There are interesting psychological and ergonomic factors that play a role here. Humans have evolved to optimally perceive and respond to visual stimuli that are their own size, give or take an order of magnitude. A normal computer display is below the middle of this range, and most objects displayed on them are at or even below the size range to which we optimally respond. I expect that blowing up the szie of a display to make displayed objects be our own size more or less will have vast and to some extent unforseen positive consequences on our ability to perceive, digest and react to information.
Also, blowing up display size such that it extends beyond our natural field of vision fully exploits our ability to process visual information, as opposed to staring down a virtual tunnel at an undersize screen. Just ask someone who wears very strong prescription glasses how much they would give to get rid of the heavy frame limiting their field of vision.
We sort of take it for granted that displays are small, be it paper, CRTs or flat screens. But this is an artificial limitation that does not need to be perpetuated. When was the last time you stared at a fixed point out in nature for more than a few minutes because this was where all the action was?
Happens very rarely, right?
I am all for fully utilizing our senses when it comes to dealing with vast quantities of data.
I beg to differ: Cycling exercises your lower back muscles extensively. It was actually recommended to me by my doctor when he realized that I was a prime target for lower back trouble.
The only caveat is that you should be careful when you have a bout of lower back pain. Other than that, cycling is highly recommended.
Well, we aren't a 4-year school and we have an edu domain (scripps.edu). However, we got our domain way before Network Solutions took over administration of domains. Then again, we grant PhD degrees which take more than 4 years usually.
But I am sure glad that the edu domain is not in the hands of a commercial entity any longer. Let's hope that the rules for getting an edu domain will be relaxed to allow any accredited degree granting institution.
Even though IT is currently a booming sector and every jerk can get in somehow, there always has been and always will be a glass ceiling for many of those who don't have proper degrees and/or credentials. You don't want to lock yourself in by skimping on education.
Myself, I am glad for every one of those 9 years I spent getting my degrees. I don't use very much of what I have learned anymore, but the degrees certainly open doors that remain locked otherwise, and the pay scale is different, too.;-)
I have extremely bright and talented friends who found that they had to go back to school and work on those darn degrees in order to advance further in their career. They were bypassed by stupid, but degreed people on a very regular basis. In hindsight, the time spent in school neither detracted much from their professional life (consulting), nor was it much of a burden, but rather it was worth every penny and minute.
In my experience, the total of your experience only starts to weigh more than your degrees once you are past 35-40 or so. It might be different in some IT sectors right now, but what will it look like five years down the line? Do yourself a favor and go to school, and use that time well. You can still work part time on the side and if you are smart, you make a killing even so.
Yes, there's a decent market for these machines. Given SGI's situation, however (they've restructured every quarter for the past 2 years) and the fact that the (non-embedded) MIPS processor line is a few generations behind similar offerings from IBM and Sun, I've a feeling that many customers with just-fat-enough wallets will take a wait and see on these machines, or just look at similar offerings from more stable companies.
I don't know about that. We benchmarked a handful of our regularly used programs - mostly molecular dynamics and quantum chemistry stuff - and SGI's 3000 system looked just as good as the competition if not better. In fact, SUN was so bad on floating point performance that we didn't bother to run the full complement of benchmarks. (That'll change late this year, but we needed systems now.)
Anyway, SGI came out on top when raw CPU performance, system scalability using our codes and I/O were considered together. So, we are going to receive a 3800 as soon as SGI can deliver one. Can't wait!
As far as company stability goes, no customer is going to buy a truly large machine without a lot of legalese in the contract that spells out what happens if things go wrong. Company failure is usually factored in. Me, I am not concerned. SGI has too much good technology, and also too much cash in the bank to simply disappear from the scene. They could be snapped up by someone else, perhaps, but would that change much? SUN did not make any major changes to the E10000 when they got it from Cray in '96, did they?
I think you are missing the real point. We are after the basic blueprint of humankind here, and by extension (we have a few dozen other genomes already done) for most of life.
What new insights this will trigger noone knows, but it is clear that having this basic but thorough knowledge is far better than a patch here and a patch there.
After WWII it was decided that we needed to know far more about the makeup of living organisms than we knew then. I forget who the players were, but the decision was to take a simple bacterium, Escherichia coli, and find out all about it that was doable with the methods available then. This was the first truly large scale biological research effort and it ammassed a ton of data. As a result we have an extremely well understood lab organism which enabled us to revolutionize genetics.
