Slashdot Mirror

Analog neural nets work best in special purpose application -- where what you want to do is to take a stream of input vectors, feed it into a MLP, and generate output vectors, ideally at the maximum rate that the analog circuits work. Or, if you're into power instead of speed, push down the power used by the net until you're just barely making the timing. This works the best when there's integrated sensors (like photoreceptors or a silicon cochlea) or actuators (micromachined stuff) on the chip, so you dispense with an A/D or a D/A by going analog as well.

There are a few products out there that use analog nets just in this way -- there's one in your Logitech Marble Trackball, computing the motion vectors of a pseudo-random dot pattern on the ball in IR.

But most problems out there actually don't fit in that nice little niche -- instead, the neural net is part of a signal processing chain, and the adjacent steps don't fit well in analog, so you need a DSP anyways. And once you have the DSP, its usually is a lose to put an analog neural net on the die, instead of just adding another ALU to the DSP, since the rest of the system is digital anyways.

One hope is this networking protocol, the address-event representation, that lets analog neural net chips communicate with each other in a very digital way, while being very efficient to implement in analog. See this paper for details ...

• Amanda
• Baikal NT-200

Here's some more cool MEMS research at Berkeley: Solid Fuel Microrockets.

They are suggested as a method for delivering Smart Dust. I think they just want to blow stuff up.

Jason

Albert Pisano, the outgoing DARPA program manager for MEMS, likes to talk about building a MEMS dandelion seed, a few mm in diameter. (I saw him deliver the talk at GWU, and he's also given it at NIST.) With current process technology, an old processor like the 8088 would easily fit in that space; in fact, you could get a few hundred 8088s in that space, so computational power isn't a problem. Add power generation, sensors, and radio communications, and you're on to something!

I work on software for a MEMS-related project (the MEMS Exchange). It's an interesting field, and one that's already having an impact in specialized areas like accelerometers, and is very close to becoming widespread.

There's a link right at the top to a page at berkley on the stuff. Check it out here

Wait a minute.... first, the G4 500 numbers are estimated, second, I have actual results for http://infopad.eecs.berkeley.ed u/CIC/summary/local/ where the K7 650's SPECfp score is 0.2 points higher than the one given by Archintosh, third... Maserati failed to mention the SPECint95 results:


CPU Int FP

K7 550 23.6/20.6
K7 600 27.2/21.6
K7 650 29.4/22.4

P3 550 22.2/15.0
P3 600 24.0/15.9

G4 450 21.4/20.4
G4 500 23.8/22.6***


please note: I estimated the G4 500's SPECint score by dividing the G4 450's score by 450 and then multiplying by 500, rounding up

the SPECfp95 estimate is from the Archintosh website...




_______________________________________________
There is no statute of limitation on stupidity.

Donald Turcotte

Dr. Steve Bramwell

John Harte

Quick! Somebody fix the SETI client! We don't want to miss any of the alien signals!!

SETI GBC Analysis

Just kiding. :)

The head of the letter says 22/7/99 and the date of the file in the dirctory is 16/8/99. Pretty outdated news for slashdot, huh? Anyway. This is a good thing to happen - at any time.
B.

A SPEC_fp number would be interesting, not some integer benchmarks from a computer magazine. I'd use a FFT for a DSP benchmark not for a state of the art microprocessor.

I'd like to second that recommendation. Marc based his book on a lot of the ideas that have swirled around this group (hypertext, idea futures, real computer security, smart contracting, etc.).

By the way, bidirectional linking is not new to the Web. It was new in 1997, when it was introduced by CritLink. I encourage you to check that out, too. It lets anybody annotate any public web page using any browser -- no software required.

-- ?!ng

And for man pages, you definitely need tkman.

--

Previous generation Alphas (21164A) are available for a song.

I recently purchased a refurbished Digital Personal Workstation 500a (500Mhz 21164A, 2MB cache) for $1100 from a local Digital reseller; had Debian GNU/Linux 2.1 running on it within an hour.

