Slashdot Mirror
Transmeta Founder Talks Chips
gManZboy writes "Dave Ditzel, CTO and Founder of Transmeta (you remember Transmeta? weren't they supposed to kick some Intel booty?) sits down and speaks with Alpha and StrongARM chip designer Dan Dobberpuhl about the history of CPUs, where they're heading, and how the heck we'll keep up Moore's Law (if we can)."
About tortilla chips? Me, I like a nice big bowl of some good salsa and sour cream.
suck it bitches
A Conversation with Dan Dobberpuhl
From Power
Vol. 1, No. 7 - October 2003
Have we maxed out yet on microprocessor power? Two industry veterans discuss the trade-offs.
Introductions
The computer industry has always been about power. The development of the microprocessors that power computers has been a relentless search for more power, higher speed, and better performance, usually in smaller and smaller packages. But when is enough enough?
Two veteran microprocessor designers discuss chip power and the direction microprocessor design is going. Dan Dobberpuhl is responsible for the design of many high-performance microprocessors, including the PDP-11, uVax, Alpha, and StrongARM. He worked at Digital Equipment Corporation as one of five senior corporate consulting engineers, Digital's highest technical positions, directing the company's Palo Alto Design Center. After leaving Digital, Dobberpuhl founded SiByte Inc., later acquired by Broadcom. In an October 1998 article, EE Times named him one of "40 forces that will shape the semiconductor industry of tomorrow." He has written numerous technical papers and is coauthor of the text, Design and Analysis of VLSI Circuits, well known to a generation of electrical engineering students. Dobberpuhl is the named inventor on nine issued U.S. patents and has several more pending patent applications in various areas of circuit design.
David Ditzel directs our conversation with Dobberpuhl. Ditzel is vice chairman and chief technology officer of Transmeta Corporation, which he cofounded in 1995 to develop a new kind of computer--one that would learn how to improve its performance and save power as it ran, by using software embedded in the processor itself. Before founding Transmeta, Ditzel was director of SPARC Labs and chief technical officer at Sun Microsystems' microelectronics division. Ditzel came to Sun from AT&T Bell Laboratories in 1987, where he was the chief architect of the CRISP Microprocessor, AT&T's first RISC chip. His work first attracted industry-wide attention in 1980, when he coauthored "The Case for the Reduced Instruction Set Computer (RISC)."
DAVE DITZEL Dan, do you want to introduce yourself and say a couple of words about your professional background?
DAN DOBBERPUHL I recently founded a new fabless chip company called P.A. Semi. I have been in the industry for 36 years, and have been developing microprocessors since 1976. I have seen a lot of changes during that period in terms of both silicon technology and microprocessor development.
DITZEL You've got a wonderfully long history here. And our topic today is really about low power. Do you remember the very first computer you worked on, and can you take a guess at what kind of power that computer probably took?
DOBBERPUHL Well, it's interesting Dave. The power dissipation of the early MOS chips wasn't all that high, because the frequencies were so low and the chips were small and the transistor counts were very low. But they were typically in the range of three to five watts.
DITZEL In those days you probably used a minicomputer or mainframe to run your CAD tools, and there was a big difference between the performance and size of those development machines and the microprocessor chips you were developing.
DOBBERPUHL Sure. So when we developed the LSI-11, our processor design environment was basically a PDP-10, which was the time-sharing machine of the era. That was a large multi-rack system. And at the time the performance ratio was very high between those devices and the chips we were developing, in factors of 10 to 100. Over the next 10 years we brought that to equivalence and then basically the CMOS devices took over.
LSI-11
DITZEL So let's set the stage. The LSI-11 started about what year?
DOBBERPUHL The original LSI-11 started in the early 70s. It was designed jointly by engineers at Digital and Western Digital.
DITZEL And that had probably a MIPS rating. How many million instructi
--When it's my time, I want to die in my sleep like my grandfather -- not screaming like all the passengers in his car
Mr Linux dosen't work there anymore, so nobloody gives a goatse about transhitta.
(you remember Transmeta? weren't they supposed to kick some Intel booty?)
Uh, 1992 called. They want their slang back (and their processors, while you're at it.)
GNAA (GAY NIGGER ASSOCIATION OF AMERICA) is the first organization which
gathers GAY NIGGERS from all over America and abroad for one common goal - being GAY NIGGERS.
Are you GAY [klerck.org]?
