Slashdot Mirror
A Peep From Transmeta And Toshiba (And RLX)
irix writes: "C|Net is reporting that Toshiba will ship a mini-notebook May 18th in Japan, coming to the U.S. later this year. The article also has some information about upcoming Transmeta CPUs." Hints and promises from Transmeta are that the next generation Crusoe will be smaller (half the size of current ones), faster (up to 800MHz) and consume less power (not quantified). U.S. notebook makers still seem reluctant to use them though -- so if Americans want a Crusoe in anything but a Sony Picturebook before the end of the year, we may have to watch dynamism.com and similar places. Update: 05/07 09:37 PM by T : OS24Ever also writes: "Linuxgram has an article about scooping RLX Technologies announcment of their new System 324 Web Server. At its optimum, the product will hold 336 Web servers running Linux or Windows (Windows costs $200 more). The Transmeta chip runs 80% cooler with 80% less power requirements, eliminating a lot of heat and need for fans, bringing single point of failure in the machine down to near zero."
no fans, no fans to break, one less point of failure.
That doen't necessarily mean those reasons have to do with the viability of the technology. It could be that, oh, most laptop companies have desktop divisions and wouldn't want to lose Intel as a supplier (or get a less-sweet deal) as a result of using Crusoe in some of their portable products.
IBM, for example, couldn't get its stor straight when announcing it was dropping Crusoe - the answer varied according to who you asked.
In light of the Toshiba and RLX/IBM announcements, I would say Transmeta's future is very bright indeed. I would take this opportunity to buy if i hadn't already. It's the last opportunity before the 5800 comes out and things really start smoking (to use a bad metaphor).
Scott Draves
The interesting thing is that this is exactly what Intel does, too. The only difference is that Crusoe's translations are softcoded, while Intel's are more-or-less hardcoded. Intel chips are actually RISC at the core, with a translation layer between IA32 and whatever they've got underneath. However, I'm willing to bet Intel's RISC machine is optimized directly for IA32 while Crusoe's is probably more general.
Also, it's interesting to note that having the RISC core give the Intel chip a boost, not a setback.
Engineering and the Ultimate
It was designed for lower power consumption and heat output. That happens to be very desirable for low-power mobile use, but *gasp* that also doesn't sound like such a bad idea for a web server farm.
If you're running a database server, CPU performance is more of an issue, but a heavily loaded web server is going to top out I/O and network throughput before it does the CPU.
Just because you can put an insanely high power server out there on the web doesn't mean you need to. A friend of mine runs a web hosting service that serves 140 domains and a number of sites hosted off of the main domain. Most of these are art and multimedia sites (what the service caters to) -- not very lightweight. The sole server is a 188 MHz Cyrix 6x86 with 128MB RAM. The loadavg is currently ... 0.08.
Crusoe isn't compute farm material, but peformance doesn't actually SUCK either.
DNA just wants to be free...
You are right that the "Cope Morphing" does consume some CPU cycles when it's first performed, though it is cached so the performance hit isn't excessive.
One major benefit however is that you drastically reduce the amount of silicon you need to build your chip. In some of the modern CPUs like the PIII, instruction decoding (the very thing Transmeta is doing in their Code Morphing) can take up half your silicon. That's a lot of savings in power, heat dissipation, and cost.
Transmeta took this in one logical direction and created a mobile CPU. This makes sense, and is a good way to get into a market where the competition from Intel/AMD wasn't as formidable. But here's another application: instead of just throwing out that 50% of silicon real estate, put in more integer and floating point units, maybe increase the cache. That 50% will buy you a lot of extra processing power.
Suddenly, they'll have a powerhouse chip that may very well outperform Intel and AMD's offerings at a comparable power/heat/cost. The hardware changes would be relatively simple, and would require modest changes to the Code Morphing software to make use of the extra hardware units. Transmeta could do this without much difficulty.
Will they? Not until they get themselves more firmly established in the mobile market. But once they have the financial muscle to take on Intel/AMD on their home turf, I think they will. The profit margins on desktop/server chips too high to pass up.
I'm as cynical as the next Slashdot'er. However, before you get yourself up in a craze of "less power, why would I want it?" You need to look at this box. It's COOL.
The possibilities of 24 servers in 3U using the power of a typical 1-2U server are incredible.
Get past the idea of using one server per function. You can setup a balanced solution between multiple servers (say 2,4,8,...24!!!) for such redundancy.
Get past the idea of need a super fast processor. For most web serving functions you're much better served by putting lots of smaller servers out there than you are by putting one big server.
Imagine wiring up 48 ports of ethernet with FOUR cables (plus another 24 ethernets coming out of a single RJ45.)