Cloning and sequencing of genes, the whole biotech industry, most of today's biological research, in fact, and many of today's medical procedures simply whouldn't be possible without that pioneering effort.
Back then you could have questioned the effort and noone would have been able to point out what would eventually come of it. That's very the nature of basic scientific research. Still, scociety as a whole places a lot of trust and hope in research, or we wouldn't see the level of funding that we have. Given past breakthroughs resulting from broad, basic efforts, there's every hope that the huiman genome project will be a huge win.
What we'll do when the human genome is completely mapped is being discussed almost daily among scientists. The post-genome era has become a big buzzword.
One very convincing idea goes like this: - you have a problem that you'd like to tackle - analyze sequence data in lioght of your problem - filter out interesting trends/data points - develop a high-throughput assay to test for what the sequence data implies - analyze test data
In other words, start on the computer, end on the computer, work in the lab in between. Sort of like what we do with literature already. You can, of course, compare the genome and related sequence data to literature anyway. It simply has be be read and understood. (No that we know a lot about the latter activity, but that's another story.)
At a recent event I attended, an intersting example was given by Dr. Wei Hu, formerly of Human Genome Sciences, Inc.: They were interested in prostate cancer and therefore looked at ESTs (expressed sequence tags) from tissue samples of various stages of prostate cancer, as well as several other unrelated tissue samples as controls. The analysis simply consisted of looking for sequence tags that consistently turn up in prostate cancer, but not elsewhere. Half a dozen or so sequences were found and most proved to be known markers for prostate tissue, especially cancerous prostate tissue, but one or two were new. This all was only a few hour's work.
Further research might then entail chasing these new markers, perhaps developing a simple and cheap assay for them, and voila, a new test for early stage prostate cancer. In practice this is of course not nearly as easy as it sounds, but you get the idea.
With the complete human genome available, one would of course compare the sequence tags against the genome to find where they are, with what other regions they are asociated, what gene they come from, etc. This would dramatically increase the information content of the simple exeperiment that was done and described.
The upshot IMHO is that biologists will dig much less in the dark, at least as far as sequence information goes. Checking the genome and other sequence databases will be just as mandatory and routine as a trip to the library is today. This in turn means that biologist will have to become much more computer-savvy, or that biologists and computer geeks need to develop closer ties.
One thing to keep in mind, though, is that the upcoming announcement is only for the mapping of the human genome, i.e. known markers will be placed along the genome in more or less regular intervals. This amounts to a lowres image plus many (most?) parts of a highres one. The actual full genome sequence is still a ways off as gaps need to be closed in difficult regions and other boring cleanup work needs to be done.
Well, then, what about all those images depicting women in lingerie, where the lingerie has moved from the parts it usually covers. This suitably disrupts the continuous blobs of 'skin tone' while fully (or at least partly) exposing strategic body parts, thus passing the blocking software. But this sort of picture content represents the vast majority of porn out there, since lingerie is thought to spice up the image.
Re:Can someone please explain: Why?
on
Pilot Synthesis
·
· Score: 2
It's not so much the MIDI synth, but more the combination of MIDI synth and sequencer software. That essentially lets you compose and arrange songs wherever you may be.
Myself, I am much more creative when not in front of my computer or even hunched over the piano. Hence, my Yamaha QY-10 (2x the size of the SG20 being discussed here) has been invaluable. Think composing while in the hammock in the garden, or in bed, or out in natute somewhere. Or playing with musical ideas on an airplane. The possibilities are really endless. The smaller, the better. Longer battery life helps, too.
Nice, but limited polyphony (24)
on
Pilot Synthesis
·
· Score: 1
This is certainly a great little product and should give other small MIDI sequencers such as the Yamaha QY series or the Boss Dr. Rythm series a run for their money.
However, the SG20 can only play 24 voices simultaneously. That is seriously lacking and well behind the competition. Think about it: Drum track, bass, snare, high hat; bass track, 2 notes; synth track, 5 notes, strings, 5 notes; guitar, 6 notes; lead voice, 3 notes. Not an extreme situation in any given MIDI file, but it will exhaust this unit already.