Are you a NIGGER [tux.org]?
Are you a GAY NIGGER [gay-sex-access.com]?
If you answered "Yes" to any of the above questions, then GNAA (GAY NIGGER ASSOCIATION OF AMERICA) might be exactly what you've been looking for!
Join GNAA (GAY NIGGER ASSOCIATION OF AMERICA) today, and enjoy all the benefits of being a full-time GNAA member.
GNAA (GAY NIGGER ASSOCIATION OF AMERICA) is the fastest-growing GAY NIGGER community with THOUSANDS of members all over United States of America. You, too, can be a part of GNAA if you join today!
# Why not? It's quick and easy - only 3 simple steps! First, you have to obtain a copy of GAY NIGGERS FROM OUTER SPACE THE MOVIE [imdb.com] and watch it. (click here to download the 280MB MPEG from BitTorrent [idge.net])
# Second, you need to succeed in posting a GNAA "first post" on slashdot.org [slashdot.org], a popular "news for trolls" website
# Third, you need to join the official GNAA irc channel #GNAA on EFNet, and apply for membership.
Talk to one of the ops or any of the other members in the channel to sign up today!
If you are having trouble locating #GNAA, the official GAY NIGGER ASSOCIATION OF AMERICA irc channel, you might be on a wrong irc network. The correct network is EFNet, and you can connect to irc.secsup.org or irc.easynews.com as one of the EFNet servers.
If you do not have an IRC client handy, you are free to use the GNAA Java IRC client by clicking here [nero-online.org].
If you have mod points and would like to support GNAA, please moderate this post up.
Politicians are all the same, they promise to build a bridge even when there is no river.
In an October 1998 article, EE Times named him one of "40 forces that will shape the semiconductor industry of tomorrow."
Hmm. I wonder what day in October 1998 that was supposed to be? I don't remember any big change.
Sure, you can twiddle with your moderation settings to read all the blatant trolls that get modded down, but what about those works of trolling art that get modded up?
A new search must be added to slashdot, the TrollDot search.
I can't find it again, but I saw an interesting discussion that took the number of processors and embedded processors and the exponential growth of these devices and also the MIPS scaling and the energy per MIPS and compared it to the amount of energy in the sun. It was very clear that at some point you will run out of energy to power all the CPUs in a surprisingly short amount of time.
I wish I could find it again. (please let me know if you know)
I'd like to see the future of computing (and I do mean desktop computing) where the whole system has no moving parts. You read me, no spinning hard drive, only solid-state MRAM drives (or something.) No fans, not even in the power supply. 5W CPUs with the more processing muscle as today's 60W beasts. Oh, and OLED screens.
Well that's enough fantasizing for one day.
Although Frito's Scoops are a pretty good deal too... Who's Moore? ;P
Un-news
DOBBERPUHL The power is dissipated mostly in the transistors, either as they switch or as they just sit there and leak.
You can calculate the dynamic power dissipation with the formula P = CV2f, where V is the power supply, C is the capacitance that is being switched, and f is the switching rate. There are some additional factors, but fundamentally the dynamic power is given by that formula.
(NT)
anti-slash.org
I'd like to see ... 5W CPUs with the more processing muscle as today's 60W beasts
It would be a fine thing, but there's no sign of it happening. Instead, the next desktop CPUs are due to dissipate more like 103 watts. It's sad.
The author of that article is clearly just a baffon that bought a ipod when he wanted to listen to N*SINK or Fity-cent, and therefore his opinion is null and void. IPod's are ment to play good music at high quality, wether or not they do is not something i can say, but i trust in Mac, because there is no way it can be as bad as Windows
peace, -ls-
Flash media, and not MRAM, thank you very much. As for fans, well, just look at some Mini-ITX boxes. And ask for something that can take a 1GHz ULV Pentium M, which outputs ~7W, and is as powerful as a 2GHz Pentium 4, which outputs ~60W. About your OLED screen, why not the billboard-grade eInk that can pull 70FPS (for your Intel Extreme Graphics 2 that can only pull 50 on a good day)?
The hard drive ... can ... be [replaced by] a ... flash disk [holding] 8 [GB}.
(1) You need more like 80-200GB to replace hard disk these days.
(2) Flash is appallingly slow writing and does not seem to be getting much faster anytime soon.
The hard disk is a moving target, and flash is not catching up.
