A computer language/compiler/architecture/OS that allows ideas to be expressed without fault, i.e. a computer system that never crashes regardless of input.
My original post postulated that the alarmist nature of the original post was out-of-place. I attempted to back this up with an argument that compared 2 chips and tried to demonstrate that the 30 or so errata in the Intel Duo was a reasonable figure.
With that out of the way.
I feel my metric is a fair, albeit rough way of measuring how many errata a chip should have as compared with its peers. I understand your argument that CPUs are complex but you are, in effect copying a design and that, in your opinion gate level effects would manifest themselves as reductions in yields, not errata items. I disagree that a memory controller is less complex than a CPU and I also disagree that gate level effects are not the root causes of some errata.
Back to the issue at hand, do you think that 30 errata are excessive in this case, why do you feel that way and what metric would you use to measure errata thersholds?
With regard to your 'dual core' argument. It is true that each core is the same, but the memory controller will be more complex (among other things) so your still assuming more complexity.
With regard to your 'algorithmic' argument. Your assertion that bugs only come from the algorithms expressed in hardware is wrong. At 65 nm tunneling electrons and short gate effects can cause otherwise functional blocks of logic to malfunction. Indeed, a chip may have many such malfunctions caused by these gate level effects. An errata list will contain issues whose base cause is the silicon not the designer.
Of course the 4x metric is fairly fantastic and a gross oversimplification, but it is a metric. In my post I should have said something to the effect that 4x would be a good upper bound on the number of bugs a chip would have with twice the complexity and half the feature sizes of another chip. If I saw a chip with 10x the number of errata of under the same circumstances, then I would be surprised.
Every chip has an errata sheet. The number of elements on the errata sheet is proportional to the complexity of the chip in question and the age of the process it is fabricated with. The Core Duo by Intel has complexity comparable to 2 PowerPC 970FX Series which its replacing and is fabricated using 65 nm technology. The PowerPC 970FX is fabricated using a 90 nm process. It has 24 items on its errata sheet (http://www-306.ibm.com/chips/techlib/techlib.nsf/ techdocs/79B6E24422AA101287256E93006C957E/$file/97 0fx_errata_dd3.x_v1.6.pdf). Therefore its expected that a chip fabricated on a substrate whose minimum feature sizes are half those of the other chip and whose complexity is double the other chip would have 4x the errata items of the other chip.
Check out tcl/tk. Its very easy to use as well as being enomously powerful. Its also supported on Windows, Linux and Solaris as well any flavor of Unix you want.
I'm gonna take the hip ol' gezzer approch. I suggest becoming a sys admin for a year. See if you really like it. A job at the enterprise level (especially in system critical applications) can be very hard. Also, I learned a great many things in college around the areas of databases, distributed applications and the like. These have helped me immensly in understanding sys administration below the GUI interface. Another feather in your hat is networking. If you can learn how to manage and layout a network you'll be dynamite. I suggest three books The UNIX System Administration Handbook and the Linux System Administration Handbook and Computer Networks - A Top Down Approch. These have great insights into all manner of networking and sys admin in general.
Remember one thing though...don't believe the hype You will use you college education, just not directly. College gives you problem solving skills that are not obtainable from practical experience.
I heard that on NPR last night too. The scary part is that the same kids that are are attratced to viloent video games are also attracted to Aryan Nation groups.
I've got another study...how many middle class boys have a problem with video game ratings?
The most insightful thing I ever grasped was that programming transends the languge with which its written. Unfortunatly we mere mortals have not devised an all encompasing language with which we can describe any system so we're left with C++, lisp, scheme etc...
I think the coolest programming languages support continuations so my vote would be lisp.
No, hollywood will never give an accurate representaion of the box or those who would like to control it. What they do do is infuse people who know nothing about computers to go out and learn about them. Its not that their representations are false, its just that they're inacuate-but they're fun to watch. So to all of you would be judges out there....shut up and enjoy what computers might be like. Use the movies to excite your curiosity about new projects, etc.
Slashdot Mirror
What a load of hog-wash.
