Thanks for the interesting responses, folks. I feel I've learned a lot. Perhaps I didn't RTFA well enough, but I was under the impression that these were two separate news items: one about wifi chipsets, and another about a new lithography technique that Intel would be using ubiquitously, including for future CPUs.
I definitely agree about the power savings from the process shrinks (thanks for the correction!); we saw those in the Coppermine->Tualatin shrinks and the Willamette->Northwood shrinks, where lower voltages could be used for a fixed speed (and indeed for higher speeds), leading to reduced power consumption and heat dissipation.
However, I was also thinking of the point brought up above, where the phenomenon of leakage gets worse the smaller the features are (and the lower the voltages are, since the relative ratio of such noises will thereby increase). Although the press release didn't mention these (and even cited the power reductions from above), I'm not entirely convinced that the problem has been solved (although I take heart the SiO2 replacement). After all, Intel was saying very recently that the Northwood->Prescott would yield power savings, but it turned out that at this scale, the leakage problems nearly negated the expected savings.
Very interesting. I wish I understood these issues better, and I'm glad that there are others here to help explain!!! Thanks -- Paul
But the leakage current problems have been increasing with process shrinks (not just at Intel, but also at IBM and AMD). So they can use even smaller lithography. Great. Will the leakage current and associated heat suck even worse than Prescott?
Shostak assumed that computer processing power will continue to double every 18 months until 2015- as it has done for the past 40 years. From then on, he assumes a more conservative doubling time of 36 months as transistors get too small to scale down as easily as they have till now.
As has been seen with 65nm Prescotts and difficulties with die shrinks to this size by IBM and AMD as well, we may already be at or near the point that they had assumed for 2015.
The comment about it being Warren Buffet, rather than Warren Beatty, is right. Warren Buffet is often referred to as the "Oracle of Omaha" for his success in understanding and predicting the financial / economic climate.
There's actually a neat story about Warren Buffet when he was young. There was a man whose young neighbor once approached him with a (set of) wild investment idea and asked him if he wanted to go in on it with him. The guy figured the neighbor for either a con artist or a hack and turned him down. Well, decades later, that young man is a very successful Warren Buffet, and I'll bet the other is kicking himself for turning down that once-in-a-lifetime opportunity.
Sorry, I looked all over for a link to that anecdote, but I couldn't find it among the sea of
"young Warren Buffet neighbor" google hits. -- Paul
On the one hand, that's the inherent risk with any technology as it becomes increasingly accessible and "user-friendly".
On the other hand, are these systems going to be cheap enough that we have to worry about script kiddies? If computers still cost $5000+, I doubt script kiddies would be such a rampant problem on the net. -- Paul
My thought is, if the hijacked machine is chocking on all these calculations, at least they'll notice that there's something wrong with their machines. (Which would be an advance in and of itself.)
The point you are trying to emphasize is a good one, but as I mentioned before, the key is testing. I've tested the code on this machine for problems that have analytical solutions, and the results agreed with the expected order of accuracy. The results also matched those obtained with non-overclocked machines.
The fact of the matter is, most machines aren't actually running precisely at spec. Do you really think that your machine is running with a precise FSB of 200.00 MHz (or whatever the spec FSB is for your machine)? It's not uncommon to see 1-5% variation on machines running "at spec".
I'm not shooting for the moon here. I'm applying a moderate overclock that's well within the range expected for this architecture (officially up to 3.6 GHz for the Northwood core), and I'm exercising due caution by conducting extensive testing.
The key to a successful overclock is to test, test, and test until you've found the top end of the stable range, and then back down a bit. An overclock that isn't stable isn't a successful overclock.
I overclock a P4 3.0C to 3.5 GHz and use it for long-term computational biophysics. (Numerical simulation of tumor growth and chemotherapy.) It's amazing what a difference a 17% speed boost can make in a 7-day calculation. I've tested both the memory and the CPU for stability and can vouch for it. Overclocking isn't just for games; it can make the hardware you can afford more productive for real work. It sure helped me to get my research done more quickly for my master's thesis.