They both like a nice hard cock up their shitter.
nm
Not that you did read the article, but here's a great paper (pdf) on low-power processor design with lots of graphs and equations showing where the architecture can tradeoff power to keep your silicon chips from melting.
The paper is out of Stanford paid for by your tax dollars.. Hopefully you won't notice the part about the address at Stanford University being the William Gates Computer Science Bldg
Isn't transmeta's new, super-kewl uberchip running at a wonder 1.1 ghz? Or was it 1.4?
try the previous one.
have a good day
How Moore's Law affects some computer users as measured in the time it takes to do something, like render a page of a document on the graphical screen in a window opened for a word processor, is shown as an example here:
- 1992 1.25 seconds
- 1993 800 milliseconds
- 1994 500 milliseconds
- 1995 320 milliseconds
- 1996 200 milliseconds
- 1997 125 milliseconds
- 1998 80 milliseconds
- 1999 50 milliseconds
- 2000 32 milliseconds
- 2001 20 milliseconds
- 2002 12500 microseconds
- 2003 8000 microseconds
- 2004 5000 microseconds
- 2005 3200 microseconds
- 2006 2000 microseconds
- 2007 1250 microseconds
- 2008 800 microseconds
- 2009 500 microseconds
- 2010 320 microseconds
- 2011 200 microseconds
- 2012 125 microseconds
When you are doing something interactively and have to wait the better part of a second (or worse) for each step to complete, it can be a big pain. A faster CPU would be nice. But once that wait gets down into a certain range (varys depending on what the task actually is), it won't really matter as much, if at all.There will still be needed even faster CPUs for many things. The use of cryptography will certainly be increasing and that is a big need for more CPU speed. Larger, more bloated (in terms of steps of code, in addition to RAM and disk space), operating systems and applications will need faster (and larger) CPUs, too (though many have learned to avoid these steps to avoid the costs of upgrades to software and hardware).
But the market for faster CPUs will gradually be leaving behind more and more people who do the kinds of things that just don't need it. The threshhold has been reached for many, and soon will be for many more. Hopefully new and expanded uses will keep (or restore) the markets in a thriving condition.
now we need to go OSS in diesel cars
They spend several paragraphs discussing NMOS capicitors in CMOS processes circa 1994, but apparently neither knew enough to speculate about MIM or Trench capacitor structures, two mature technologies used in DRAM. Yes, they were leading in to the gate leakage issue, but the substance of that boiled down to, "Leakage sure is a big problem." Their solution is low-voltage chips with fewer transistors. Revolutionary!
There's way more substance in press releases from Intel.
Who do you get to be an expert to tell you something's not obvious? The least insightful person you can find? -J Roberts
But at least he did repost it to the correct article
They should be talking about why they didn't drop down on their knees and give pleasure to Linus Torvalds while he was still there, now they have to drive to his house to give him due credit.
This is so goddamn true.
first post?! yay!!!!!!!
Moore's Law is probably a limited phenomenon.
<pedantic>
Probably? Assuredly, I would say. If transistor count continues to double every 2 years, with 42M transistors per CPU in 2000, you would have 43 billion in 2010, 44 trillion in 2020, 47*10^21 in 2050, and 53*10^36 in 2100. If that hasn't reached the number of atoms in the known universe, then keep counting years and it will.
</pedantic>
9 megabytes per second is not good enough for you?
I wish I could believe that spec is realizable in a real system, but even if so, no, it's really not good enough for me. I can push at least 5 times that into my hard disk, and if anything I want and need more, not less.
I notice you stopped your calculations in 2012, much like the Mayan Calendar. This clearly points to the end of time and the coming singularity when computers become sentient and take over the world.
ACM Queue... "Tomorrow's computing today"
so tomorrow, I get to look forward to more underpowered web servers?
Being a Moore, I can't help but comment on Moore's law. In my lifetime, there have been a number of unforseen and incredible advances that have helped Moore's law significantly, besides the usual annual technological improvements. Moore's law will continue to advance mostly because of these unforseen advances, and I believe that the annual technological improvements that have become commonplace will also continue. Long live progress!
stuff |
One of the best Transmeta features was supposed to be the replaceable "translator layer" code, so it could run as ix86, motorola, alpha, or whatever CPU you wanted. (so you could boot Amiga, Mac and PC stuff on the same box, just picking upload of proper code on bootup. But AFAIK only x86 translator code was ever created. Anybody knows about progress with other platforms?
