Slashdot Mirror
Transmeta Claims Five Year Lead Over Intel/AMD
safariman writes: "An article on Yahoo news reports that Transmeta is claiming a five year lead over Intel and AMD. Does anyone else think this claim is a bit excessive? After all, Transmeta itself is not five years old. Besides, once an idea is public, it is a lot easier to copy."
If the strategy of the Transmeta chips is to emulate other CPUs, its kind of ironic that they will be using "5 year off technology" to emulate present CPUs?
;-)
Thats like using a Pentium III as basically a really fast 386.....er....um.....nevermind
Comment removed based on user account deletion
Besides, once an idea is public, it is a lot easier to copy.
...and illegal if they hold patents. Course, I can forgive you if you weren't aware they had any :)
This is a luidicrous statement. No technology is that much ahead. that would be like the AMD having the athlon at the same time as Intel came out with the P200mmx. It is basically inconceivable that Transmeta would be that much ahead of industry leaders like Intel, and AMD. -wyn
Perhaps this is an effort to help build up their stock value, along with a general PR effort.
Transmeta now has a five year lead over Intel in the announcement of products that I cannot purchase yet
I thought M$ was the leader in this field.
Anyone remember microcode? You could put your CPU control unit into a set of microinstructions in ROM that would tell your ALU what operations to take and you wouldn't have to design a complex controller. The above sounds similar. Is that, essentially, what Crusoe does? I know it is a lot more complex than the Mircocode of the 1970s and 1980s but one of the coolest aspects of Microcode is that you could emulate other instructions and so it made it easier to make a CPU compatible with earlier units.
It seems to me like Crusoe is a very advanced implementation of microcode, but purely in software. Correct me if I'm wrong but isn't one of the primary features of Crusoe that it emulates the Instruction Set of different processors, such as x86, in Software?
How is that 5 years ahead?
5 years is a bit of a stretch, but on the other hand, they may have a significant lead. Of course, to admit that they have a lead is to concede that their technology is the right way to go. If you accept that, then, yes, it will take the other chip makers quite a while to switch to that type of processor design. Consider that it takes Intel several years to develop each chip generation, and that's when the fundamental design technology is similar.
So it could well take Intel or AMD 5 years to develop a processor that used the same technology as Transmeta has now. Of course, that's no reason to believe that Transmeta will succeed in making competitive processors.
Quote:
I have yet to see a product out there that uses a processor from Transmeta. Aside from that, this is just good old FUD-spewing by the boys who made Crusoe. Both Intel (with their P6 core) and AMD (since the K6 days) have stopped executing x86 instructions directly. They do hardware realtime translation of x86 instructions. I'm going to guess that Crusoe does this as well, but partially in software (much like the first PPC-based Macs).
I have yet to see any benchmarks on battery life or processor performance either. I don't care if the battery on my Crusoe-based notebook lasts 1hr more than my Intel-based Thinkpad, if the bloody thing runs at 486-speeds. To substantiate the claim, I would at least like to see performance benchmarks!
--
This is like me claiming I have a three year lead in a computer engineering degree over a physics major... so what? Transmeta is just trolling, and I'm sure their competitors will continue to ignore them most of the time, as well they should.
If you are modding me down because you disagree with me, use the "Flamebait" category, not the "Troll" one.
Most PR stunts like this are done because:
a) They're looking for more funding from VC.
b) They're going to make an IPO soon, so they are trying to get their name out into the investment community.
c) They want to boost consumer opinion of their product (or non-product, as the case may be).
d) They are trying to intimidate another company (e.g., AMD) into working out some sort of deal behind the scenes (which we know nothing about)...
I'm still waiting to see a Crusoe chip; if it's everything they say it is, then Transmeta can crow all they want.
Lucas
--
Spindletop Blackbird, the GNU/Linux Cube.