Other people are using overclocked computers for useful things all the time. Consider the thousands who use overclocked computers to pump out WU's for Stanford's protein folding project (find out how proteins fold and misfold, help cure diseases, etc.). The top-producing teams are overclockers.au, HardOCP (lots of overclockers), and overclockers.com at Number 3.
I used to use TeraTermPro / TTSSH as well. It was very nice, but alas, TTSSH only has SSH 1.5 and most likely won't be updated to SSH protocol v. 2.0. AFAIK, That means that you won't get the most recent security fixes, as well as other nice features of SSH v. 2.0 (like compression).
A good alternative is PuTTY. Works like a charm in all flavors of Win32.
Their main problem seems to be summarized as such: How can we get rid of mesh nodes while still accurately representing the surface?
A related approach is: Given that we have N node points available, how can we distribute them among the surface to provide the most accurate representation? If we can add a few nodes, where would it be best to add them? If the mesh is good enough, where would it be best to remove them?
Then, they explain their approach a bit more:
Desbrun explains that his accomplishment was to simplify such a mesh, by combining as many of the little triangles as possible into larger elements without compromising the actual shape. Nearly flat regions are efficiently represented by one large, flat mesh element while curved regions require more mesh elements.
My colleagues, who do advanced computational fluid mechanics (applied to drop-drop collisions with surfactant diffusion, simulation of tumor growth, biomathematics, etc.) do similar work, but in a very different way. For a given surface mesh, they assign a sort of mesh energy to the whole structure, where the energy of any point depends upon the local curvature of the surface, etc., and they add a constraint or two. (I don't remember the full details.) It's very similar to a network of springs. Then, they apply standard optimization techniques to minimize the mesh energy. The result: a mesh that has the highest density where the surface has the highest curvature, and the lowest density where the surface is flattest. The method is fully adaptive and even works during and after surface topological changes, which is very useful in CFM, where your surfaces can break, merge, etc.
So, this idea of modeling surfaces by concentrating mesh nodes where the surface curvature is greatest with an automated, adaptive algorithm is not new, and it is very applicable in making more efficient use of limited computational resources.
WinSCP: free secure ftp client to move work around
PuTTY: free SSH client as mentioned above
MinGW & MSYS: free C++ compiler and environment, with useful things such as grep
Matlab: student version of linear algebra sofware. Mostly used for visualization these days
Miktex: fantastic distro of LaTeX for Windows; for writing homework and scientific papers
For additional productivity:
MS Powertoys / TweakUI: free download to specify documents folders, etc., and tweak other behavior; available at MS's site. Also "command prompt here" powertoy, etc.
Google Toolbar: iexplorer is good enough for me (might as well use i0t, since it's already in system memory) when coupled with the google bar and popup blocking.
Norton AntiVirus: not dumb enough to open attachments, but it's good to scan downloads before unzipping
stanford folding project: great for testing stability of the overclock while contributing to a good cause
For fun:
Flavor of the Month Game Demo: I'm too cheap to buy most games, but I'm getting a lot of fun out of the Halo and FarCry demos at the moment.
Linux (gentoo build from stage 1): For working:
X: because X tunneling is nice
OpenBox: I like their no-nonsense window manager
xv: great for quick viewing of graphics files
Intel's C++ Compiler: free C++ compiler in linux, and beats the pants off of G++ for fast-running code
Matlab: student version of linear algebra sofware. Mostly used for visualization these days
LaTeX: for writing homework and scientific papers
For additional productivity:
Firefox: You all know this one.
nvidia drivers: gives better performance in matlab
stanford folding project: great for testing stability of the overclock while contributing to a good cause
Yes, except that if any one of your 2000 members abuses the system, you'll be the one in trouble for it. (Since personal.mail is registered to you.) And then you'll be facing the money-back demands of the other 1999 members who suddenly lost their service. -- Paul
Good point there. Actually, as I thought about it more, I realized that non-volatile RAM would probably never be as fast as "regular" RAM.