And how does it compare to a G3 or 604 PPC? Well, the Alpha is 64-bit, which enforces good coding habits ;-) and the integer performance is comparable. Though GNU libm on Alpha is still not so very optimized, the floating point performance of the 500Mhz 21164A soundly flattens my 350Mhz G3 PowerMac :)

Here are some specs I retrieved from this site:

CPU /clock / SPEC95int / SPEC95fp

Alpha 21164A / 500Mhz / 15.0 / 20.4

PPC 750 (G3) / 350Mhz / 15.4 / 11.2

BTW, PC164SX main logic boards with 533Mhz Alpha 21164PC have appeared on Ebay for $250! Standard ATX form factor and accepts SDRAM DIMMs.

Just to let everyone know... ;-)

~AC

It would be cool to get the various distributed computing initiatives (d.net's RC5, SETI, OGR, etc.) together onto one CD image to burn and run on Dreamcasts. Just pop the disc in when you're done playing and let it crack/spook/compute... It might be a good first step before tackling getting to the onboard 3D.

Well, this is just one application, but look at the statistics for Seti by cpu type. There is an Alpha churning out seti work units in 57 minutes!. It makes my 8 3/4 hr average look downright slow.

Get a clue. For example, read Berkeley security class notes

Ok, so I did something wrong - here . The error of my ways was that slashdot no longer likes single quotes around the URL - need to use double quotes..

Erik

Has it ever occurred to you that God might be a committee?

Well, the HPPA 8500 has 140 million transistors :-) (1.5 MiByte L1 (yes level-one) cache).

Here are some approx numbers:

Intel: PII was 7.5 M transistors in a .25 micron process and an area of 131 mm^2. The PIII is 9.5 M transistors but I've got no area for it, I believe it's made in the same .25 micron process (not 100% certain) but you can't just scale the area (different types of transistors take a different amount of area, and if the placement is 'suboptimal'..)

AMD: K6-III 21.3 M transistors and 135 mm^2 in .25 micron process (not same as intels process) and Athlon shouldn't be too different (in that process).

IBM/Moto (PPC): PPC750 is 6.35 M transistors 40 mm^2 in IBMs .22 micron process and dissipates 5.5 W at 466MHz. IBMs POWER3 is 160 mm^2 in a .2 micron (should this be the same .22 micron as above?) process.

SUN (SPARC): USII is 5.4 M transistors and 126 mm^2 in some .25 micron process and dissipates 21 W at 400 MHz.

COMPAQ (ALPHA): 21264 is 15.2 M transistors in a .35 (!) micron process and takes 302 mm^2 and dissipates 72 W at 666 ( ;-) ) MHz

SGI (MIPS): The R10k was 6.8 M transistors (no numbers for R12k) and took 298 mm^2 and dissipates ca 30 W at 195 MHz in a .35 micron process (this CPU shipped '96, don't have numbers for any later).

HP (PA-RISC): 140 M transistors and an unknown area and an unknown heat-dissipation. The process is intels .25 micron.

Larger (in area) processors are usually more expensive to manufacture because the errors are per-area more than per-transistor. And that means that if you make larger chips your 'yield' (percentage of the made chips that works) becomes lower at the same time as you get fewer chips per wafer... Now compare the 21264 and PPC750... The PPC750 designers weren't incompetent - they had other goals than the Alpha designers..

Intels processes has traditionally been very well tuned with very high yields.

The raw data is availible here

Erik

Has it ever occurred to you that God might be a committee?

Check out UCB logo at ftp://ftp.gnu.org/pub/gnu/ucblogo/ Trust me, there's a lot more educational value in it than you think (ask all the current 20 year olds what they used in grade school) and, in my opinion, has more educational value than dummy down software like Encarta. But if you don't believe me, ask someone who has done extensive research in education (also the writer of this software). Mail Brian and tell him that you're interested in using logo for educational purposes and he'll gladly give you some advice.