I have seen plenty of MIDI files (and have myself created some) that severely tax my software synth due to the large quantity of simultaneous notes (32-48). The SG20 could never handle those and would sound awful, because any new note immediately silences an old one that should still be playing. That sounds worse than a bad mp3.
Surgery based on this technology would be a bad idea! Keep in mind that the human body changes over time. This is living tissue, not a machine! Most eyes go from normal or nearsighted (whichever the case may be) in their youth to (slightly) farsighted in their middle age and get worse from there.
I expect that the slight local aberrations which this adaptive optics technology measures and corrects change even more over time. That would make surgical correction a bad move, as the correction would develop into more aberrations over time.
Also, current LASIK and other laser surgery techniques are rather crude and can leave you with less than perfect vision. Furthermore, they are known to introduce glare, halos and other gost images of things with very high contrast. i.e. the quality of local visual perfection actually goes down, especially in the periphery. You'd most likely need more adaptive optics after LASIK than before.
Laser surgery produces scar tissue in an otherwise perfectly clear tissue which had a lot of clean, local structure (neat hexagonal patches, for example). I just can't see why healed, scarred tissue should be superior to what grows naturally, even if imperfectly.
Finally, adaptive optics improve vision especially in low light situations. LASIK is known to make your eyes worse under these same conditions. Doesn't sound like a good match to me.
Frankly, I prefer an external device that can be periodically retuned to perfectly (or as closely as can be, at least) match the current state of my eyes.
Simple: It depends on the computation/communication ratio. As Greg pointed out, many problems have higher communication needs than a distributed approach can satisfy, or in other words, you don't want to compute 2 seconds and then wait 10 seconds to have your results acknowledged and integrated by the server.
Parallel computations can be classified into intrinsically (embarassingly) parallel, highly parallelizable and hard to parallelize. The first class may lend itself to distibuted.net-type approaches (if client size - which is dictated by the problem/algorithm - isn't going to discourage you, that is). The second class is what we are talking about here and typically does not lend itself to distributed.net-type approaches. The third class remains in the realm of vector supercomputers, mostly.
You're right. these have been around for a very long time. In the traditional Unix world they have been used as database accelerators and such.
It is nice to see this technology trickle down into the PC world, though. Now we just have to wait for PCI's successor to see some real throughput improvements.
Interesting article. However, the lapping advice really sucks!
You can't lap the heat sink mounting plate much flatter with sandpaper on your granite countertop (or any old glass plate for that matter). As a woodworker who tunes his handtools, I can tell you that lapping requires a VERY flat surface. These can be had relatively cheap, but you have to go out and look for them. I got mine from Lee Valley Tools. Then mount the sandpaper very carefully. It's best to use plastic backed abrasives, rather than cloth or paper backed stuff. Or simply use loose grit.
You also have to hold the tool to be lapped at a very consistent angle and apply consistent pressure/force over the entire surface to be lapped and over the entire lapping stroke. No rocking, no twisting, just smooth moves.
If you use just any hard surface that appears flat to you and don't practice your lapping technique, you might end up with a severely dished surface!
Re:Great, now when can I get Slate?
on
New Mega Alphas
·
· Score: 1
I have seen and touched a Slate at a customer site in San Diego two months ago. Running Linux, no less. Persumably these people now have a rack full of them. Officially, Slates have been on the market for a month or so. I am not surprised to see that people are still waiting for them, though. I suspect that high-profile customers got first dibs.
Re:Just got back from the Atlanta rollout thingie
on
New Mega Alphas
·
· Score: 1
Compaq can provide better info than this, but a rollout is probably the wrong time for this. A short while ago we had one of their high performance computing / enterprise computing managers out here, and we got a lot of meat. I am not sure to what extent I am still bound by the NDA, but let me assure you that the new GS boxes are very competitive with Sun's E6500 and E10000, as well as SGI Origin2000. A lot of good design decisions went into these servers, and they should shine in both, the commercial database market as well as in the technical computing arena.
What's cool about these is they've got a crossbar switched architecture, so they scale better than a bus or omega switched network. [snip]... these are sweet boxen and should deliver truly obscene tpc-c and tpc-h results.
I agree completely. These servers should provide some healthy competition for SUN and SGI (and others) in the high-end server market. From all the specs I saw under NDA, these are very competitive designs with all the right things in place.
Now drop a bunch of these into a SC-class framework (Compaq's high speed, low latency clustering solution) and you have a killer supercomputer.