However, what would probably be a redeeming "in-between" solution would be a cheap, non-volatile solid state disk. Non-volatile RAM that's cheap enough and reliable enough to use a really fast hard drive. Not to store all data on, but big enough to store the OS on. One could always have a larger, traditional slow hard drive for data and a smaller, faster hard drive for the OS. Some people already do this with compact flash, etc., but those media aren't exactly made to be read / written a huge number of times like a partition with an OS needs to be...
Won't this all be moot after non-volatile RAM (that's as fast as current RAM) is developed and widespread? AFAIK, fast non-volatile RAM technology is well under way and within reach in the next several years. The only reason we currently need to read from the hdd is that the RAM is volatile and loses its contents on a power down.
We can already see how this would work in an S3 suspend: the entire PC but the memory (and perhaps the CPU) is powered down. Press the keyboard, and you're back in your OS in seconds. Replace present-day memory with non-volatile memory, and you're completely there. -- Paul
Re:Is this really going to change the "big picture
on
AT&T Labs' Brain Drain
·
· Score: 1
I'm told 3M is one such company, that does lots of "pure research" in order to come up with new products. Of course, IBM and Microsoft are mentioned - because they're already known to do this and have deep pockets. But R&D is the key to long term advancement of ANY business.
Except that W. James McNerney, Jr., a product of the GE management machine, became the Chairman of the Board and CEO of 3M in recent years.
I remember the chills that went through the research department when that happened (I was an intern in fiber optics at the time), and indeed, the writing was on the wall for imminent changes. The traditional "15% rule", whereby researchers could spend 15% of their time on their own research ideas which often lead to very innovative, lucrative new products (e.g., Post-It notes, UV-hardened polymer), was in jeopardy when I left. I'm not sure where that went since then, as I'm not privy to the 3M internals.
Here's an interesting link on 3M research culture, BTW. -- Paul
I remember being really gung ho about Linux and Open Source after trying my first distro: Mandrake 8.0. At that time, Win98SE and WinME were the dominant flavors (WinXP was just starting to come out), and I found that the Mandrake install did a better job of detecting most of my hardware than the MS install.
Eager to support the cause, I plucked down $100 to preorder the Pro version of the upcoming Mandrake distro. "Cool, I'm supporting open source. I'm doing my part," I thought, and I'd even get some of the CD's early for my pre-order. So I ordered, my credit card was charged, and day after day, week after week, no product arrived. And day after day, my emails to the company weren't answered. There were no real announcements anywhere to be seen about what was causing the delay. Finally, after a bit more than a month of this, I finally called the company at my own expense and had my order cancelled. (And even that required quite a run around, as the number listed on Mandrake's site didn't seem to be a direct number, so I had to call a few times to connect with anybody.)
And this is how they treated an eager customer. Hardly the way to treat a paying customer! I sure wouldn't want to run my business this way.
Granted, things are better now, but when your business isn't run like a business, don't expect customers to stick around. -- Paul
Hmm, it looks "cute". I'd imagine that the "form before function" people will buy this, female or male. It'll look great in that new VW bug along with the Michael Graves teapot.
That's a good strategy. An efficient heatsink with a quiet fan can go a long ways. I installed a Thermalright SP-94 (all-copper with heatpipes) with a 92mm Panaflo M running at 8V. It keeps my P4 3.0C (overclocked to 3.5 GHz) pretty cool and pretty quiet. My GF Ti4200 is passively cooled with a Zalman heatsink. My current power supply is pretty quiet, but I fixed my old one up by removing the airflow-restricting punched vents with a Dremel and replacing the cheapo loud fan with a quiet Panaflo L. Also added a direct intake of airflow right about the CPU / HSF in the side of a case with an 80mm Panaflo L running at 7V.
It's amazing how quiet this system is. It's even quieter than my previous PIII system was at half the speed. Up until recently, it was much quieter than my wife's HP Celeron 800 system. (We just finished modding that together to quiet it down...)
Really, if you're willing to put in a little bit of elbow grease and research here and there, you can wind up with a high-performance system that's quieter than most OEM systems.