Stop thinking...they're new...I'm too cool for this. It will change they way you approach web architectures...hell other systems infrastructures too.
In the interests of full disclosure, I am an RLX employee. Mid to late 2000, a friend called me to come look at this box. I quit my job and moved to Texas from DC. Don't particularly like much about the area, having lived in an urban location like Georgetown, and moving to a suburban area like the Woodlands, but that didn't matter. I came running.
So, like I said...look at the product. And think.
No I'm not a marketing guy...
I'm in IT--and you think that you're cynical.
220 node beowulf setup in under 2 hours.
I talked to the guy after he was done using the hardware. My mouth dropped.
TBH, I am a contractor at RLX, so my opinions may be biased. I also work on the Debian-based default linux stack.
There's just one more thing I will say about this product. -- Bring a clean pair of underwear, because you are going to soil the ones you currently are wearing when you see this thing in action.
It's a Transmeta laptop with a CD-ROM and a fullsize screen, in the kind of nice aluminum casing that Panasonic has been using on their M1 laptop (and that Apple has more recently been copying with their titanium laptops). It has a 7 hour battery life, 9 if you replace the drive bay with an additional battery.
Check it out!
Jon
All opinions expressed herein are my own, and not those of my employers, who are appalled.
No need to imagine, my friends. Heard a piece on NPR about RLX technologies. Beowulf clusters are the next phase. They want to sell them to accounting firms and stock market analysts.
I'll dig around NPR and see if I can't find the story.
by Mike Buddha -- Someday the mountain might get him, but the law never will.
We've got a beta of the RLX Razor at my work. For a managed services company, this thing is sweet!
For that first slot "control tower" one can do all sorts of nifty things with the other blades from power cycling to bios settings... which inclines me to want to have that first blade not be generally accessable to the outside world. This reduces your available work horses to 23 per 3U density. That's still a rackspace bargain, though!
The hard drives may be a point of contention for some people, but the way I envision these is to boot them via net and have your web cluster get data from network attached storage rather than local. This way you may boot these diskless and reduce the power consumption by about half! A fully populated chassis will then suck less than 200 watts.
All in all, a very tasty product. Can't wait to throw these in our colos =)
A quick blurb on Yahoo news mentions that IBM is going to resell RLX boxes. It also calls RLX a Compaq offshoot, although I'm not sure this is a valid use of the term 'offshoot'.
Not really, with typical smp boxes, if a cpu goes bad then your entire machine crashes. There's a difference between smp machines and fault tolerant machines. Maybe if those cpus where run in a cluster with the same task and data running on multiple cpus and with invisible failover but I don't think the server in question has that capability.
"When you sit with a nice girl for two hours, it seems like two minutes. When you sit on a hot stove for two minutes, it
I think the whole point is that the processor runs fanless. So while there is one processor, and it is one point of failure, under normal circumstances you need both the CPU and the fan to be working or the server dies.
Perhaps more to the point the CPU doesn't have any moving parts. If any one thing is going to be a single point of failure, I'd rather it didn't have any moving parts.
Dave
I write a blog now, you should be afraid.
This became a FAQ when the crusoe came out. The official line is there is no such thing as a native instruction set, basically because they want to retain the flexibility to change it between one model and another. Or possibily even just between revisions of processor cores.
It's all quite clever really.
Now, when are they going to write the emulator for PPC? Or Sparc?
Dave
I write a blog now, you should be afraid.
80% less heat and 80% less power (do the two really translate so directly like that?),
Actually yea they kinda do, as a perfect processor that would release absolutly no heat, it would use absolutly no energy though such is basically impossible, it is a thoretic possibility. Of course the monitor and other moving part devices would use energy.
I think it's a great idea to get those Crusoes into the mass market web server market. Transmeta has been so focused on the highly competitive sub-notebook market that they missed Cobalt selling RAQ servers hand over fist at amazing mark ups. Any concerns of power can be removed by noting that this things pack an AMD K6-2 450 and still sell for $1500 - $3600.
If Sun can sell a K6-2 450 at $3600, Transmeta has got to be able to find some room in there for profit.
-jhp
/. -- the Free Republic of technology.
If you read the article *cough*, you would see that it's can be up to 336 servers (processors) in a 42U rack. Very redundant. Surely you'd agree that such a setup would eliminate problems with any single processor failing.
"If I am such a genius, how come that I am drunk and lost in the desert with a bullet in my ass?" --Otto (Malcom ITM)
Actually, for repetetive server tasks, dynamic meta level translation can be just the right thing.