I'm going to side track your main point a little as I think this still relates. I have always been quite worried whenever I see people talk about educating our kids using computers. Not that there isn't any pedagogical value to it but that whenever it is used in schools with children it is almost always misused. I can't get into too much detail here but you should read this article which I find to be a very thoughtful paper about the uses of computers in education. It was written by one of the creators of UCB logo who has done some quite extensive work in this area. The point I'm trying to make is that there has been so much hype recently with computer XXX in general that we're often quick to jump the gun and assume that anything the slightest bit related the computers is always a good thing. Keep in mind that computers are not teachers and we should never try to substitute human beings for them. They're just tools and I strongly encourage people to spend money on books and teachers before blowing it all away on "high tech" gadgets and network equipment. There is some educational value in computers but it has nothing to do with the way most people think they do. Again, read the article and you'll see what I mean. There are also many other papers you can read at http://www.cs.berkeley.edu/~bh/papers.ht ml

I'm going to side track your main point a little as I think this still relates. I have always been quite worried whenever I see people talk about educating our kids using computers. Not that there isn't any pedagogical value to it but that whenever it is used in schools with children it is almost always misused. I can't get into too much detail here but you should read this article which I find to be a very thoughtful paper about the uses of computers in education. It was written by one of the creators of UCB logo who has done some quite extensive work in this area. The point I'm trying to make is that there has been so much hype recently with computer XXX in general that we're often quick to jump the gun and assume that anything the slightest bit related the computers is always a good thing. Keep in mind that computers are not teachers and we should never try to substitute human beings for them. They're just tools and I strongly encourage people to spend money on books and teachers before blowing it all away on "high tech" gadgets and network equipment. There is some educational value in computers but it has nothing to do with the way most people think they do. Again, read the article and you'll see what I mean. There are also many other papers you can read at http://www.cs.berkeley.edu/~bh/papers.ht ml

Actually, the consensus among complexity theorists is that quantum turing machines are more powerful, even theoretically, than regular probabilistic turing machines. This has not been proven, for similar reasons that ~(P=NP) has not been proven either (namely, that it is freakin HARD problem to do so).

For more on this, look at this paper by Umesh Vazirani (one of the big names in computational complexity): Quantum Complexity Theory.

Sigh. Too bad more aren't using a software engineering OO language since there is now such a rich choice available under the GNU license: Eiffel, Ada95, Sather (among the safest). Even Modula3, Smalltalk, and Java are worthy of consideration. Here's some URLs,
GNU Eiffel
GNU Ada95
GNU Sather

In their technical news sections the following is listed: July 17, 1999 We are awaiting the arrival of a RAID controller card and database server software for our Sun 450 server. When these items arrive we'll do a major (and hopefully final) upgrade on our server architecture.

That is the link to the alpha binary. Most people probably want the x86 binary which is h ttp://setiathome.ssl.berkeley.edu/software/setiath ome-1.3.i386-winnt-cmdline.exe.

If you're looking for an alternative to CVS's complexity, check out PRCS. I've found it to be an amazingly clean system to set up and use, with a simpler model of what's going on. No client/server yet; version 2.0 will support full disconnected distributed operation.
--
Employ me! Unix,Linux,crypto/security,Perl,C/C++,distance work. Edinburgh UK.

Personally, I think it's wonderful. My most extensive experience with it has been in the glx project, and it's worked very well. Not being a professional programmer, it's the most sophisticated system I've every used. From talking to some of my friends, I understand it has some advantages over a lot of the commercial offerings, so we may not be missing much. :) I also use it to keep track of the data and analysis routines for my scientific work.

That said, there are lots of problems. It's not been terribly stable in my experience. It has poor support for binary files. Administration isn't fun (or easy) and it's difficult to set up securely. It's not very smart: particularly glaring ommisions are that you can't remove directories once they're added to the source tree (!), and moving files around is a pain. The notion of "code branches" could be more powerful and easier to use. (This is probably the part linus objected to--I couldn't imagine trying to track all the kernel patches that way. maybe with a gui front-end and database to keep track of the options... :) It's not easy to perform clean backups or mirrors either, and the command line options are neither elegant nor consistent. Oh yes, and it could be faster. That's my personal list so far.