On the other hand they have a chip design with a billion gated clocks - not something you can do to an existing design overnight (except at a very gross level) - so in the sense that it will take one design cycle for the big guys to be doing what they are
On the other hand all it takes is another small startup to get an async logic x86 clone to market - for those who don't know asynchronous logic has held a promise or lower power, faster design for years - but the CAD tools don't support it - a number of async designs have been done including Amulet (an async ARM).
Async CPUs are in effect clockless - everything internal is self timed, nets only switch when they need to saving power and, in effect self-clock-chipping :-)
Once something has been done would it take you as the creator took to create it. It really depends. Would take Intel or AMD 5 years to create a "clean room" of transmeta's code morphing technology? I would think it would take them less time maybe 2 years most 3. But you never know. It could be so complicated that it will take 5 years to recreate.
No, hardware is only faster than software at doing simple things. CPUs are built (mostly) out of transistors and the more raw output you can produce with the least transistors, the better off you are (usually). That said, RISC systems were designed because some smart engineers realised that they could pull off higher CPU speeds if they simplified CPU design and put the onus on the compiler to generate complex code, instead of using a complex chip.
... and the move to software is a good idea, but can be emulated quite quickly with enough programmers. That is, AMD or Intel just has to take their translation hardware and write code that does the same thing (if they wanted to do it).
Transmeta has just done something similar -- put the onus on the translation software to optimise a given program in a given instruction set for their CPU.
The problem is that the big guys (Intel and AMD) are already doing this in hardware
- Michael T. Babcock (Yes, I blog)
Me: Ok, so you are 5 years ahead of Intel, right?
Transm: Yes, that's right. And AMD.
Me: But this technology, you've only been working on it for less than 5 years.
Transm: Correct. We were ahead even then.
Me: Ok, so Intel decided that want to be like you tommorow, it is still going to take then 5 years just to get to the point you are at right now?
Transm: yes.
Me: What makes you so great?
Transm: Linus Torvalds. He made linux in less than 5 years, too.
Me: No no no. Linus only makes you famous.
Transm: Well, I don't see Linux Torvalds working for Intel, do I??
Me: Or AMD. That's not the point. I don't see why Intel is going to take longer than you to do something when they are bigger.
Transm: Ok, five years is a long time. Doesn't that impress you??
Me: No! It doesn't matter if you are lying!
Transm: What?? We really do have Linux Torvalds! That is not a lie!
I can't spell or type, but that doesn't mean I'm unusually stupid.
Transmeta seems to be succumbing to the same thing AMD succumbs to: marketing arrogance. Remember those claims by AMD that the Athlon was faster? And remember how they were shot down when it was found that a similarly clocked P3 could perform better in high-demand applications? Sure was a wake-up call to AMD's marketing team; they couldn't use faster performance in their ads 'cause it just wasn't true.
Also, who the hell names these add-on instruction sets? At least Intel uses some variety in names, but "3d-Now!" and "Power-Now!" just seem to reflect the "instant gratification, damnit!" mentality of the general AMD user base. These names sound more like ransom demands from a terrorist faction, which I'm starting to believe is what AMD is quickly becoming.
"Ancillary does not mean you get to rule the world." --U.S. Circuit Judge Harry Edwards, speaking to the FCC's lawyer
We all know the marketing strategy before an IPO:
Everything is about to generate buzzes and keep up the momentum.
So, are you ready to buy machines with Crusoe chip, and throw in your money to invest in the company?
No? How come?
The quote this whole thing's based on says, ``For them to catch up, they would also need a software based approach. That means they would have to start from scratch and from my 20 years of experience, it would take at least five years to get a new microprocessor out the door."
Is anybody here seriously suggesting he is wrong to say that it would take five years to start from scratch designing a chip - especially when that's how long it took Transmeta? That's where the five years come from and I would have thought that's accurate. People have had decades to copy Intels X86 architecture (and several years for the Pentium and most of its guises) so if it was that easy there would be more alternatives.
Which means the question is the first point: do Intel/AMD/ARM etc have to use a design which implements software? Well, that I don't know. But considering the problems and delays Intel has had bringing the Pentium 4 to market there is evidence to say he might be right.
