coins flat, cut them into triangles and then dip them in that lovely ferric citrate extra-spicy salsa.
:)
So sue me