Basically, it's a hack on top of rcs, and it's starting to show. It could probably benifit from a complete rewrite in the next year or so, with the addition of things like a security model and support for distributed (and mirrored) repositories. bitkeeper is something like this, but not free software. prcs is another, gpl'd, attempt headed by Josh MacDonald, author of xdelta; there's no client/server for it yet, though. Personally, I'd like to see a cross between cvs, an eternity server, and Debian's apt package tool. :)

Nevertheless, I think it works fine for medium sized projects and really helps facilitate/speed up internet-based development. Beats mailing patches around!

Personally, I think it's wonderful. My most extensive experience with it has been in the glx project, and it's worked very well. Not being a professional programmer, it's the most sophisticated system I've every used. From talking to some of my friends, I understand it has some advantages over a lot of the commercial offerings, so we may not be missing much. :) I also use it to keep track of the data and analysis routines for my scientific work.

That said, there are lots of problems. It's not been terribly stable in my experience. It has poor support for binary files. Administration isn't fun (or easy) and it's difficult to set up securely. It's not very smart: particularly glaring ommisions are that you can't remove directories once they're added to the source tree (!), and moving files around is a pain. The notion of "code branches" could be more powerful and easier to use. (This is probably the part linus objected to--I couldn't imagine trying to track all the kernel patches that way. maybe with a gui front-end and database to keep track of the options... :) It's not easy to perform clean backups or mirrors either, and the command line options are neither elegant nor consistent. Oh yes, and it could be faster. That's my personal list so far.

Basically, it's a hack on top of rcs, and it's starting to show. It could probably benifit from a complete rewrite in the next year or so, with the addition of things like a security model and support for distributed (and mirrored) repositories. bitkeeper is something like this, but not free software. prcs is another, gpl'd, attempt headed by Josh MacDonald, author of xdelta; there's no client/server for it yet, though. Personally, I'd like to see a cross between cvs, an eternity server, and Debian's apt package tool. :)

Nevertheless, I think it works fine for medium sized projects and really helps facilitate/speed up internet-based development. Beats mailing patches around!

Personally, I think it's wonderful. My most extensive experience with it has been in the glx project, and it's worked very well. Not being a professional programmer, it's the most sophisticated system I've every used. From talking to some of my friends, I understand it has some advantages over a lot of the commercial offerings, so we may not be missing much. :) I also use it to keep track of the data and analysis routines for my scientific work.

That said, there are lots of problems. It's not been terribly stable in my experience. It has poor support for binary files. Administration isn't fun (or easy) and it's difficult to set up securely. It's not very smart: particularly glaring ommisions are that you can't remove directories once they're added to the source tree (!), and moving files around is a pain. The notion of "code branches" could be more powerful (and easier to use). It's not easy to perform clean backups or mirrors either, and the command line options are neither elegant nor consistent. That's my personal list so far. :)

Basically, it's a hack on top of rcs, and it's starting to show. It could probably benifit from a complete rewrite in the next year or so, with the addition of things like a security model and support for distributed (and mirrored) repositories. bitkeeper is something like this, but not free software. prcs is another, gpl'd, attempt headed by Josh MacDonald, author of xdelta; there's no client/server for it yet, though.

Personally, I think it's wonderful. My most extensive experience with it has been in the glx project, and it's worked very well. Not being a professional programmer, it's the most sophisticated system I've every used. From talking to some of my friends, I understand it has some advantages over a lot of the commercial offerings, so we may not be missing much. :) I also use it to keep track of the data and analysis routines for my scientific work.