In the context of promoting his own company he's not making unreasonable assumptions.
On the other hand, it would be nice to market a "student version" of Crusoe.
Students could design their own architecture, up to the instruction set, including a MyArch -> Crusoe instruction translator, and then let Crusoe execute their code.
Also interesting would be a console emulator (SNES!) designed to run into Crusoe.
--Pazu
...that they feel their technology is that far ahead of their competitor's technology...
They aren't saying they know what the technology will be in five years...
BlackNova Traders
They don't see an upper MHz limit for Crusoe. The aim is to go for a simple architecture which produces little heat for the performance, and then up the performance. If they stay in business, Moore should be on their side. Right now they cannot get the performance they want though.
I think SMP would be hard. But I could be wrong.
They can do other instruction sets. Eventually they would like to do multiple instruction sets in parallel. (Think moving the JVM into the machine.) However they will likely have a harder time squeezing performance out of RISC than CISC.
I don't know the status of 64-bit. Internally their chip is 128 bit though, so it should be doable. But I think they prefer AMD's approach. (I think that Microsoft will discover the hard way that AMD left them an upgrade path they can live with while Intel did not - the barrier to entry that Microsoft erected is working against IA64 now.)
I believe that Transmeta will do something extreme. If they can hang on, they have enormous potential. But if they cannot survive this critical period, they will leave a hole in the ground.
Oh, the one technical detail everyone seems to ignore. There have been many micro-controller architectures. But x86 was not intended to be one and is rather difficult to emulate. They claim to have real breakthroughs and I believe them.
Cheers,
Ben
My usual seat in the cluetrain is at A HREF="http://pub4.ezboard.com/biwethey.ht
INTL is Inter-Tel Inc. Found it on yahoo.
INTC is Intel.
If at first you don't succeed, skydiving is not for you.
Intel is INTC, not INTL. Here's the corrected link: INTC
// zyqqh
IIRC AMD is already in cahoots with Transmeta to make lower power comsumption CPUs. From what I read, AMD wasn't all that impressed with marginal gain.
--
Chief Frog Inspector
A feeling of having made the same mistake before: Deja Foobar
Transmeta's design, while not revolutionary (IBM has been doing microcode on their mainframe processors for decades) is something new in the microcomputer space. However, consumers typically don't buy CPU's for their design -- they buy strictly on price/performance. This is why AMD is currently slaughtering Intel at the low end: even though their chips are just imitations of Intel's, their process and scale are set up so that they can deliver large yields at low prices.
Transmeta can win on price/performance if it can get good production yields.
The real test will be on 64-bit, though: when Intel finally releases the Itanium, how fast will Transmeta be able to re-code its magical morpho-chip to the IA-64 instruction set? If they can do it quickly, they'll have a production Itanium-compatible chip on the market with a significantly lower R&D cost, because the hardware part of the chip is already paid for. Intel will have to charge a super-premium for Itanium, because of all the zillions they spent building it. If Transmeta can do a software-only upgrade to the Crusoe to make it Itanium-compatible, they'll be able to sell it at less than half of what Intel will be charging. If they can do that, they've got it made -- and they will indeed be five years ahead of Intel.
--
Tired of FB/Google censorship? Visit UNCENSORED!
AMD or Intel just has to take their translation hardware and write code that does the same thing (if they wanted to do it). Not the same thing. Hardware does one thing, fast; software can be more flexible. Transmeta's does lots of things that hardware cannot - branch probability marking, incremental compilation, etc.
Preferential Voting: easy as 1-2-3
I have to agree with you on this one Lucas, looks
more like a PR move than anything.
My big problem with it is that they chose such a
stupid time frame to claim that they were ahead
by and provided nothing but their own word
as evidence that they were ahead.
Perhaps if I really felt motivated to find out
about them beyond that single article, I might
just find that, but their initial statement has
turned me off toward them in general.