The next big thing:
A computer language/compiler/architecture/OS that allows ideas to be expressed without fault, i.e. a computer system that never crashes regardless of input.
My original post postulated that the alarmist nature of the original post was out-of-place. I attempted to back this up with an argument that compared 2 chips and tried to demonstrate that the 30 or so errata in the Intel Duo was a reasonable figure.
With that out of the way.
I feel my metric is a fair, albeit rough way of measuring how many errata a chip should have as compared with its peers. I understand your argument that CPUs are complex but you are, in effect copying a design and that, in your opinion gate level effects would manifest themselves as reductions in yields, not errata items. I disagree that a memory controller is less complex than a CPU and I also disagree that gate level effects are not the root causes of some errata.
Back to the issue at hand, do you think that 30 errata are excessive in this case, why do you feel that way and what metric would you use to measure errata thersholds?
Thank you for your well reasoned response.
With regard to your 'dual core' argument. It is true that each core is the same, but the memory controller will be more complex (among other things) so your still assuming more complexity.
With regard to your 'algorithmic' argument. Your assertion that bugs only come from the algorithms expressed in hardware is wrong. At 65 nm tunneling electrons and short gate effects can cause otherwise functional blocks of logic to malfunction. Indeed, a chip may have many such malfunctions caused by these gate level effects. An errata list will contain issues whose base cause is the silicon not the designer.
Of course the 4x metric is fairly fantastic and a gross oversimplification, but it is a metric. In my post I should have said something to the effect that 4x would be a good upper bound on the number of bugs a chip would have with twice the complexity and half the feature sizes of another chip. If I saw a chip with 10x the number of errata of under the same circumstances, then I would be surprised.
Every chip has an errata sheet. The number of elements on the errata sheet is proportional to the complexity of the chip in question and the age of the process it is fabricated with. The Core Duo by Intel has complexity comparable to 2 PowerPC 970FX Series which its replacing and is fabricated using 65 nm technology. The PowerPC 970FX is fabricated using a 90 nm process. It has 24 items on its errata sheet (http://www-306.ibm.com/chips/techlib/techlib.nsf/ techdocs/79B6E24422AA101287256E93006C957E/$file/97 0fx_errata_dd3.x_v1.6.pdf). Therefore its expected that a chip fabricated on a substrate whose minimum feature sizes are half those of the other chip and whose complexity is double the other chip would have 4x the errata items of the other chip.
Check out tcl/tk. Its very easy to use as well as being enomously powerful. Its also supported on Windows, Linux and Solaris as well any flavor of Unix you want.
I'm gonna take the hip ol' gezzer approch. I suggest becoming a sys admin for a year. See if you really like it. A job at the enterprise level (especially in system critical applications) can be very hard. Also, I learned a great many things in college around the areas of databases, distributed applications and the like. These have helped me immensly in understanding sys administration below the GUI interface. Another feather in your hat is networking. If you can learn how to manage and layout a network you'll be dynamite. I suggest three books The UNIX System Administration Handbook and the Linux System Administration Handbook and Computer Networks - A Top Down Approch. These have great insights into all manner of networking and sys admin in general.
Remember one thing though...don't believe the hype You will use you college education, just not directly. College gives you problem solving skills that are not obtainable from practical experience.
Just my two cents...
I heard that on NPR last night too. The scary part is that the same kids that are are attratced to viloent video games are also attracted to Aryan Nation groups.
I've got another study...how many middle class boys have a problem with video game ratings?
The most insightful thing I ever grasped was that programming transends the languge with which its written. Unfortunatly we mere mortals have not devised an all encompasing language with which we can describe any system so we're left with C++, lisp, scheme etc...
I think the coolest programming languages support continuations so my vote would be lisp.
No, hollywood will never give an accurate representaion of the box or those who would like to control it. What they do do is infuse people who know nothing about computers to go out and learn about them. Its not that their representations are false, its just that they're inacuate-but they're fun to watch. So to all of you would be judges out there....shut up and enjoy what computers might be like. Use the movies to excite your curiosity about new projects, etc.