Slashdot Mirror
Fujitsu Coming Out With Crusoe Machines
Pulzar writes: "Fujitsu will release two notebooks containing Crusoe processors from Transmeta in November, the company said today, bringing the total number of companies coming out with Transmeta-based products to seven."
You mean a Pentium MMX isn't 5x faster than a Pentium at the same clock speed? And a 486 isn't faster than a Pentium? I'm outraged!
You mean "I need numbers"?
.config
I want to see thee very very very simple things:
1. network pipe over loopback under linux (memory to memry + some CPU).
2. standard mysql benchmark test
3. linux kernel compile for a reasonably populated
In btw: I do not want top notch performance as the laptop disks and IO will cripple it anyway. I just want to finally see what a hell are we talking about.
Baker's Law: Misery no longer loves company. Nowadays it insists on it
http://www.sigsegv.cx/
Wouldn't it make sense for Transmeta to rush in the delivery of the much hyped webpads that was supposed to showcase the primary benefits of the Crusoe processor?
It seems to be coming to light that the crusoe can't compete with Intel on performance and may not provide enough power savings in a traditional laptop to make a difference, so they need to find a niche market and quickly. I though that was their business plan from the start but it looks like we'll just see some regular notebooks coming out with fairly normal features and eventually Transmeta will be unable to justify their existence to manufacturers or consumers.
Transmeta's x86->microcode translation is no more dynamic than an Athlon's or Pentium's is. Let's look at their strategies:
Crusoe
Translate/reorder x86 instructions on the fly in software using "code morphing". The results will be in Cruesoe-native VLIW (Very Long Instruction Word) format, which is really just a block of 4 "normal" instructions that are guaranteed to have no dependence on each other/create any hazards if executed in parallel. Recent translations are kept in a cache.
Advantages: - Hardware is really cheap and doesn't consume much power since there is no instruction reordering/translation done in hardware.
- Performance will be good if you have a high hit rate in your translation cache.
Disadvantages
- Absolutely awful performance when you have nearly any misses in your translation cache.
AMD/Intel
x86 instructions are converted into RISC-like micro-ops in hardware on the fly. Most common x86 instructions have a one-to-one correlation to a micro-op, and instructions are reordered dynamically to try to exploit parallelism.
Advantages
- Fast: the worst case translation here is the same as the best case translation on a Crusoe (i.e., it's as if all instructions are a hit on the translation cache, as translation functionally incurs no overhead).
- Instruction reordering should be no worse than Crusoe's best effort.
Disadvantages
- The reordering/translation hardware is very, very complex and takes up a ton of power.
As for the argument that "normal benchmarks cannot be applied to the Crusoe," don't believe Transmeta's marketing BS. An industry standard benchmark like SPEC9x does use real applications (like gcc, etc.) for longer than a few milliseconds, and should be a good indication of Crusoe's real world, steady state performance.
Don't get me wrong, I certainly like the Crusoe. However, if you're looking for performance on x86 applications equivalent to even a mobile Celeron, you'll probably need to look elsewhere. To me, the real intriguing aspect of Crusoe is the prospect of running "native" applications. Running an os (read: linux) and user apps compiled for the Crusoe architecture would eliminate the need for the "code morphing" overhead, as the packing/reordering of instructions into VLIW format would happen at compile time. Such a system should be very fast (I would imagine equal or better performance per clock to a mobile pentium), and have killer battery life to boot.
Anyhow, I look forward to seeing the first Transmeta-based machines hit the market. Should be fun!
The speed complaints need to be taken with a grain of salt. The comments are mostly from people with 500 horsepower cars, complaining that 300 horsepower cars are slow and worthless.
Stop and consider some of the most impressive high-tech games of the 1990s: Quake, Flight Unlimited, Mario 64, System Shock. Now look back at what kind of top of the line machines were being used for the development of those games. Quake was wrapped up when 90 and 120 MHz Pentiums were the best you could get, for example. Now suppose you could have told the developers of these games about a chip with:
* A raw clock speed 3.5 times higher.
* A much faster bus (100 vs. 66 MHz).
* A much larger cache.
* A significantly better processor design featuring out of order execution and less need to pipeline by hand.
* 3D video cards at least 5x faster than what was being sold in $100,000 SGI machines in 1995. (Remember, in 1995 software rendering was the norm.)