Oh, as for hard drive noise that was mentioned above, the trick to try is to mechanically isolate the hard drive from the case, such as by suspension. See www.silentpcreview.com and www.overclockers.com for some ideas.:) -- Paul
Slashdot Mirror
Thanks for the interesting responses, folks. I feel I've learned a lot. Perhaps I didn't RTFA well enough, but I was under the impression that these were two separate news items: one about wifi chipsets, and another about a new lithography technique that Intel would be using ubiquitously, including for future CPUs.
I definitely agree about the power savings from the process shrinks (thanks for the correction!); we saw those in the Coppermine->Tualatin shrinks and the Willamette->Northwood shrinks, where lower voltages could be used for a fixed speed (and indeed for higher speeds), leading to reduced power consumption and heat dissipation.
However, I was also thinking of the point brought up above, where the phenomenon of leakage gets worse the smaller the features are (and the lower the voltages are, since the relative ratio of such noises will thereby increase). Although the press release didn't mention these (and even cited the power reductions from above), I'm not entirely convinced that the problem has been solved (although I take heart the SiO2 replacement). After all, Intel was saying very recently that the Northwood->Prescott would yield power savings, but it turned out that at this scale, the leakage problems nearly negated the expected savings.
Very interesting. I wish I understood these issues better, and I'm glad that there are others here to help explain!!! Thanks -- Paul
But the leakage current problems have been increasing with process shrinks (not just at Intel, but also at IBM and AMD). So they can use even smaller lithography. Great. Will the leakage current and associated heat suck even worse than Prescott?
Shostak assumed that computer processing power will continue to double every 18 months until 2015- as it has done for the past 40 years. From then on, he assumes a more conservative doubling time of 36 months as transistors get too small to scale down as easily as they have till now.
As has been seen with 65nm Prescotts and difficulties with die shrinks to this size by IBM and AMD as well, we may already be at or near the point that they had assumed for 2015.
The comment about it being Warren Buffet, rather than Warren Beatty, is right. Warren Buffet is often referred to as the "Oracle of Omaha" for his success in understanding and predicting the financial / economic climate.
There's actually a neat story about Warren Buffet when he was young. There was a man whose young neighbor once approached him with a (set of) wild investment idea and asked him if he wanted to go in on it with him. The guy figured the neighbor for either a con artist or a hack and turned him down. Well, decades later, that young man is a very successful Warren Buffet, and I'll bet the other is kicking himself for turning down that once-in-a-lifetime opportunity.
Sorry, I looked all over for a link to that anecdote, but I couldn't find it among the sea of "young Warren Buffet neighbor" google hits. -- Paul
On the one hand, that's the inherent risk with any technology as it becomes increasingly accessible and "user-friendly".
On the other hand, are these systems going to be cheap enough that we have to worry about script kiddies? If computers still cost $5000+, I doubt script kiddies would be such a rampant problem on the net. -- Paul
That was my fist question, too.
My thought is, if the hijacked machine is chocking on all these calculations, at least they'll notice that there's something wrong with their machines. (Which would be an advance in and of itself.)
The point you are trying to emphasize is a good one, but as I mentioned before, the key is testing. I've tested the code on this machine for problems that have analytical solutions, and the results agreed with the expected order of accuracy. The results also matched those obtained with non-overclocked machines.
The fact of the matter is, most machines aren't actually running precisely at spec. Do you really think that your machine is running with a precise FSB of 200.00 MHz (or whatever the spec FSB is for your machine)? It's not uncommon to see 1-5% variation on machines running "at spec".
I'm not shooting for the moon here. I'm applying a moderate overclock that's well within the range expected for this architecture (officially up to 3.6 GHz for the Northwood core), and I'm exercising due caution by conducting extensive testing.
That just isn't so.
The key to a successful overclock is to test, test, and test until you've found the top end of the stable range, and then back down a bit. An overclock that isn't stable isn't a successful overclock.
I overclock a P4 3.0C to 3.5 GHz and use it for long-term computational biophysics. (Numerical simulation of tumor growth and chemotherapy.) It's amazing what a difference a 17% speed boost can make in a 7-day calculation. I've tested both the memory and the CPU for stability and can vouch for it. Overclocking isn't just for games; it can make the hardware you can afford more productive for real work. It sure helped me to get my research done more quickly for my master's thesis.