The Alpha is not limited to 2-way SMP. The Alpha can also be put into much larger SMP configurations. The AlphaServer ES40 is 4-way, the GS60 is 8-way, and the GS140 can have up to 14 Alpha processors in it. There is even a new SC that can have 64-512 processors! These are SMP machines, not Beowulf clusters.
Correction: The Compaq SC systems are not SMP boxes. They are ES40s clustered via high-speed, low-latency Quadrics interconnects. Nice systems, but a few quirks.
Compaq is working on very large SMP systems. Ask your sales rep about details.
Doubletwist is sort of a spinoff/evolution of Pangea Systems, Inc. and has been formed to be a ASP for genomic science. Pangea wasn't such an unknown company in the biotech field. What Doubletwist desperately needed right now was a campaign to increase their name recognition. I guess they now got that...
Whether the announced annotations to the known genome sequence data are really worth the hoopla will be known in a few days or weeks when genomics scientists have had a chance to look it over. In the meantime, relax! There will be lots more such announcements in this hotly contested field. It's just like chip wars.
The real question is this: If the same money were spent on, say, Athlon nodes connected with channel bonded fast ethernet (or even myrinet); could you get even more performance? I figure that you could build a cluster of stripped down Athlon-700's on channel bonded ether for around $2k per node including switches, etc. That would allow up to 7500 nodes (though I imagine that network bandwidth/latency would kill your performance at that scale). Hmmm...
You got it all wrong. What drives such a purchase is: How many CPUs can my app use advantageously? If the app scales to a few hundred CPUs, but not to thousands, there is absolutely no point to build a equivalent (in terms of theoretical Teraflops) system from thousands of cheaper CPUs. You absolutely want a few hundred of the fastest possible chips.
As to interconnects: For many MPI-based codes you don't want to go over ethernet or any other TCP/IP-based interconnect. You want low level protocols over low latency/high bandwidth interconnects. TCP/IP really sucks when it comes to latency, no matter how many channels you bond together.
You would need a benchmarking standard, not just a standard benchmark. And that standard needs to be protected by a license or something to ensure that everyone plays by the rules.
The benchmarking standard should state that code changes are not allowed, and it should detail how the benchmark was to be run, how it was to be reported and where.
We already have standard benchmarks and benchmarking standards with SPEC. It's just not really open source. But as another poster detailed here, a SPEC equivalent should be pretty straightforward to implement as open source. Just who or what mechanism would make sure that noone cheats is another issue, though.
That is why open source benchmarks are a good idea -- not only does it allow people to improve on the code directly, but it lets people see exactly what is going on behind the scenes.
That is a very bad idea, indeed. If the code base of the benchmark changes at all, none of the numbers are comparable between releases. This is exactly why tightly controlled benchmarks like SPEC have been successful. SPEC only changes every few years, there are clear rules of what you can do and what not while compiling and running the benchmarks and there are rules about how to report the resulting numbers.
Inasfar as one can trust generic benchmarks, SPEC has held up nicely and allowed us to superficially compare systems from Unix vendors with different CPUs, different architecture and different OS. Even with infrequent updates, the transition from one version of SPEC to the next gets in the way sometimes. I can only imagine how bad a true open source solution without additional rules would be.
I know, SPEC isn't open source in the strict sense, but it IS a broadly accepted benchmark suite of which source is available, and it has served us well for the past 11 years.
As well as generic benchmark can serve anyway. There is of course no substitute to check out a box with your own apps and workload.
Apparently, a lot of people here like to shout "beowulf cluster" and don't understand much about the problems of massively parallel computing.
The bottleneck is inter-processor communication. If all you are doing is trying to brute force a cipher the processors are almost independent and easily reach the theoretical aggregate performance figures. But if you are doing complex physical simulations you can end up waiting for data most of the time and using only a fraction of your theoretical parallel power. Well spoken!
We've just been through an evaluation cycle where we pitted all major HPC systems and most Intel/Linux clustering solutions against each other. While I can't post numbers (yet), the results are that Intel/linux clusters come out at the bottom, mostly because of high latencies in the interconnect. I think it is not only the hardware per se, but immature drivers also.
