It most certainly IS true, unless you've come up with some revolutionary new way to circumvent the laws of physics. You said "processor power versus heat output." A procesor that consumes thirty watts puts out about 102 BTUs of heat per hour. A processor that consumes 60 watts will put out about
205 BTUs of heat per hour. As I said, unless you can find some exotic way to radiate a lot of power as RF energy instead of heat, there's a linear relationship.
And here I was thinking that a company might be willing to pay a few bucks more for a system that consumed a fraction of the power, while performing just as well,
When buying supercomputers (which is what I was talking about, if you'd actaully bothered to read what I wrote), that's almost never an option.
If that was the case, there wouldn't be anything even resembling mainframes, as they would need a large ammount of space and heavy-duty cooling systems for all of their AMD Duron processors.
This shows a remarkable lack of understanding as to what a mainframe is. Get half a clue some time. Mainframes aren't known for having exceptionally high-performance processors; these days often an inexpensive desktop machine can outperform them on pure number crunching. Mainframes, however, have one heck of a lot more I/O bandwidth than a desktop machine, which is why they're better suited to running high-end database applications, transaction processing, and such.
If you want heavy-duty number crunching, you don't want a mainframe, you want a supercomputer.
That must be why Google's setup is so famous, because they are doing the same thing as everyone else, and disregarding the issue of heat.
Presumably this statement was intended to be sarcastic? I must be dense because I don't understand what you're getting at, or how it relates to the discussion.
And when Intel bought StrongARM, the architects left to found a startup called Alchemy, and created a great little MIPS-based processor core. Since then AMD bought Alchemy, so it must be like old home week at AMD.:-)
Red Hat doesn't have anywhere NEAR the resources needed to start building state-of-the-art (or even nearly-s.o.t.a.) microprocessors.
Not to mention that HP wants Alpha to dry up and blow away, since it competes with their home-grown stuff. If it weren't for contractual obligations, they'd drop it in a New York minute.
Nice dream, though.
At the moment it looks like the only way that mainstream processors are going to move to a (slightly) less ugly architecture is if AMD succeeds with the Hammer parts (Athlon 64 and Opteron). At least they've managed to increase the number of general registers.
I've noticed lately that Agilent is deemphasizing their old slogan, "innovating the HP way". Maybe they are trying to distance themselves from the HP fiasco. At this point, the HP name has lost most of its value, and people who buy Agilent products know this.
When they announced the spinoff, it was immediately obvious to almost everyone I know that the wrong part of the company was getting the HP name.
The reason why HP favors the itanium is because they spent billions of dollars co-creating it with Intel. [...] They want a return on their investment.
So you're basically saying that HP management doesn't understand the concept of "sunk costs"?
The fact that they've already spent billions of dollars on it doesn't necessarily mean that continuing to push it is the best business decision. Sometimes you have to realize that you made a bad choice, and write off the investment.
Otherwise you may just lose many billions more.
is basically a linear factor for all processors, unless you can find a processor that is radiating one heck of a lot more RF than the others.
OK, it's a joke. I know what you meant. But power consumption is only a relatively minor factor in the purchase decisions for massively parallel supercomputers. The main criterion is Gflops/dollar.
The FTC was sleeping when they allowed HP to acquire it.
If HP hadn't acquired Compaq, Alpha would still be just as dead. In fact, quite probably more dead. The FTC has screwed up on allowing many acquisitions, but I don't think this was one of them.
On the other hand, I think HP was incredibly stupid to buy Compaq. What a brilliant plan: "There's no money to be made on commodity PCs, so we'll buy another PC vendor that's in a world of hurt. We'll still lose money on each one, but we'll make it up on volume." I'm pretty sure it's dragging them down, but they can play all sorts of accounting tricks to make it look OK on the books.
And all C programmers get screwed over by it, sooner or later. I've had to fix code that was out in the field for years with one of these stupid bugs in it. I would much rather have the language defined so that the compiler can catch this.
Sounds like you belong to the "I don't want any damn compiler telling me what I can and can't do" camp. If you don't like having the compiler check for obvious errors, you should write in BCPL, or maybe assembly language.
I looked over the FAQ, and am thoroughly unimpressed with some aspects, like the decision not to fix the behavior of the switch statement (i.e., falling into the next case if there's not a break).