I suspect that an implementation is at least several years off, but it should theoretically be possible to generate specialised execution paths with system calls turned into method calls (because the dynmaic translation software could proove no information would escape the specified code path). Inlining the system method calls, you could then potentially get rid of the copying between user and kernel space... potentially, your application could inline the parts of the kernel, all the way down through the networking stack.
As to whether this is likely to happen... who knows. IIRC cdrom.com maxes out an OC-3 on a dual pentium pro, so perhaps it would be overkill. But theoretically possible nonetheless
And some of it could be attributed to investors diversifying their portfolios as they haven't been allowed to do for 6 months. It is a good investment idea.
---
The stock lost 23% today as insiders were released from their 1 year lockup-IPO deal and sold. This story suggests that Transmeta is not viewed with great enthusiasm by the people on the inside. Of course some of this selling could be attributed to pressures from other losses to raise capital, but you cant read it as good news regardless.
Someone you trust is one of us.
.
"The flexibility of the software-translation approach comes at a price: the processor has to dedicate some of its cycles to running the Code Morphing software, cycles that a conventional x86 processor could use to execute application code."--Transmeta Crusoe Whitepaper
VLIW (Very Long Instruction Word) technology employeed by the Crusoe is essentially a software emulation layer for the majority of the CPU instruction set. This means that the "Code Morphing" software that translates instruction sets into VLIW words sucks CPU time. In other words, just because is a 633Mhz CPU, doesnt mean it will perform like a PIII 633Mhz CPU. This sounds like a step in the wrong direction.
Someone you trust is one of us.
he Transmeta chip runs 80% cooler with 80% less power requirements, eliminating a lot of heat and need for fans, bringing single point of failure in the machine down to near zero."
Almost wholly falacious (fellatious?) reasoning. You may think the processor is less likely to fail if it runs cooler, but if there's only one of them, it is still a single point of failure. A single point of failure either is or isn't. It is not "nearly" anything. (If you entirely do away with the need for a fan, it helps some.)
--
"that's not encryption - it's a new perl script that I'm working on..." - from some Matrix parody
"The Transmeta chip runs 80% cooler with 80% less power requirements, eliminating a lot of heat and need for fans, bringing single point of failure in the machine down to near zero."
What kind of pointy-haired boss nonsense is that? By what yardstick of single points of failure has it approached zero, and how close is "near?" What about unplugging it? Bumping the reset switch? Drive failure? Flood? Fire? Or running Windows with IIS?
Okay, it's hard to classify Windows as a mere single point of failure.
...
The way RLX has managed to tuck 24 servers into a 3U enclosure is to stick them in vertically.
Not that much of a conundrum. A 3U enclosure containing 24 servers.
J.
The article states that the Toshiba will use a polysilicon display, which, according to this article, is one of the new low-power display technologies that is competing against the new organic displays.
As long as this new laptop does not include CD, floppy, or DVD drives, it should be very power efficient. I wonder what the power bottle kneck for such a laptop is. Does the 10GB harddrive zap too much juice? Or is it the graphics chipset? I bet the speakers are the most energy hungry parts on laptops such as this new Toshiba and the newer Sony Powerbooks.
For a web site that needs high bandwith but has uneven bandwith requirements (ie - gets a big rush in the middle of the day), you could configure a load balancer to point to Transmetta powered servers when the load was light, and allow the bigger UtraSparc or Xeon servers to go into standby mode. When the hits start to exceed te capacity of the Transmetta servers, the load balacer could wake up the bigger iron to handle the heavier load. No loss of capacity or increase in latency (well, not much, anyway), and dramatically reduced power consumption.
I can't imagine I'm the first person to think of this...
--
In spite of the suggestions and all the tests that I have made, I have not cavato a spider from the hole.
The interesting thing about Crusoe processors is that the raw hardware is abstracted from the instruction set.
Why is this important? Instruction sets are really an abstraction themselves. They are what allow us to not have to program in binary. Problem is that until Crusoe, they were always hardware. Wanted some features the instruction set didn't provide? Build a new chip at a cost of billions of dollars. Want to port your software to a different architecture? Better be careful whate instruction set features you depend on. Or instead you can buy a Crusoe processor and it's simply a software update.
But it gets better. With a Crusoe chip, any CPU can be emulated. Want your machine to be a MIPS today and an UltraSparc tomarrow and a Pentium the day after? It's possible, if perhaps not incredibly practical. (though it would be great for testing software on multiple systems without having to buy multiple systems)
Additionally the underlying hardware can be optimized for the real tasks being undertaken without requiring a redesign of the software application or operating system. Transmeta could create a server chip that is highly optimized for I/O or even possibly have several chips work together while the operating system is none the wiser.