Anyone who would make outrageous claims to bolster
business should've gotten into disk jockeying
or rock stardom, because that's where
that sort of PR belongs.
I'm just mean today, I guess... oh well!
~- Llah -~
Really, I know some people have been dissing Transmeta (TMTA in October, when they IPO) for:
1. running slower than Intel
2. doing PR prior to the IPO
But. It doesn't matter.
Yes, you read that correctly. Noone gives a frog's hind quarters about the speed anymore, it's HOW LONG IS THE BATTERY LIFE. And they can deliver substantial gains on battery life for laptops and webpads, so they are 5 years ahead.
And they're not that slow either, pretty close in speed and they are shipping in quantity to a large number of disparate vendors.
Besides, most of us just care about Bandwidth, Batteries, Butting Up Against Bill And Beating Him Silly. And on all three they win round one - better than the US Govt, that's for sure!
Will in Seattle
Transmeta has chips that are kinda slow. But if their burn rate doesn't exceed the speed at which VCs throw money at them, their CPUs will get faster.
SMP's hard, but then I'm an idiot, so what do I know.
Transmeta might try other instruction sets. But they might not work very well.
I have no idea what I'm talking about now, but I'll throw in the term "64-bit" 'cause it makes me look 1337. Oh, BTW, Intel and AMD don't get along. MS will probably make an OS compatible with at least one of them. But I might be wrong and MS might start selling bedwarmers for elderly people. Mmm... bedwarmers.
I believe that Transmeta might make lots of money, but not if they go bankrupt.
Guess what? x86 wasn't a MPU architecture - it was a recipe for lasagna. That makes it hard for Transmeta to emulate. But hey, they say they've got their act together, know where their towels are, and are going to make some insanely great lasagna.
But then, I might be wrong.
Now to understand the context, keep in mind that Transmeta does not see itself as a head to head competitor with Intel or AMD in the x86 market. Transmeta is really going after the embedded space and the mobile computing space. While notebook manufacturers are intending to implement Crusoe, Transmeta is really targeting the Palm sized computers, the mobile phones, etc.
What Ditzel is saying then is that Transmeta has a Quantum leap on AMD and Intel in this area. Transmeta's technology allows them to shave off 1/4 to 1/3 of the transistors needed for a CPU. If, and this is a big if, Transmeta's technology scales down (not up) they have the potential to be the embedded king of the processor because, in theory, the chips of the competition will always be more complicated.
Now, I don't know if Ditzel is right on this. Intel's StrongARM looks mighty fine in comparrison. I'm still waiting for Rebel (formerly Corel) to come out with a poratable Netwinder around the StrongARM. The Netwinder desktop runs a nice little Linux desktop class machine on 15 Watts. That's less juice than some x86 CPUs alone (let alone the hard drive, the fan, etc.). And of course if Palm does move to the StrongARM as they are rumored to be doing, it will get very interesting....
I'm not counting Transmeta out, just not holding my breath for them to achieve world domination. It seems like they've got a decent product and given the slow acceptance of non x86 CPU's, they might have a good shot at gaining enough marketshare to make some money.
have a day,
-l
This is all under the assumption that AMD and Intel haven't got any secret dark projects that they are hiding from the media. Just because Transmeta has been whored around by the techworld (primarily slashdot) doesn't mean that the bigger companies (the man) are doing nothing new and unique. This is to be expected from slashdot, though, because if you follow the money, you'll probably find out a lot of the higher ups in VA are going to invest in Transmeta.
:) It's about money. If you still think it's about freedom, leave the 90's, and join us in 2000. (or support Debian)
Oh, freedom and software? I'm sorry, it's not about that anymore
If you are that annoyed by Slashdot, then I must ask: Why are you still here?
Go away. Slashdot isn't yours. It has no responsibility to you. The creators of Slashdot don't care about you, or your opinions. If you really want something better, go find it or start it yourself. But by clogging up story threads with crap like this, you're only making yourself look silly.