45 5F E1 04 22 CA 29 C4 93 3F 95 05 2B 79 2A B2
are you guys saying that a CPU only uses as much power as a regular lamp [bulb]
Absolutely. But grab a 60-100W light bulb that's been on a few minutes (PLEASE DON'T REALLY!) and tell me what it feels like. That is one heck of a lot of wasted heat energy.
BTW, the body heat of one human is also approximately the same as this figure, and look how much food (energy) we use up each day. It's just spread over a lot of surface area so the peak temperature isn't as high.
OK, this is off topic, but there is no appriate forum for this - has anyone else noticed the **occasional** LARGE ads at the top of the right hand column on the /. homepage?
I think you meant Amdahl's Law.. the improvement to the user is only as noticeable as the original experience was poor.
The faster the original redraw, the less of an effect the speedier redraws have on the user's interaction experience.
Transmeta a Go-Go followed by his encounter with Riker and Geordie... here. :-)
I've been looking for a 100% solid-state DVR.
Why? On-board camera for my race car.
If I can get it to turn on recording at the same time as I push the DATA RECORD switch on the datalogger, then I get video and sound synched to the data log - and that would be a HUGE advantage.
Why solid-state? Because race cars take a lot of abuse. 1.6G to -1.6G in the space of half a second or so.
I figure an MPEG2 capture card, an audio capture card, the OS on EPROM and Compact Flash as the filesystem. Video IN and stereo audio IN. Record at full-speed every time the RECORD pin goes to ground. Operate at 10V-16V.
I've found a number of VERY similar devices (for security cameras), but nothing yet that does full speed video and sound. Build one, price it cheap, and I'll buy it.
DG
Want to learn about race cars? Read my Book
Hmm.. I'm probably wrong over here, but I thought if you combine N>1 clusters you get 1 (bigger) cluster.
No idea how its computational would scale, but surely not exponentially?
No, don't ask someone that has a fab, because their answers will be strongly colored by whatever technology their fab is using.
Semiconductor fabs are so extraordinarily expensive that their rate of change is massively limited by the long time it takes to recoup investment. Unless you want a major loss on your hands, you have to churn out chips on each fab line for as long as possible without change, so you certainly would not advocate publicly any change in technology that would limit the lifespan of a current fab line.
I believe that the Opteron and Athlon 64 are both SOI designs (and Opteron has been in production for 6+ months).
Isn't the G5 also SOI? Isn't that a specialty of IBM Microelectronics?
Why would they imply that this (obviously) mature technology isn't?
And I was just listening to the Infectious Grooves today! :-)
Moore's law applies directly to the number of transistors on a chip, but since we are all using Maas Biochips these days, none of that applies anymore. Unless you own one of those old Ono-Sendai peices of crap...
TallGreen CMS hosting
But programmers don't use them.
With my command of assembly language, there probably aren't many coders out there that could write faster code than I. I'm not bragging; it's a simple fact that if you can fit the entire executable into the processor's cache, all of the optimizing compilers in the world don't matter. Instructions executed out of cache are completed around 7 to 10 times faster than those fetched from memory.
But nobody cares. Programmers have become so used to the increase in processing power that they are willing to write ever more bloated code, knowing that somewhere, somehow, there's an EE major who will invent a fab process that will make up for the shortcomings of programmers.
For every doubling of clock speed or halving of power consumption, there's some intelligent idiot (tm) designing a better, slower Java. It doesn't matter how fast the processor clock is if you're programmers are ever willing to invent less-efficient languages.
Instead, if you taught programmers to write efficient code, the speed of the processor would be a moot point. Today's processors are overkill for the average consumer - unless, of course, said consumer writes an interpreted language in Java and runs the interpreter in a JVM... Yeah, you laugh, but trust me, someone is already doing it....
The society for a thought-free internet welcomes you.
The original SOI was silicon on sapphire and was developed in the 60s for rad-hardened military chips and used by HP for workstation processors in the 70s and 80s. And yes, those AMD and IBM processors use SOI.
Two wrongs don't make a right, but three lefts do.
No mention at all of the early developments done by Acorn (Sophie Wilson and others) and after that by ARM Ltd on the first ARM design followed by succesive new generations of ARMs. Something that is still going on. The Digital and Intel developers could build on the already available power/mips performance of the ARM architecture.
So far that ratio has been better in true ARM designs than in the American offspring.