Just because the chip doesn't perform the same at the same Mhz is basically irrelevant to the value of the Crusoe chip. Unfortunately Transmeta is doing a really bad job of promoting this. I can't understand why the heat/power thing is the centerpiece of their marketing.
Duh. It's already here. The PCG-C1VN, Sony's Picturebook, was available in the US only a little after it was in Japan. This improves their previous record -- the PCG-C1 took almost a year to get to the US.
What activities would help "the masses" get their hands on the technology?
I have a different question for you: Why do you care? Whether a laptop uses a TM chip or Intel or AMD is totally invisible. The only question is performance versus battery life versus cost.
Given that no one seems to be interested in using TM's chip so far, there must be valid reasons. One is that the performance seems to be much worse that the competitors. And two is that the processor is only a minor source of battery drain.
So why are you so interested in demanding that manufacturers supply with apparently inferior technology?
This is not to say, by the way, that TM's chip may not find uses in applications that are more interested in heat dissipation than performance.
--
Sometimes it's best to just let stupid people be stupid.
I have mixed feeling on this one.
But I do not know how viable it will be as a laptop replacement.
Check out the Vinny the Vampire comic strip
"It is a greater offense to steal men's labor, than their clothes"
To me, that means that the device generates 20% the heat that it used to, which means instead of running at 322 deg Kelvin, it would run at 64degK, or about -373 degrees Fahrenheit, which is damn cool. It would keep the whole notebook at a cold-ass temperature, definitely removing any need for a fan!
Knowing slashdot's userbase and how anti-large-corporation they are, and since transmetta seems to stand for anti-intel almost as much as amd does, I expect this to be viewed as flamebiat, so flame away! But here are my thoughts. Doing Embedded Engineering professionally I can attest to the various strengths and weaknesses of most embedded platforms out there. Is crusoe low power? Yes! Is it low heat? Yes! Is it the first processor out there to do "code morphing"? No. Is it the the lowest power consumption and lowest heat processor out there? No again. CodeMorphing (as transmetta calls it) is nothing amazingly great. Ever since Intel and AMD both had to preserve backwards compatibility with 16bit processors, and have had to register rename, they've been taking in one set of instrunctions and actually executing a different set. They've accomplished this through *gasp* software built into the hardware! *gasp* They call it microcode. If you want to get down to the nitty gritty, the Pentium4's trace cache is pretty much the same thing that Transmetta's CodeMorphing system does. It decodes an instrunction into other instrunctions and stores (or caches) the decoded instrunction fragments. It's not rocket science. It's not revolutionary. But it sure is cool. That aside, let me get back to the power issue. What most of the readers on slashdot don't seem to know is that both AMD and Intel make mobile versions of their processors. No wait.. everyone knows that but most people don't actually know how different the mobile version is from it's desktop brother. If you were to compare numbers between Crusoe, Mobile Intel and Mobile AMD processors, you'd find that their power consumption is all actually pretty darned close. The mobile version of the desktop processors actually are different right down to some of the core. Many of the mobile processors can turn off part of their cpu's. They can all regulate their frequency. They're flat out not the same processor other than that they have the same name and the same features from a software standpoint. All that said, I personally think XScale/StrongARM are the best performance/power processors out there. Too bad you can't run native x86 code on them.
Anybody want to tell me what this means? I sure as hell can't make head nor tail of it.
When it arrives on the continent, the notebook will join Sony's Crusoe-powered Vaio, NEC's Versa Ultralight and a new entrant from Casio.
When it arrives on the continent no one will notice! My laptop screams look at me! I come in five different colors of florescent!
Murphy's Law of Copiers
I love the smell of Karma in the morning
I can't wait to see what happens when someone teams up mini-fuel-cell type batteries (like the ones Motorola has in development) with a low-power proc like the Crusoe or StrongARM to build a tablet PC that lasts for a month on a single charge
Where's my lobbyist? Right here.
Sony is pretty good about bringing out products in the US eventually, and when they do, the best part will be the compatability with Linux. Look at the Vaio, for example. Almost totally supported, which is pretty rare with notebooks. I'm happy with mine. I'm sure that when the Crusoe shows up on American shores, it'll be in a great product too... gotta love the toys
I'm an Angry Clam. You would be angry too if you were a ball of snot in a shell.
Not quite. Any non-reversible operation increases entropy and thus produces some heat. Changing a 1 to a 0 (by draining a charge to ground, say) will produce heat. You could sure reduce it below where we are now, and there are some interesting designs around for almost-reversible-process-computers, but even they have to do something irreversible (e.g. tell you the answer) to really qualify as "computers".
Alright! Somebody had to do it.... Can you imagine a Beow.....