_______________
you may quote me
I could have sworn that as part of the Intel vs. Digital lawsuit that Digital sold their fab to Intel. Perhaps, Compaq is now fabless, but I don't believe Samsung is.
have a day,
-l
that's right. microsoft wasn't the first to the market - it was apple or digital research or IBM, depending on which particular market you're referring to, and which version of history you believe.
and transmeta's not the first to their market, either. they're following intel, AMD, motorola, etc into the processor fray with something new and different. they're not defining a new market so much as redefining an old one.
of course, the point is moot if they can't deliver.
- Entertaining Bits from the Ancient Kernel Tree
I agree with the other reply - this is the only post here that clarifies the article.
"There is more worth loving than we have strength to love." - Brian Jay Stanley
This post is five years ahead of the rest of you. This means that by the time you will be replying or modding this, it will already have lost my interest. Have a nice life.
People replying to my sig annoy me. That's why I change it all the time.
Async does have major promise, especially for low-power. However, if you just sprinkle magic async dust over an existing chip design, you get a processor that is only as fast as its poorest transistor. Transmeta, which has a fraction of the transistors (and a head start on the software which could adapt more intelligently to the vagiaries of a clockless chip), would benefit far more from this transition than traditional x86 designs.
In effect, you have parallel advances. There's fabrication technology, where the big guys have the advantage, but that's just a matter of money. There's sync/async, which hasn't been opened up yet. And there's software, where Transmeta does indeed have a head start of 5 years... minus however long intel/AMD have been running secret initiatives to do Transmeta-like tricks in software.
Preferential Voting: easy as 1-2-3
On the other hand, Intel/AMD didn't hire the Rather Bright group of engineers of whom Linus Torvalds is merely one in a cast of dozens.
On the gripping hand, Transmeta has obtained patents on many of the more interesting technologies that they developed in the process, thus meaning that Intel/AMD would need to work around them, which I expect is an underlying assumption in the assertion that it would take some years to replicate it all...
If you're not part of the solution, you're part of the precipitate.
VLIW is pretty close to software based CPU functions. It essentially pumps out the out-of-order execution and completion units to software.
Sure, Transmeta pushes out some more things out of CPU, but it's hard to argue it's a 5 year lead.
sigs are a waste of space
Well, at least Intel doesn't make insipid commercials just to appeal to the morons of society. Real computer users know to stay away from AMD and their "92.813% x86 compatibility" goal. In the computer world, you get what you pay for. If you seem to have struck a deal, you always end up paying for it because of some bug that the AMD designers could've avoided in the first place.
"Ancillary does not mean you get to rule the world." --U.S. Circuit Judge Harry Edwards, speaking to the FCC's lawyer
That's a very strange perspective.
Intel has already developed a next-generation architecture-- IA64 -- so any statement that they can't develop a next-generation architecture is just plain wrong.
Yes, I'm aware that Intel is having a horrible time delivering IA64-- but you can't question their commitment to IA-64, which I would have to call a "radical shift" from their previous approach.
The issues that I think would be there with SMP would involve synchronizing locks etc. They are trying to do a lot of out of order operations, falling back on careful code when their optimizations fail. That keeps their pipeline full. But to get to a safe lock they need to flush that to get to a state where they can. Which will lose performance.
Cheers,
Ben
My usual seat in the cluetrain is at A HREF="http://pub4.ezboard.com/biwethey.ht
One of the biggest barriers everyone is facing right now is having the chip melt. It is a constant battle.
Transmeta is not facing that particular barrier. Plus their chip is simpler hence easier to iterate through generations.
Therefore over time the performance gap should narrow and reverse, if only they can stay alive.
Cheers,
Ben
PS This opinion is based on a co-worker's conversation with some key people at Transmeta last Feb.
My usual seat in the cluetrain is at A HREF="http://pub4.ezboard.com/biwethey.ht
and I suppose you thought the "flat police" commercial was tactful? How about that stupid Athlon commercial which had their claim of being faster? This is the claim which I challenge.