Other people are using overclocked computers for useful things all the time. Consider the thousands who use overclocked computers to pump out WU's for Stanford's protein folding project (find out how proteins fold and misfold, help cure diseases, etc.). The top-producing teams are overclockers.au, HardOCP (lots of overclockers), and overclockers.com at Number 3.
I used to use TeraTermPro / TTSSH as well. It was very nice, but alas, TTSSH only has SSH 1.5 and most likely won't be updated to SSH protocol v. 2.0. AFAIK, That means that you won't get the most recent security fixes, as well as other nice features of SSH v. 2.0 (like compression).
A good alternative is PuTTY. Works like a charm in all flavors of Win32.
Computational Fluid Mechanics.
Their main problem seems to be summarized as such: How can we get rid of mesh nodes while still accurately representing the surface?
A related approach is: Given that we have N node points available, how can we distribute them among the surface to provide the most accurate representation? If we can add a few nodes, where would it be best to add them? If the mesh is good enough, where would it be best to remove them?
Then, they explain their approach a bit more:
Desbrun explains that his accomplishment was to simplify such a mesh, by combining as many of the little triangles as possible into larger elements without compromising the actual shape. Nearly flat regions are efficiently represented by one large, flat mesh element while curved regions require more mesh elements.
My colleagues, who do advanced computational fluid mechanics (applied to drop-drop collisions with surfactant diffusion, simulation of tumor growth, biomathematics, etc.) do similar work, but in a very different way. For a given surface mesh, they assign a sort of mesh energy to the whole structure, where the energy of any point depends upon the local curvature of the surface, etc., and they add a constraint or two. (I don't remember the full details.) It's very similar to a network of springs. Then, they apply standard optimization techniques to minimize the mesh energy. The result: a mesh that has the highest density where the surface has the highest curvature, and the lowest density where the surface is flattest. The method is fully adaptive and even works during and after surface topological changes, which is very useful in CFM, where your surfaces can break, merge, etc.
So, this idea of modeling surfaces by concentrating mesh nodes where the surface curvature is greatest with an automated, adaptive algorithm is not new, and it is very applicable in making more efficient use of limited computational resources.
"And now, your highness, we will discuss the location of your hidden rebel base..."
Poor Nash is either going to find a mistake in this guy's work or go insane trying to ...
Ah, heck, here goes. :)
Windows:For working:
- WinSCP: free secure ftp client to move work around
- PuTTY: free SSH client as mentioned above
- MinGW & MSYS: free C++ compiler and environment, with useful things such as grep
- Matlab: student version of linear algebra sofware. Mostly used for visualization these days
- Miktex: fantastic distro of LaTeX for Windows; for writing homework and scientific papers
For additional productivity:- MS Powertoys / TweakUI: free download to specify documents folders, etc., and tweak other behavior; available at MS's site. Also "command prompt here" powertoy, etc.
- Google Toolbar: iexplorer is good enough for me (might as well use i0t, since it's already in system memory) when coupled with the google bar and popup blocking.
- Norton AntiVirus: not dumb enough to open attachments, but it's good to scan downloads before unzipping
- stanford folding project: great for testing stability of the overclock while contributing to a good cause
For fun:- Flavor of the Month Game Demo: I'm too cheap to buy most games, but I'm getting a lot of fun out of the Halo and FarCry demos at the moment.
Linux (gentoo build from stage 1):For working:
- X: because X tunneling is nice
- OpenBox: I like their no-nonsense window manager
- xv: great for quick viewing of graphics files
- Intel's C++ Compiler: free C++ compiler in linux, and beats the pants off of G++ for fast-running code
- Matlab: student version of linear algebra sofware. Mostly used for visualization these days
- LaTeX: for writing homework and scientific papers
For additional productivity:- Firefox: You all know this one.
- nvidia drivers: gives better performance in matlab
- stanford folding project: great for testing stability of the overclock while contributing to a good cause
For fun:Yes, except that if any one of your 2000 members abuses the system, you'll be the one in trouble for it. (Since personal.mail is registered to you.) And then you'll be facing the money-back demands of the other 1999 members who suddenly lost their service. -- Paul
Good point there. Actually, as I thought about it more, I realized that non-volatile RAM would probably never be as fast as "regular" RAM.