We looked at Myrinet, Giganet and SCI. All of them were pretty close in performance, but don't come near more established SMP, NUMA or clustering technologies under traditional Unices. Our benchmark is molecular dynamics code with a reasonably large solvated protein.
Slashdot Mirror
There are interesting psychological and ergonomic factors that play a role here. Humans have evolved to optimally perceive and respond to visual stimuli that are their own size, give or take an order of magnitude. A normal computer display is below the middle of this range, and most objects displayed on them are at or even below the size range to which we optimally respond. I expect that blowing up the szie of a display to make displayed objects be our own size more or less will have vast and to some extent unforseen positive consequences on our ability to perceive, digest and react to information.
Also, blowing up display size such that it extends beyond our natural field of vision fully exploits our ability to process visual information, as opposed to staring down a virtual tunnel at an undersize screen. Just ask someone who wears very strong prescription glasses how much they would give to get rid of the heavy frame limiting their field of vision.
We sort of take it for granted that displays are small, be it paper, CRTs or flat screens. But this is an artificial limitation that does not need to be perpetuated. When was the last time you stared at a fixed point out in nature for more than a few minutes because this was where all the action was?
Happens very rarely, right?
I am all for fully utilizing our senses when it comes to dealing with vast quantities of data.
I beg to differ: Cycling exercises your lower back muscles extensively. It was actually recommended to me by my doctor when he realized that I was a prime target for lower back trouble.
The only caveat is that you should be careful when you have a bout of lower back pain. Other than that, cycling is highly recommended.
Well, we aren't a 4-year school and we have an edu domain (scripps.edu). However, we got our domain way before Network Solutions took over administration of domains. Then again, we grant PhD degrees which take more than 4 years usually.
But I am sure glad that the edu domain is not in the hands of a commercial entity any longer. Let's hope that the rules for getting an edu domain will be relaxed to allow any accredited degree granting institution.
Even though IT is currently a booming sector and every jerk can get in somehow, there always has been and always will be a glass ceiling for many of those who don't have proper degrees and/or credentials. You don't want to lock yourself in by skimping on education.
;-)
Myself, I am glad for every one of those 9 years I spent getting my degrees. I don't use very much of what I have learned anymore, but the degrees certainly open doors that remain locked otherwise, and the pay scale is different, too.
I have extremely bright and talented friends who found that they had to go back to school and work on those darn degrees in order to advance further in their career. They were bypassed by stupid, but degreed people on a very regular basis. In hindsight, the time spent in school neither detracted much from their professional life (consulting), nor was it much of a burden, but rather it was worth every penny and minute.
In my experience, the total of your experience only starts to weigh more than your degrees once you are past 35-40 or so. It might be different in some IT sectors right now, but what will it look like five years down the line? Do yourself a favor and go to school, and use that time well. You can still work part time on the side and if you are smart, you make a killing even so.
Yes, there's a decent market for these machines. Given SGI's situation, however (they've restructured every quarter for the past 2 years) and the fact that the (non-embedded) MIPS processor line is a few generations behind similar offerings from IBM and Sun, I've a feeling that many customers with just-fat-enough wallets will take a wait and see on these machines, or just look at similar offerings from more stable companies.
I don't know about that. We benchmarked a handful of our regularly used programs - mostly molecular dynamics and quantum chemistry stuff - and SGI's 3000 system looked just as good as the competition if not better. In fact, SUN was so bad on floating point performance that we didn't bother to run the full complement of benchmarks. (That'll change late this year, but we needed systems now.)
Anyway, SGI came out on top when raw CPU performance, system scalability using our codes and I/O were considered together. So, we are going to receive a 3800 as soon as SGI can deliver one. Can't wait!
As far as company stability goes, no customer is going to buy a truly large machine without a lot of legalese in the contract that spells out what happens if things go wrong. Company failure is usually factored in. Me, I am not concerned. SGI has too much good technology, and also too much cash in the bank to simply disappear from the scene. They could be snapped up by someone else, perhaps, but would that change much? SUN did not make any major changes to the E10000 when they got it from Cray in '96, did they?
I think you are missing the real point. We are after the basic blueprint of humankind here, and by extension (we have a few dozen other genomes already done) for most of life.
What new insights this will trigger noone knows, but it is clear that having this basic but thorough knowledge is far better than a patch here and a patch there.