That machine is a 333 MHz Pentium II with a Voodoo 2 card. Can you imagine the power? Wow, could you develop some mind boggling games on such a box. And most people are just surfing, downloading porn and MP3s, and using Word. Fast forward, and now we have people putting down 400-500MHz as "slow crap." Personally, I'd take a Crusoe that gave equivalent performance to such a machine, especially considering that it would be cheaper and use much less power. Blindly going for more megahertz is not the way to progress.
low size/weight are the two most important qualities for most people.
Still the Crusoe won't make that much of a difference until the
power usage of other components starts to come down. I recall that
the Crusoe has some support for reducing usage of other online
components by building a model of their operation ob board. If
successors of the Crusoe can actually simulate other circuitry and so
reduce the chipset count, that would be another big win both for power
consumption and performance.
done by released Intel processors. Rememebr that the Crusoe is a VLIW
processor which is capable of executing four microcode instructions at a
time. I guess the new VLIW Intels will do the same, but for now the
Crusoe technology is much more advanced than the Intel stuff.
I agree about performance, but the engineering really does live up
to the promised `very cool'. I think the performance delivered in
January was a disappointment: I think they had been hoping to beat
Intel chips of a comparable clock rate.
Can't wait to see Crusoes in set-top/embedded boxes. Decent performance, no heat, and no loud fans!
I have played with numerous Transmeta systems. I even have one by my desk now. It's real simple. Power consumption does matter. I read that the Intel P4 will consume up to 66W, compared to the TM5600 2-3W. This is INSANE. I don't want a 66W CPU anywhere near my hifi amp. Why do people bother arguing about audio codec quality when they have two or more fans with roughly the sound of a hovercraft in their room. Simple test for which chip is "cooler". When decoding a DVD in software, put your finger directly on the chip packaging. Not the heatsink, the packaging. HINT: Don't try this at home. You WILL get burnt on one of them. As for performance. I'm happy. Just remember, not everyone needs 1.4GHz P4s, lots of Celerons get sold, even today.
- Paul
sigh If you don't know how a CPU works, please just stay out of the Crusoe pool.
Your nice Intel (or AMD, or Cyrix or other) chip performs all its calculations in software - or just as much as the Crusoe does.
Which is to say, each opcode is broken down into microcode, which is then translated to circutry on the actual chip. Generally, once you get past the opcode level, you stop calling it software.
What the Crusoe has is the ability to manipulate the microcode programming realtime. In a Pentium, or AMD, or any other chip (including those RISC chips that are not really so RISC anymore), that microcode programming is fixed and can't change. In Crusoe, the potential is for the chip to adapt and allocate internal, on-chip resources to the current task.
In a simplified way, when you play MP3s, the chip takes on the characteristic of a dedicated MP3 decoder. When you run SETI at Home, it takes on the characteristics of a dedicated SETI chip. When you run Windows, it takes on the characteristics of Rodney Dangerfield.
The simple fact is that the Crusoe chip offers loads of potential, has a great idea that should be explored, and looks like it came out (in the first batch) slower than the competition, so they tried to pitch the (coincidental) lightweight power consumption.
Incidently, they were right about one thing. Normal benchmarks are not applicable to this category of processor. You can't measure it by running through a set of simple computations for a short amount of time (milliseconds). If you do, it won't adapt. In real life usage, however, you are more likely to be running KWord or Quake III for more than a few milliseconds, giving it time to reconfigure to an optimal setting.
With today's huge advance in hardware, I don't see much use for such "software oriented" chips.
Yeah. Down with software. Firmware's so much better. Gimmie the days of slapping in carts into the back of a TI computer. Who needs magnetic or optical media anyway. :)
--
Evan
"$30 for the One True Ring. $10 each additional ring!" -- JRR "Bob" Tolkien
Comment removed based on user account deletion
Perhaps I should have bolded or italicised the word "potential" I used throughout the message. I figured that simply leaning heavy on the qualifiers would show my opinion. It's hard not to outright say: "Crusoe looks like crap on a die".
Personally, I remain unconvinced that Transmeta will ever be a real sucess story. A little publicity, some OEM support, and they will (IMO) probably fade from view.
BUT - the concept is good. I just don't see it as having been translated to a practical product.
--
Evan
"$30 for the One True Ring. $10 each additional ring!" -- JRR "Bob" Tolkien
Comment removed based on user account deletion
Both users report that they are very happy!
Ahh - My eye!