However, what would probably be a redeeming "in-between" solution would be a cheap, non-volatile solid state disk. Non-volatile RAM that's cheap enough and reliable enough to use a really fast hard drive. Not to store all data on, but big enough to store the OS on. One could always have a larger, traditional slow hard drive for data and a smaller, faster hard drive for the OS. Some people already do this with compact flash, etc., but those media aren't exactly made to be read / written a huge number of times like a partition with an OS needs to be ...
Great points! -- Paul
Won't this all be moot after non-volatile RAM (that's as fast as current RAM) is developed and widespread? AFAIK, fast non-volatile RAM technology is well under way and within reach in the next several years. The only reason we currently need to read from the hdd is that the RAM is volatile and loses its contents on a power down.
We can already see how this would work in an S3 suspend: the entire PC but the memory (and perhaps the CPU) is powered down. Press the keyboard, and you're back in your OS in seconds. Replace present-day memory with non-volatile memory, and you're completely there. -- Paul
Except that W. James McNerney, Jr., a product of the GE management machine, became the Chairman of the Board and CEO of 3M in recent years.
I remember the chills that went through the research department when that happened (I was an intern in fiber optics at the time), and indeed, the writing was on the wall for imminent changes. The traditional "15% rule", whereby researchers could spend 15% of their time on their own research ideas which often lead to very innovative, lucrative new products (e.g., Post-It notes, UV-hardened polymer), was in jeopardy when I left. I'm not sure where that went since then, as I'm not privy to the 3M internals.
Here's an interesting link on 3M research culture, BTW. -- Paul
I remember being really gung ho about Linux and Open Source after trying my first distro: Mandrake 8.0. At that time, Win98SE and WinME were the dominant flavors (WinXP was just starting to come out), and I found that the Mandrake install did a better job of detecting most of my hardware than the MS install.
Eager to support the cause, I plucked down $100 to preorder the Pro version of the upcoming Mandrake distro. "Cool, I'm supporting open source. I'm doing my part," I thought, and I'd even get some of the CD's early for my pre-order. So I ordered, my credit card was charged, and day after day, week after week, no product arrived. And day after day, my emails to the company weren't answered. There were no real announcements anywhere to be seen about what was causing the delay. Finally, after a bit more than a month of this, I finally called the company at my own expense and had my order cancelled. (And even that required quite a run around, as the number listed on Mandrake's site didn't seem to be a direct number, so I had to call a few times to connect with anybody.)
And this is how they treated an eager customer. Hardly the way to treat a paying customer! I sure wouldn't want to run my business this way.
Granted, things are better now, but when your business isn't run like a business, don't expect customers to stick around. -- Paul
Hmm, it looks "cute". I'd imagine that the "form before function" people will buy this, female or male. It'll look great in that new VW bug along with the Michael Graves teapot.
That's a good strategy. An efficient heatsink with a quiet fan can go a long ways. I installed a Thermalright SP-94 (all-copper with heatpipes) with a 92mm Panaflo M running at 8V. It keeps my P4 3.0C (overclocked to 3.5 GHz) pretty cool and pretty quiet. My GF Ti4200 is passively cooled with a Zalman heatsink. My current power supply is pretty quiet, but I fixed my old one up by removing the airflow-restricting punched vents with a Dremel and replacing the cheapo loud fan with a quiet Panaflo L. Also added a direct intake of airflow right about the CPU / HSF in the side of a case with an 80mm Panaflo L running at 7V.
It's amazing how quiet this system is. It's even quieter than my previous PIII system was at half the speed. Up until recently, it was much quieter than my wife's HP Celeron 800 system. (We just finished modding that together to quiet it down...)
Really, if you're willing to put in a little bit of elbow grease and research here and there, you can wind up with a high-performance system that's quieter than most OEM systems.
Oh, as for hard drive noise that was mentioned above, the trick to try is to mechanically isolate the hard drive from the case, such as by suspension. See www.silentpcreview.com and www.overclockers.com for some ideas. :) -- Paul