After WWII it was decided that we needed to know far more about the makeup of living organisms than we knew then. I forget who the players were, but the decision was to take a simple bacterium, Escherichia coli, and find out all about it that was doable with the methods available then. This was the first truly large scale biological research effort and it ammassed a ton of data. As a result we have an extremely well understood lab organism which enabled us to revolutionize genetics.
Cloning and sequencing of genes, the whole biotech industry, most of today's biological research, in fact, and many of today's medical procedures simply whouldn't be possible without that pioneering effort.
Back then you could have questioned the effort and noone would have been able to point out what would eventually come of it. That's very the nature of basic scientific research. Still, scociety as a whole places a lot of trust and hope in research, or we wouldn't see the level of funding that we have. Given past breakthroughs resulting from broad, basic efforts, there's every hope that the huiman genome project will be a huge win.
What we'll do when the human genome is completely mapped is being discussed almost daily among scientists. The post-genome era has become a big buzzword.
One very convincing idea goes like this:
- you have a problem that you'd like to tackle
- analyze sequence data in lioght of your problem
- filter out interesting trends/data points
- develop a high-throughput assay to test for what the sequence data implies
- analyze test data
In other words, start on the computer, end on the computer, work in the lab in between. Sort of like what we do with literature already. You can, of course, compare the genome and related sequence data to literature anyway. It simply has be be read and understood. (No that we know a lot about the latter activity, but that's another story.)
At a recent event I attended, an intersting example was given by Dr. Wei Hu, formerly of Human Genome Sciences, Inc.:
They were interested in prostate cancer and therefore looked at ESTs (expressed sequence tags) from tissue samples of various stages of prostate cancer, as well as several other unrelated tissue samples as controls. The analysis simply consisted of looking for sequence tags that consistently turn up in prostate cancer, but not elsewhere. Half a dozen or so sequences were found and most proved to be known markers for prostate tissue, especially cancerous prostate tissue, but one or two were new. This all was only a few hour's work.
Further research might then entail chasing these new markers, perhaps developing a simple and cheap assay for them, and voila, a new test for early stage prostate cancer. In practice this is of course not nearly as easy as it sounds, but you get the idea.
With the complete human genome available, one would of course compare the sequence tags against the genome to find where they are, with what other regions they are asociated, what gene they come from, etc. This would dramatically increase the information content of the simple exeperiment that was done and described.
The upshot IMHO is that biologists will dig much less in the dark, at least as far as sequence information goes. Checking the genome and other sequence databases will be just as mandatory and routine as a trip to the library is today. This in turn means that biologist will have to become much more computer-savvy, or that biologists and computer geeks need to develop closer ties.
One thing to keep in mind, though, is that the upcoming announcement is only for the mapping of the human genome, i.e. known markers will be placed along the genome in more or less regular intervals. This amounts to a lowres image plus many (most?) parts of a highres one. The actual full genome sequence is still a ways off as gaps need to be closed in difficult regions and other boring cleanup work needs to be done.
Well, then, what about all those images depicting women in lingerie, where the lingerie has moved from the parts it usually covers. This suitably disrupts the continuous blobs of 'skin tone' while fully (or at least partly) exposing strategic body parts, thus passing the blocking software. But this sort of picture content represents the vast majority of porn out there, since lingerie is thought to spice up the image.
It's not so much the MIDI synth, but more the combination of MIDI synth and sequencer software. That essentially lets you compose and arrange songs wherever you may be.
Myself, I am much more creative when not in front of my computer or even hunched over the piano. Hence, my Yamaha QY-10 (2x the size of the SG20 being discussed here) has been invaluable. Think composing while in the hammock in the garden, or in bed, or out in natute somewhere. Or playing with musical ideas on an airplane. The possibilities are really endless. The smaller, the better. Longer battery life helps, too.
This is certainly a great little product and should give other small MIDI sequencers such as the Yamaha QY series or the Boss Dr. Rythm series a run for their money.
However, the SG20 can only play 24 voices simultaneously. That is seriously lacking and well behind the competition. Think about it: Drum track, bass, snare, high hat; bass track, 2 notes; synth track, 5 notes, strings, 5 notes; guitar, 6 notes; lead voice, 3 notes. Not an extreme situation in any given MIDI file, but it will exhaust this unit already.