The stated argument is that it's desirable to make the language "look like" C. If you want a language that looks like C, use C. Personally, I want someting better. In fact, I think that making Java look like C was one of Java's biggest mistakes; it lulls the programmer into a false sense of familiarity when in fact Java's semantics are much different than C's.
And if you just want to make sure that C programmers don't get burned by their old habits, there's a simple solution to that as well. Require that each case end with a "break" or "fallthrough" statement. That way there's no
ambiguity if the user forgets the break, it will produce a compile error.
What does the Z8 Encore bring to the party that you can't get from other, better-supported microcontrollers? For instance, the
TI MSP430 parts? It's a low-cost 16-bit microcontroller with a PDP-11-like architecture. Very nice for writing assembly, AND it is well-suited to a C compiler (unlike the Z8).
Or, if you want to attach the nodes to Ethernet, the Ubicom IP2022. It's still reasonably inexpensive, but has 64K of Flash, 20K of RAM, and built-in 10baseT Ethernet support. That way you don't have to invent any new protocols to wire the things up.
The only drawback of the IP2022 is that the SDK is somewhat expensive. If you just want basic tools (a compiler, assembler, linker, and debugger), you can use the GNU tools. But the SDK includes the Ethernet driver, TCP/IP stack, small HTTP server, etc., which would be useful for an application like this.
Disclaimer: I've worked for Ubicom for a little over four months. Before that, I was a satisfied customer, having designed their SX part into the first generation ReplayTV box to handle IR remote functions.
Not true. I've purchased multiple "flying machines" in the US without any approval from the FAA, and have operated them in the US without regard to FAA rules and regulations, and they don't care in the slightest. Not every "flying machine" is under their jurisdiction.
Even if a working Solotrek would be covered by FAA regs, if they sell a scrapped non-functional prototype (which it is), they don't have to make the buyer promise not to fly it in order to make the FAA happy. It would be up to the buyer to meet any FAA requirements.
This is just like the situation if I sell you a scrapped non-functional prototype of an automobile. I don't have to meet any DOT regs for it, since it isn't a car. If you turn it into a car, you have to deal with the regs.
If a developer wants to pursue this dual-license scheme and also to accept contributions of patches or extensions from the user community, it is important for the developer to inform contributors of this up front, and to require that ownership of contributions transfer to the developer.
Certainly you can't release a piece of code under the GPL, allow an active user community to invest a lot of work improving it under the assumption that it is always going to be GPL'd, then take the result commercial. Without copyright assignments in place, you'd be liable for infringement.
I've released some minor GPL'd code for which I've also sold a commercial, non-GPL license. In my case, the point is that I'm perfectly willing to allow the community to use it under the GPL, and I'm perfectly willing to let my commercial licensee pay me for the right to use it without the GPL obligations. But as the developer, I am NOT granting anyone else the right to modify it then sell their own version as closed source.
I view the GPL obligations as being in a sense "payment" for the work I put into the code. You can pay me by honoring the GPL terms, or by negotiating to pay me cash. Why should I allow a third party to take my work "closed" without paying me?
Note that I am not criticizing people who chose to release code under a non-copyleft license, such as the BSD or X licenses. It is clearly the perogative of the copyright owner to choose what terms he or she wishes to use.
When Microsoft started lobbying Congress trying to convince them that publicly funded software shouldn't be GPL's, I started trying to push this argument. The GPL is ideal for publicly funded software, because the organization that holds the copyright (for instance, a University) can charge money for commercial licenses. What Microsoft was really trying to do was to make sure that they continued to get free handouts of BSD-licensed code.
I've published a few obscure pieces of code under the GPL. Though I never actively tryied to sell commercial licenses, I was approached by one company and was paid several thousand dollars for a commercial license. Not bad for somthing obscure that I never expected to make money on. Imagine what a person could make writing more mainstream GPL'd code. For such things, public release under the GPL is like free advertising.
I recommend that software authors who are interested in trying this with the GPL'd code place prominent notices in their README files and web pages, because some potential customers may not be aware that they can negotiate a non-GPL license from the copyright owner.
This may be the first no-program guide-fee commercial PVR.
The first several generations of ReplayTV boxes
didn't (and still don't) require a paid subscription, though their currently offered models do. Thus RCA definitely isn't the first to do that.
(I was the third employee of Replay, which was originally Pacific Digital Media and has since been acquired by Sonic Blue.)
This is truly an amazing milestone for those of us who once spent $500 for the fantastically large 10MB models.