That said, there are lots of problems. It's not been terribly stable in my experience. It has poor support for binary files. Administration isn't fun (or easy) and it's difficult to set up securely. It's not very smart: particularly glaring ommisions are that you can't remove directories once they're added to the source tree (!), and moving files around is a pain. The notion of "code branches" could be more powerful (and easier to use). It's not easy to perform clean backups or mirrors either, and the command line options are neither elegant nor consistent. That's my personal list so far. :)

Basically, it's a hack on top of rcs, and it's starting to show. It could probably benifit from a complete rewrite in the next year or so, with the addition of things like a security model and support for distributed (and mirrored) repositories. bitkeeper is something like this, but not free software. prcs is another, gpl'd, attempt headed by Josh MacDonald, author of xdelta; there's no client/server for it yet, though.

It's rather well explain ed on the SETI@Home web site, just look more closely.

• The 6502 appeared in 1975.
• Apollo was designed in the 1960's and the last mission was in 1972.

Here are a few more links for more information about HTTP and some neat things that are being done with it...

• Get the latest dirt from the World Wide Web Consortium.
• RFC 2616: Hypertext Transfer Protocol -- HTTP/1.1 ( text, PostScript, PDF)
• Berkeley's TranSend service is a cluster of workstations working together to act as a massive HTTP proxy. This proxy "transforms" Web pages based on clients settings. Was the basis of the ( now-commercial) Top Gun Wingman Web browser for the PalmPilot.
• The Anonymizer acts as a proxy that strips out all the unwanted/unneeded header lines that your Web browser sends.

I had started hacking together an HTTP/1.1-compliant proxy in perl that did on-the-fly compression if the client supported it, but I never got around to completing it. Initial results were impressive, especially when it was paired with a caching proxy like Squid or a CacheFlow box. Of course, with DSL and cable modems getting more widespread use, people like myself that are still pinned to a 33.6k connection are being left behind.

Caching/compressing/proxying is still in widespread usage outside North America (most notably Australia and European countries). Their problem was (is!) outrageous access prices and relatively slow overseas connections, so they've been using caching for a long time to help solve it. The US and Canada have solved their "problem" of Web pages not instantaneously loading by throwing more bandwidth at it...

Here are a few more links for more information about HTTP and some neat things that are being done with it...

• Get the latest dirt from the World Wide Web Consortium.
• RFC 2616: Hypertext Transfer Protocol -- HTTP/1.1 ( text, PostScript, PDF)
• Berkeley's TranSend service is a cluster of workstations working together to act as a massive HTTP proxy. This proxy "transforms" Web pages based on clients settings. Was the basis of the ( now-commercial) Top Gun Wingman Web browser for the PalmPilot.
• The Anonymizer acts as a proxy that strips out all the unwanted/unneeded header lines that your Web browser sends.

I had started hacking together an HTTP/1.1-compliant proxy in perl that did on-the-fly compression if the client supported it, but I never got around to completing it. Initial results were impressive, especially when it was paired with a caching proxy like Squid or a CacheFlow box. Of course, with DSL and cable modems getting more widespread use, people like myself that are still pinned to a 33.6k connection are being left behind.

Caching/compressing/proxying is still in widespread usage outside North America (most notably Australia and European countries). Their problem was (is!) outrageous access prices and relatively slow overseas connections, so they've been using caching for a long time to help solve it. The US and Canada have solved their "problem" of Web pages not instantaneously loading by throwing more bandwidth at it...

I've also heard that someone got documentation for the mpeg2 daughterboard matrox sells for the g100/200/400 graphics cards. They were going to write a driver for it, but I haven't heard anything since. I think there's supposed to be a website at livid.on.openprojects.net sometime soon, but there's nothing there as of this posting.

See also mpeg.openprojects.net for a what progress there's been on an open software decoder. Unfortunately, the MPEG2/DVD standard is pretty well tied up with patents (worse than mp3).

Finally, the original Berkeley group demonstration code is still available.