I have seen plenty of MIDI files (and have myself created some) that severely tax my software synth due to the large quantity of simultaneous notes (32-48). The SG20 could never handle those and would sound awful, because any new note immediately silences an old one that should still be playing. That sounds worse than a bad mp3.
Surgery based on this technology would be a bad idea! Keep in mind that the human body changes over time. This is living tissue, not a machine! Most eyes go from normal or nearsighted (whichever the case may be) in their youth to (slightly) farsighted in their middle age and get worse from there.
I expect that the slight local aberrations which this adaptive optics technology measures and corrects change even more over time. That would make surgical correction a bad move, as the correction would develop into more aberrations over time.
Also, current LASIK and other laser surgery techniques are rather crude and can leave you with less than perfect vision. Furthermore, they are known to introduce glare, halos and other gost images of things with very high contrast. i.e. the quality of local visual perfection actually goes down, especially in the periphery. You'd most likely need more adaptive optics after LASIK than before.
Laser surgery produces scar tissue in an otherwise perfectly clear tissue which had a lot of clean, local structure (neat hexagonal patches, for example). I just can't see why healed, scarred tissue should be superior to what grows naturally, even if imperfectly.
Finally, adaptive optics improve vision especially in low light situations. LASIK is known to make your eyes worse under these same conditions. Doesn't sound like a good match to me.
Frankly, I prefer an external device that can be periodically retuned to perfectly (or as closely as can be, at least) match the current state of my eyes.
Check the I Know Why Refractive Surgeons Wear Glasses site for more details on laser eye surgery.
Simple: It depends on the computation/communication ratio. As Greg pointed out, many problems have higher communication needs than a distributed approach can satisfy, or in other words, you don't want to compute 2 seconds and then wait 10 seconds to have your results acknowledged and integrated by the server.
Parallel computations can be classified into intrinsically (embarassingly) parallel, highly parallelizable and hard to parallelize. The first class may lend itself to distibuted.net-type approaches (if client size - which is dictated by the problem/algorithm - isn't going to discourage you, that is). The second class is what we are talking about here and typically does not lend itself to distributed.net-type approaches. The third class remains in the realm of vector supercomputers, mostly.
You're right. these have been around for a very long time. In the traditional Unix world they have been used as database accelerators and such.
It is nice to see this technology trickle down into the PC world, though. Now we just have to wait for PCI's successor to see some real throughput improvements.
Interesting article. However, the lapping advice really sucks!
You can't lap the heat sink mounting plate much flatter with sandpaper on your granite countertop (or any old glass plate for that matter). As a woodworker who tunes his handtools, I can tell you that lapping requires a VERY flat surface. These can be had relatively cheap, but you have to go out and look for them. I got mine from Lee Valley Tools. Then mount the sandpaper very carefully. It's best to use plastic backed abrasives, rather than cloth or paper backed stuff. Or simply use loose grit.
You also have to hold the tool to be lapped at a very consistent angle and apply consistent pressure/force over the entire surface to be lapped and over the entire lapping stroke. No rocking, no twisting, just smooth moves.
If you use just any hard surface that appears flat to you and don't practice your lapping technique, you might end up with a severely dished surface!
I have seen and touched a Slate at a customer site in San Diego two months ago. Running Linux, no less. Persumably these people now have a rack full of them. Officially, Slates have been on the market for a month or so. I am not surprised to see that people are still waiting for them, though. I suspect that high-profile customers got first dibs.
Compaq can provide better info than this, but a rollout is probably the wrong time for this. A short while ago we had one of their high performance computing / enterprise computing managers out here, and we got a lot of meat. I am not sure to what extent I am still bound by the NDA, but let me assure you that the new GS boxes are very competitive with Sun's E6500 and E10000, as well as SGI Origin2000. A lot of good design decisions went into these servers, and they should shine in both, the commercial database market as well as in the technical computing arena.
What's cool about these is they've got a crossbar switched architecture, so they scale better than a bus or omega switched network. [snip] ... these are sweet boxen and should deliver truly obscene tpc-c and tpc-h results.
I agree completely. These servers should provide some healthy competition for SUN and SGI (and others) in the high-end server market. From all the specs I saw under NDA, these are very competitive designs with all the right things in place.
Now drop a bunch of these into a SC-class framework (Compaq's high speed, low latency clustering solution) and you have a killer supercomputer.