I dreamed of being able to get 10MB for only $500. My first hard drive was 5MB, and cost $1100 for the drive, $350 for the controller, and $200 for the host adapater. That was only 22 years ago; people were paying over $7000 for a 30MB drive (8-inch or 14-inch) around 1980.
And no, I'm not going to regale you with tales of having to walk to school 22 miles in the snow, uphill both ways. I only had to walk about 1.5 miles, and it was fairly flat, though there was in fact snow.
Penrose argues convincingly that consciousness is a QM phenomenon exhibited by most life forms,
s/convincingly/unconvincingly/
It was hard for me to tell what Penrose's argument was, but it seems to be:
Computers are, by design, deterministic. Barring hardware failures, if you start with the computer in the exact same state twice, and give it the same inputs at the same times, you will get the same outputs (provably false, due to metastability of flip-flops when inputs are changed too soon before or after a clock transition)
The behavior of neurons is nondeterministic due to QM effects (unproven but plausible)
The non-determinism is necessary to intelligence (unproven)
Therefore computers can't be intelligent
The argument is unconvincing due to point one being false and the total lack of proof of points 2 and 3.
Even if point 1 was true and points 2 and 3 were proven, that still wouldn't rule out the possibility of designing a computer that is deliberately non-deterministic, using a noise source.
I thought maybe that Penrose was claiming that QM effects are not really random, but instead are biased toward supporting intelligence. I don't know how one could hope to prove such a claim, but even if it were true it wouldn't preclude building a machine to exploit that effect.
I could buy a kit of new-old-stock pads from one supplier for about $50, but that just isn't the proper way to restore an old vintage computer, it's got to be done right, and that means canniballizing old hardware on the cheap, not buying new parts off the shelf..
IMNSHO, what you should use is new pads, not NOS. Even NOS pads will have deteriorated some over the years. Eventually the foam rubber turns to disgusting sticky goo. But AFAIK, they are no longer made.:-(
I've got two Sol 20 systems, and between the two have almost enough good pads to make one keyboard fully functional.
I recently got some third-party Z-80 and 80-column upgrade daughterboards, which I have yet to try out. I'm not sure whether I have a suitable CP/M BIOS for use with the 80-column display; if not I'll have to hack it myself.
Slashdot Mirror
Unless SCO has some patent that covers System V but not System III or 7th Edition, and Linux uses the patented technique, it's not a problem.
If you want heavy-duty number crunching, you don't want a mainframe, you want a supercomputer.
Presumably this statement was intended to be sarcastic? I must be dense because I don't understand what you're getting at, or how it relates to the discussion.And when Intel bought StrongARM, the architects left to found a startup called Alchemy, and created a great little MIPS-based processor core. Since then AMD bought Alchemy, so it must be like old home week at AMD. :-)
(reference to a Inspector Dim of the Yard, from a Monty Python sketch)
Before Samsung, Mitsubishi had an Alpha license. That went nowhere also.
Not to mention that HP wants Alpha to dry up and blow away, since it competes with their home-grown stuff. If it weren't for contractual obligations, they'd drop it in a New York minute.
Nice dream, though.
At the moment it looks like the only way that mainstream processors are going to move to a (slightly) less ugly architecture is if AMD succeeds with the Hammer parts (Athlon 64 and Opteron). At least they've managed to increase the number of general registers.
When they announced the spinoff, it was immediately obvious to almost everyone I know that the wrong part of the company was getting the HP name.
The fact that they've already spent billions of dollars on it doesn't necessarily mean that continuing to push it is the best business decision. Sometimes you have to realize that you made a bad choice, and write off the investment. Otherwise you may just lose many billions more.
OK, it's a joke. I know what you meant. But power consumption is only a relatively minor factor in the purchase decisions for massively parallel supercomputers. The main criterion is Gflops/dollar.
Ah, but they wouldn't have to pay for it twice. That's the beauty of it.
On the other hand, I think HP was incredibly stupid to buy Compaq. What a brilliant plan: "There's no money to be made on commodity PCs, so we'll buy another PC vendor that's in a world of hurt. We'll still lose money on each one, but we'll make it up on volume." I'm pretty sure it's dragging them down, but they can play all sorts of accounting tricks to make it look OK on the books.
Sounds like you belong to the "I don't want any damn compiler telling me what I can and can't do" camp. If you don't like having the compiler check for obvious errors, you should write in BCPL, or maybe assembly language.
