The sun is about 30% brighter than it was long ago, and it will continue getting brighter. The other poster just has his time-scales all wrong - it won't be a critical problem for hundreds of millions of years.
And the sun is getting bigger, as hydrogen turns to helium. But again, the time scale is long.
When I read about information technology on/., I learn stuff. When I read about science, I'm dismayed. Most of this global warming thread is ridiculous.
I am bored almost out of my mind. And I get to do algorithm development too, not just coding.
Its largely a kind of alienation....the illusion of being a part of a team is long, long gone. Even feeling like a cog in a big wheel would be a significant improvement. Work is more like being a part of a food chain - a tool for people above you who want to climb the salary scale.
Maybe this is nobody's fault, I don't know. It seems like it has happened to nearly everyone.
I'm hoping there are still niches I can find later where the magic isn't all gone, but this is probably wishful thinking.
If a computer can sort through the synapses, find the ones that are looking to match "man with RPG in the distance", and figure out when they fire, it can perhaps bring something up on the display faster than the person can. Computers, after all, can process a small number of things faster than we can. They just can't process as many complex things in parallel. No such device will be able to 'sort through the synapses' - it won't have that kind of access. At best it can detect that something was noticed. Furthermore, given a detection, the human is (in my opinion) still much better at recognizing what it is, even subconciously. Although computers do process certain types of problems faster and better than humans, that does not include this type of pattern matching, even for a single shape.
As someone who works as a government contractor, my guess is it is because government bureaucracy stifles innovation. Most smart minds would rather work in academia where they get more freedoms, less restrictions, and are more easily able to surround themselves with likeminded individuals. I think acadmia is to some degree a 'play science' pyramid scheme. Certain types of useful research can be accomplished in that setting, but for many worthwhile topics its impossible to find funding, and its always a treadmill. Private sector R&D would be better, if corporations were less fixated on short-term stock gains, and more interested in long term investment. Actually the market looks pretty bleak to me for people who want to do real research, and not just publish rigorous yet largely meaningless papers and pad their resumes.
Paypal claimed the seller had the money, and told the seller we (the buyer) had the money, and lied and stonewalled for months. Finally they just gave the money back without explanation after a state Attourney General inquired on our behalf.
The previous time I posted this on/., I was modded down as a liar. But unbelievable as the stody is, it is the truth.
In fairness to Paypal, our experience was shortly after it was purchased by EBay, so probably EBay cleaned it up some since then.
The South China Sea all the way to Malaysia and Philippines as being a part of China. Needless to say, the map treats all other seas, gulfs, and bays on the globe as being international waters.
Chinese culture seems to me to still be in the 19th century in many regards, and unable to engage in self-criticism. I've never met a Chinese person who could admit opposing points in relation to Tibet for example, notwithstanding that these people are all intelligent and decent in other regards.
I'd be able to feel more sorry for Tibet if the exiled government wasn't stuck in the middle ages though.
Megatest Q2/52 testers are still being used. As I recall, the host computer is a PDP-11. In some contexts there's no compelling reason to upgrade to more expensive equipment if the old equipment is getting the job done. The same company may use ancient equipment alongside latest-greatest, depending on requirements and cost tradeoffs for different products.
Aerial imagery is impersonal, and like ink blots in the sense that its possible for law enforcement to fit it to plausible seeming yet completely inaccurate crime scenarios. And even if it doesn't stand up in court, just being accused costs an innocent defendent a lot of time, money, and other problems. This stuff is seriously bad news. And we need to get past this being a partisan issue - its not coming from the top, from the 'Bush administration'. Granted that the Bush administration bears much responsibility, a lot of the people who are developing and pushing this technology are Democrats, and the issue isn't going to just go away when there's new leadership at the very top.
Do you have oil? If you do, then this corruption is a worldwide tragedy which must be stopped, we'll send troops^Wobservers right away. Argentine oil is much, much more important to the US interests than Iraq's, and the US has not reacted in such a manner to events there.
You can't get out of the "gravity well" with a "spaceplane". Where does the energy come from? That's what the big booster rockets and gigantic hydrogen tank on the shuttle are for.
If anyone is inclined to mod me a troll for contradicting facts well established in "Star Trek", please at least have integrity to respond with a counter-argument.
Something like the manned moon mission was cool for its own sake as a demonstration of engineering prowess. But the idea that any of this has been or can be steps towards a space-faring future is utter BS. From energy considerations, it doesn't work. If you think of planets as barren or far away, like say, a far away, barren continent, then it seems to work metaphorically. But if you calculate how far away they really are, and how hot or cold they really are, it never works, not even close. Not suprisingly, space-travel components always avoid those calculations. For example...people talk of extracting hydrogen from the Martian atmosphere for a return trip. How long would this take, using optimistic, back-of-the-envelope calculations? Serious thinking about space exploration should start with such estimates. But they are always mysteriously vague or absent. While I'm at it, "private" space flight involving low-cost airplane-like vehicles is also a con. Yes, such craft can get into "space". But to reach a useful orbit requires something like a huge rocket - just calculate the potential energy difference. The press releases never mention that. NASA scientists must of course make a living. It would be nice to see more earth science and big telescopes. Give the honest scientists more money, and let the rest find something else to do for a living.
(They are mostly males between 25 and 44 years of age.) The number of clicks on an ad campaign is also not strongly correlated with brand awareness for the ads' subject, according to the study.
I'm guessing that the typical top 6% clicker uses a different metric than brands when clicking.