The Alpha is not limited to 2-way SMP. The Alpha can also be put into much larger SMP configurations. The AlphaServer ES40 is 4-way, the GS60 is 8-way, and the GS140 can have up to 14 Alpha processors in it. There is even a new SC that can have 64-512 processors! These are SMP machines, not Beowulf clusters.
Correction: The Compaq SC systems are not SMP boxes. They are ES40s clustered via high-speed, low-latency Quadrics interconnects. Nice systems, but a few quirks.
Compaq is working on very large SMP systems. Ask your sales rep about details.
Doubletwist is sort of a spinoff/evolution of Pangea Systems, Inc. and has been formed to be a ASP for genomic science. Pangea wasn't such an unknown company in the biotech field. What Doubletwist desperately needed right now was a campaign to increase their name recognition. I guess they now got that...
Whether the announced annotations to the known genome sequence data are really worth the hoopla will be known in a few days or weeks when genomics scientists have had a chance to look it over. In the meantime, relax! There will be lots more such announcements in this hotly contested field. It's just like chip wars.
The real question is this: If the same money were spent on, say, Athlon nodes connected with channel bonded fast ethernet (or even myrinet); could you get even more performance? I figure that you could build a cluster of stripped down Athlon-700's on channel bonded ether for around $2k per node including switches, etc. That would allow up to 7500 nodes (though I imagine that network bandwidth/latency would kill your performance at that scale). Hmmm...
You got it all wrong. What drives such a purchase is: How many CPUs can my app use advantageously? If the app scales to a few hundred CPUs, but not to thousands, there is absolutely no point to build a equivalent (in terms of theoretical Teraflops) system from thousands of cheaper CPUs. You absolutely want a few hundred of the fastest possible chips.
As to interconnects: For many MPI-based codes you don't want to go over ethernet or any other TCP/IP-based interconnect. You want low level protocols over low latency/high bandwidth interconnects. TCP/IP really sucks when it comes to latency, no matter how many channels you bond together.
You would need a benchmarking standard, not just a standard benchmark. And that standard needs to be protected by a license or something to ensure that everyone plays by the rules.
The benchmarking standard should state that code changes are not allowed, and it should detail how the benchmark was to be run, how it was to be reported and where.
We already have standard benchmarks and benchmarking standards with SPEC. It's just not really open source. But as another poster detailed here, a SPEC equivalent should be pretty straightforward to implement as open source. Just who or what mechanism would make sure that noone cheats is another issue, though.
That is why open source benchmarks are a good idea -- not only does it allow people to improve on the code directly, but it lets people see exactly what is going on behind the scenes.
That is a very bad idea, indeed. If the code base of the benchmark changes at all, none of the numbers are comparable between releases. This is exactly why tightly controlled benchmarks like SPEC have been successful. SPEC only changes every few years, there are clear rules of what you can do and what not while compiling and running the benchmarks and there are rules about how to report the resulting numbers.
Inasfar as one can trust generic benchmarks, SPEC has held up nicely and allowed us to superficially compare systems from Unix vendors with different CPUs, different architecture and different OS. Even with infrequent updates, the transition from one version of SPEC to the next gets in the way sometimes. I can only imagine how bad a true open source solution without additional rules would be.
I know, SPEC isn't open source in the strict sense, but it IS a broadly accepted benchmark suite of which source is available, and it has served us well for the past 11 years.
As well as generic benchmark can serve anyway. There is of course no substitute to check out a box with your own apps and workload.
Apparently, a lot of people here like to shout "beowulf cluster" and don't understand much about the problems of massively parallel computing.
The bottleneck is inter-processor communication. If all you are doing is trying to brute force a cipher the processors are almost independent and easily reach the theoretical aggregate performance figures. But if you are doing complex physical simulations you can end up waiting for data most of the time and using only a fraction of your theoretical parallel power. Well spoken!
We've just been through an evaluation cycle where we pitted all major HPC systems and most Intel/Linux clustering solutions against each other. While I can't post numbers (yet), the results are that Intel/linux clusters come out at the bottom, mostly because of high latencies in the interconnect. I think it is not only the hardware per se, but immature drivers also.
We looked at Myrinet, Giganet and SCI. All of them were pretty close in performance, but don't come near more established SMP, NUMA or clustering technologies under traditional Unices. Our benchmark is molecular dynamics code with a reasonably large solvated protein.