"Ancillary does not mean you get to rule the world." --U.S. Circuit Judge Harry Edwards, speaking to the FCC's lawyer
You're just too afraid to face the fact that the Athlon was outperformed by a lesser-clocked Intel chip. Did you read any of the other pages of that review? You were probably too afraid to do so, since you couldn't stand seeing a hardware review site trash talking der Fuhrer AMD GmbH. Take those blinders off your eyes, and rip that green swastika armband off of your khaki shirt.
"Ancillary does not mean you get to rule the world." --U.S. Circuit Judge Harry Edwards, speaking to the FCC's lawyer
If Transmeta can do a software-only upgrade to the Crusoe to make it Itanium-compatible, they'll be able to sell it at less than half of what Intel will be charging.
Not.
The Code Morphing(TM) emulation technology used in Crusoe(TM) processors is designed to emulate CISC (complex instruction set computing) machines such as x86, 68k, and JVM. It's much harder to emulate RISC machines such as MIPS, Alpha, and PowerPC, let alone machines that are already VLIW like ia64.
<O
( \
XGNOME vs. KDE: the game!
Will I retire or break 10K?
When my Palm III (to take a nice cache-less example) wants to read a word from RAM it asks for the address and then waits (I think) 3 clock cycles. If it read it any sooner, the memory chip might not be ready yet - the 0's don't change to 1's in a nice ideal square wave, they relax there exponentially, and if the processor reads something that's halfway from 0 to 1 who knows what it will think.
Why 3 clock cycles? Because that's the worst case - if I have a relatively crappy RAM chip, and my battery is close to dead, and it happens to be a really hot day, that's how long it will take. But what if my processor watched the voltage carefully enough to know when the memory was ready? After all, if the voltage reaches 1 after two clock cycles, you know that's where it's supposed to be.
And once you start thinking that way, why do you need a clock at all? You ask the memory for a value, and wait only long enough until it's ready; ask the addition ciruitry to add that number to register n, and wait only long enough until it's ready; ask the pipeline for your next instruction, etc. If async were a factory, instead of having some fixed-speed conveyer belt running through it, you just have individuals handing each other tasks, and a very careful setup so nobody starts to do anything until all the materials they need arrive at their station.
This obviously takes some high-level wizardry to design. Instead of just doing the next step on the next digital tick of the clock, each part of the chip must make sophisticated analog judgements about when its input is ready and when it's possibly still just a glitch. But what you get is a chip that overclocks itself to exactly its own limits. You'd sell computers rated to "greater than xxx Mhz" instead of just "xxx Mhz", and cooling your computer would give you immediate speed improvements.
Probably the first use for async is for extremely low power devices. If you just have a teeny little solar cell which delivers fluctuating power (or a heat-engine running off of daily temperature differences or body heat, or whatever), well, the chip will take as much power as it gets and just operate slower when the power drops. I am sure your sci-fi imagination can come up with possible applications.
Preferential Voting: easy as 1-2-3
For the record, the Linux Kernel Compilation Benchmark being 'the only test that matters' was a tongue in cheek remark. Next time I'll remember the smiley for the humor impaired; as for the rest of your 'argument':
You invoked a fascist reference; you automatically lose. Go away Troll.
--Parity
--Parity
'Card carrying' member of the EFF.
I wouldn't ever argue that hardware and/or software can't do certain things because there is a very fine line these days. When you look at the translation code and/or microcode in some CPUs, there _is no_ difference.
...
At any rate, modern Intel/AMD CPUs are actually able to mark branch probabilities as I understand it -- and its not hard when you consider they already have R/W registers galore to work with. Why not store it in another? Read up on modern processor specs
http://www.x86.org/ when you're done.
- Michael T. Babcock (Yes, I blog)
Actually, one of the things they'll be able to accomplish in the multi-CPU market is to hide the fact that there are multiple CPUs from the OS entirely. The software driving the CPUs can (probably) properly disperse the calculations among CPUs the same way it currently fills pipelines.
- Michael T. Babcock (Yes, I blog)