The stated argument is that it's desirable to make the language "look like" C. If you want a language that looks like C, use C. Personally, I want someting better. In fact, I think that making Java look like C was one of Java's biggest mistakes; it lulls the programmer into a false sense of familiarity when in fact Java's semantics are much different than C's.
And if you just want to make sure that C programmers don't get burned by their old habits, there's a simple solution to that as well. Require that each case end with a "break" or "fallthrough" statement. That way there's no ambiguity if the user forgets the break, it will produce a compile error.
Or, if you want to attach the nodes to Ethernet, the Ubicom IP2022. It's still reasonably inexpensive, but has 64K of Flash, 20K of RAM, and built-in 10baseT Ethernet support. That way you don't have to invent any new protocols to wire the things up.
The only drawback of the IP2022 is that the SDK is somewhat expensive. If you just want basic tools (a compiler, assembler, linker, and debugger), you can use the GNU tools. But the SDK includes the Ethernet driver, TCP/IP stack, small HTTP server, etc., which would be useful for an application like this.
Disclaimer: I've worked for Ubicom for a little over four months. Before that, I was a satisfied customer, having designed their SX part into the first generation ReplayTV box to handle IR remote functions.
Even if a working Solotrek would be covered by FAA regs, if they sell a scrapped non-functional prototype (which it is), they don't have to make the buyer promise not to fly it in order to make the FAA happy. It would be up to the buyer to meet any FAA requirements.
This is just like the situation if I sell you a scrapped non-functional prototype of an automobile. I don't have to meet any DOT regs for it, since it isn't a car. If you turn it into a car, you have to deal with the regs.
Certainly you can't release a piece of code under the GPL, allow an active user community to invest a lot of work improving it under the assumption that it is always going to be GPL'd, then take the result commercial. Without copyright assignments in place, you'd be liable for infringement.
I view the GPL obligations as being in a sense "payment" for the work I put into the code. You can pay me by honoring the GPL terms, or by negotiating to pay me cash. Why should I allow a third party to take my work "closed" without paying me?
Note that I am not criticizing people who chose to release code under a non-copyleft license, such as the BSD or X licenses. It is clearly the perogative of the copyright owner to choose what terms he or she wishes to use.
I've published a few obscure pieces of code under the GPL. Though I never actively tryied to sell commercial licenses, I was approached by one company and was paid several thousand dollars for a commercial license. Not bad for somthing obscure that I never expected to make money on. Imagine what a person could make writing more mainstream GPL'd code. For such things, public release under the GPL is like free advertising.
I recommend that software authors who are interested in trying this with the GPL'd code place prominent notices in their README files and web pages, because some potential customers may not be aware that they can negotiate a non-GPL license from the copyright owner.
(I was the third employee of Replay, which was originally Pacific Digital Media and has since been acquired by Sonic Blue.)
And no, I'm not going to regale you with tales of having to walk to school 22 miles in the snow, uphill both ways. I only had to walk about 1.5 miles, and it was fairly flat, though there was in fact snow.
It was hard for me to tell what Penrose's argument was, but it seems to be:
- Computers are, by design, deterministic. Barring hardware failures, if you start with the computer in the exact same state twice, and give it the same inputs at the same times, you will get the same outputs (provably false, due to metastability of flip-flops when inputs are changed too soon before or after a clock transition)
- The behavior of neurons is nondeterministic due to QM effects (unproven but plausible)
- The non-determinism is necessary to intelligence (unproven)
- Therefore computers can't be intelligent
The argument is unconvincing due to point one being false and the total lack of proof of points 2 and 3.Even if point 1 was true and points 2 and 3 were proven, that still wouldn't rule out the possibility of designing a computer that is deliberately non-deterministic, using a noise source.
I thought maybe that Penrose was claiming that QM effects are not really random, but instead are biased toward supporting intelligence. I don't know how one could hope to prove such a claim, but even if it were true it wouldn't preclude building a machine to exploit that effect.
I've got two Sol 20 systems, and between the two have almost enough good pads to make one keyboard fully functional.
I recently got some third-party Z-80 and 80-column upgrade daughterboards, which I have yet to try out. I'm not sure whether I have a suitable CP/M BIOS for use with the 80-column display; if not I'll have to hack it myself.
If you had an Apple I, though...
If it was present, will Apple release fixes for those, or just force everyone to buy the 10.2 upgrade?