I've also heard that someone got documentation for the mpeg2 daughterboard matrox sells for the g100/200/400 graphics cards. They were going to write a driver for it, but I haven't heard anything since. I think there's supposed to be a website at livid.on.openprojects.net sometime soon, but there's nothing there as of this posting.

See also mpeg.openprojects.net for a what progress there's been on an open software decoder. Unfortunately, the MPEG2/DVD standard is pretty well tied up with patents (worse than mp3).

Finally, the original Berkeley group demonstration code is still available.

What ELSE are you doing with your system? I'm running SETI@Home on a Celeron 400 pushed to 450. It pumps out a data unit every 13 hours or so running Win98.

Why be a noone on Team /.? If you're going to join a SETI@Home team, you should really join the Quake Team.
A team where you can actually find your own stat istics.

The SETI project is important...who knows what we may find with this much computing power all interlinked working towards the same goal. this is much more important then decrypting :).


while your at it:
join the Black Belts SETI Team!


Sensei

Once you have this installed, you have a nice RF pipe that any malware tojan can use to broadcast your data, even if you are not connected to the net at the time. All it needs to do is toggle DTR and RTS, and poof, you are on the air.

does anyone know what the table full of numbers on the bottom
of http://seti athome.ssl.berkeley.edu/about_seti/about_seti_at_h ome_1.html
is for? at least they don't seem to correspond to ascii, or do they?
we might need a new distributed computing client to solve this ;)

Even if these folks had the compilers that would allow you to take large chunks of code, convert it into a hardware representation and program the FPGA to execute it you still have to get have some DATA to feed the instruction stream! The only people that seem to understand true parallel programming models seem to be the people at Tera Computer). They have the only architecture that can do a context switch on each instruction to allow the processors to execute those instructions that happen to be executable because the operand data fetches are complete. Everyone else (Compaq(DEC), Intel, AMD, Sun, SGI, etc.) consume huge amounts of chip real estate with primary & secondary caches rather than really solving the problem of memory latency. The old CPU/Cache IS DEAD in the long run (the chips get too hot). What will work are architectures like Tera's and/or approaches like " Processor in Memory"/" Intelligent RAM"/" Embedded DRAM" that are innovative ways of dealing with the problem of operand latency and memory bandwidth.

I'll take that bet...

Their basic premisse is that Look at all the ways you use radio transmissions (...) On earth, we've already decided on the most efficient method! ( seti)

Excuse me, we have decided? Dinosaurs roamed the earth 65 million years ago, and we have used radio transmissions now for only slightly more than a hundred years. What if we find something better than radio signals in say a hundred, a thousand, or even a hundred thousand years.

These people have zero respect for time. These people are not scientists.

Their database only has Intel and one clone (AMD). What about real processors? They have some links for Alpha (just comments, nothing in their database), but what about G3 (PowerPC), Sparc, etc., etc.? Of course, with Intel processors it's more important because you have to overclock them just to get them up to the regular speeds of the other processors . . .

Anyway, in case anyone is interested, here's an overclocking page for Mac hardware, and here's one for Amigas. I'm sure there are lots of others out there.

Not so. The SETI project at Arecibo is only recording 1/40th of the total bandwidth available through the spectrometer. Even factoring in a pessimistic 50% observing efficiency (time the telescope is actually availble for SETI data taking) we need 5.6 million 2 day average computers!

Mike
--

Yup, read the FAQ.

They can increase the amount of data available to process by getting a second recorder.

O.K.,

Go to their web site and click on Groups. Then click on clubs and then Team Slashdot and then Join this group.

Better yet, just click HERE

O.K.,

Go to their web site and click on Groups. Then click on clubs and then Team Slashdot and then Join this group.

Better yet, just click HERE

O.K.,

Go to their web site and click on Groups. Then click on clubs and then Team Slashdot and then Join this group.

Better yet, just click HERE

O.K.,

Go to their web site and click on Groups. Then click on clubs and then Team Slashdot and then Join this group.

Better yet, just click HERE