The doctor said I'm not supposed to get Slashdot in it!
Yes, but Linus works there.
Still makes it vapourware...
--- Can i borrow your Clue-Stick(tm)? I need to go beat a few people with it...
It's good to see an intelligent x86 clone that's aimed towards power management than raw speed. Afterall, with Intel/AMD locked into that 'Clash of Titans' thing we're hardly likely one of them producing this sort of processor.
(Ok, StrongARM perhaps, but you know what I mean).
-- Hob - Java Spectrum Emulator
Indeed, but if you can get the chip to run cool enough, you won't need a fan. A traditional CRT monitor uses so much power that there's no point using a low power chip on such a system. But that's not their market. They're being aimed at the portable market -- notebooks and webpad type devices. Yes, LCD screens still suck large amounts of power, but advances are being made in this area (hopefully LEP screens will have low power requirements). Also, consider the CPU in a set top box (e.g., a satellite or cable decoder box). How many people would put up with them if they needed a noisy fan in them? With its low power requirements, a Crusoe is ideal here, a market that's inaccessable to Intel and AMD (with their current offerings).
"The invisible and the non-existent look very much alike." -- Delos B. McKown
For example a K6 will have a BogoMips value roughly twice that of the clock Speed in MHz, whearas a Pentium will have a BogoMips value of about 40% of the clock speed in MHz.
Only use BogoMips when comparing processors of the same type.
Oh really? let me see..
/var/log/messages | grep -i bogo
[root@formatter kde2]# cat
Sep 25 15:32:50 localhost kernel: Calibrating delay loop... 1464.73 BogoMIPS
And thats on Pentium III 733. Do you think that my Pentium III is 3 times faster then yours? I really don't think so!
Try to compare it with standard tools - then flame!
Hetz (Heunique)
Oh how wrong you are! Ever been on an 8 hour flight when your notebook goes dead after 2 hours? I thought not. I take these trips several times a month and I will be first in line to get a Crusoe notebook. Look to see Crusoe notebooks sell like hotcakes in the Fortune 500 domain.
On the PC Watch website:
PC Watch: Fujitsu LOOX S/T
And here:
Fujitsu LOOX
Hope this helps...
Comment removed based on user account deletion
But Dave Ditzel, Transmeta's CEO, has said that Crusoe-based notebooks will perform just as well as competing chips in real-world circumstances. A 700-MHz Crusoe 5600 will perform just as well as a 700-MHz Pentium III, he said.
Wasn't it stated somewhere that, due to code-morphing, the Transmeta chips would have about 80% the horsepower of a regular x86 chip of the same MHz?
I hope this wasn't something PR said. I'm a fan of Transmeta's stuff, and I'd hate to see 'em make promises they can't keep.
Of course school children need them. How else are they going to learn anything?! Have you seen the state of the TV and VCR sets in public school today? They're practically ancient!
What are the reasons to use Crusoe over Intel and AMD? Its not power useage
p 061900.htm
Intels chips use 1-4 watts in full energy save mode. The Crusoe Gets between 0.789 and 2 watts in its energy save mode. Amazing. http://www.intel.com/pressroom/archive/releases/m
AMD's chips use as little a 3 watts in energy save mode also.
And has anyone seen any REAL benchmarks? You know Sandra any thing to compare it besides Window Desktop Idle, MP3 Playback and DVD playback. For all we know it could run just as fast as a winchip or a Cryix (but be a whole lot cooler)?
In Transmeta OWN report, when they compare Office 2K scores, it took the P3 chip 0.133 hrs (8 min) to compleate the benchmark and the Crusoe 0.164 hrs (10 min). Thats a prety sizeable difference.
Sanchi
"They said we couldn't do it [Athlon]... but we built it, we shipped it... and we didn't have to recall it." Rich Heye
From the article:
But Dave Ditzel, Transmeta's CEO, has said that Crusoe-based notebooks will perform just as well as competing chips in real-world circumstances. A 700-MHz Crusoe 5600 will perform just as well as a 700-MHz Pentium III, he said.
Weren't they saying not so long ago that the 700Mhz TM5600 would give `comparable performance' to a PIII at 500Mhz[1]? Looks like their Code Morphing{tm} software is morphing the marketing figures as well as the x86 code...
[1] And the Transmeta Vaio Picturebook is going to be of `similar performance' to the 400Mhz PII in the current Picturebook, using a 500Mhz TM5600...