Right, though as I recollect equation editor only spams you when it thinks you're frustrated. Years back equation editor was significantly more efficient to use than it is now (at least as I use it), then they crippled it and added those messages.
(The quadword instructions have 6 cycle latencies.)
If, by quadword instruction, you mean an instruction that operate on 128-bit registers, this is incorrect.
First of all, 95% of the SPU instructions operate on 128-bit registers. More than half of them have a latency of only 2 or 4 cycles. And the vast majority of them have a throughput of 1 per cycle (ie. 0-cycle stall). The only instructions with a 6-cycle stall are double-precision instructions and 'fscrrd' (floating-point status control register read). See table B-1 in the Cell Broadband Engine Programming Handbook: http://www.ibm.com/chips/techlib/techlib.nsf/techdocs/9F820A5FFA3ECE8C8725716A0062585F?Open&S_TACT=105AGX16&S_CMP=LP
This means for example that you can execute 1 'fm' per cycle (4-way SIMD single-precision floating-point multiply), or 1 'and' per cycle (4-way SIMD word add).
Yes, mostly you just repeated what I said, though perhaps I was insufficiently clear. Almost all of the single precision floating point operations have 6 cycle latencies and 1 cycle throughputs, which is why IBM's CBE Tutorial lists the "single-precision floating-point class" as having a latency of 6 cycles (Table 3.2). Most of the integer related operations relevant to my signal processing operations are also 6 cycles. A few are 7 cycles, but it doesn't matter much, since I was able to get 1 cycle throughput with loop unrolling, as I indicated. For the sake of brevity I didn't go into more detail. Saying that only double precision operations have unavoidable stalls is also just what I indicated.
Another point of clarification....most of the important parts of my applications can be coded entirely and efficiently using the SIMD instructions, without selects, shifts, or shuffles. However, in some places approximations can be made which are actually faster overall, even though they involve what are essentially scalar operations. I made these improvements where uncertainty analysis showed that they did not significantly degrade the alorithm accuracy, and where they improved performance on CBE. Similar improvements were much easier and more efficient on other processors however, namely on Xeon (5310) and Opteron. Without a dozen page post I can't completely spell everything out, but I think these comments are generally applicable to other scientific and signal processing applications, because they involve common features like interpolations and complex arithmetic.
I did invite feedback, and there was no way to tell from my original post how careful I was in reaching my conclusions, so I appreciate the response.
These instructions are only required for scalar code. Vectorizable code does not generally require additional selects, shifts and shuffles, unless the compiler can't guarantee that the data is aligned. However, since all instructions are SIMD, for scalar code, the compiler has to emit those additional instructions in order to mask the effect of loads and stores and the rest of the instructions being SIMD. This makes it slow and bloated for scalar code, but not for SIMD code. The algorithms in question have a few spots that are inherently scalar, so I have to put those intrinsics in by hand to transition to the subsequent vectorizable calculations. If I let the compiler do it (the gnu compiler provided by IBM for the SPU), then it is very, very inefficient, much worse than what I can do myself. But on Xeon or Opteron, the same transitions between scalar and vector calculations are handled very efficiently by the compiler.
Whoops...sorry for the HTML snafus. This should have been
It's easy enough to measure total power consumption of the system with an AC power meter. The original 90nm PS3 used 200W. The 65nm PS3 cut that down to 135W. We'll see how it goes with the 45nm. Thanks, I've seen that. But IBM is pitching the cell processor as having a future in signal processing and high performance computing. Developing those kinds of applications on PS3 and current (very expensive) blade servers might make sense if there are going to be more power and cost effective solutions available later. So my interest was in knowing what the figure is for the processor itself, not for PS3.
It's easy enough to measure total power consumption of the system with an AC power meter. The original 90nm PS3 used 200W. The 65nm PS3 cut that down to 135W. We'll see how it goes with the 45nm.
Thanks, I've seen that. But IBM is pitching the cell processor as having a future in signal processing and high performance computing. Developing those kinds of applications on PS3 and current (very expensive) blade servers might make sense if there are going to be more power and cost effective solutions available later. So my interest was in knowing what the figure is for the processor itself, not for PS3.
They probably won't reveal the absolute power consumption for competitive reasons. But they're pitching it as fast and low power, and the apples-to-apples practical speed comparisons I have seen, it doesn't seem to be that fast. (With carefully optimized codes its about equal to 2 Quad Core Xeons, in my experience, with the Xeons running at half the clock rate.) So that leaves power consumption as a major remaining selling point. If it was really low power as advertized, wouldn't revealing the power consumption benefit them? If they won't reveal it, they are hiding it. Knowing that it will be 40% better is of limited value if we don't know how good it is now.
Relevance of CBE beyond PS3 of course depends in large degree on its computing performance. For the applications I've looked at, I haven't been very impressed. They say it does 204GFLOPS, but approaching that requires being able to use all multiply-add instructions, which count as two operations. (Some sources say the two operations per clock cycle per SPU is due to there being two pipelines, however, only one of the pipelines handles arithmentic operations and the other is exclusively for load, store, control, and a few shift operations.) Also, it seems to take a lot of select, shift, and shuffle instructions to make efficient use of the quadword (SIMD) instructions. With Xeon and Opteron, use of the quadword instructions seems to require far fewer other additional cycles. And this is with floats, with instruction related stalls completely eliminated on CBE through careful loop unrolling and other methods. (The quadword instructions have 6 cycle latencies.) I can only get performance comparable to 2 quad-core Xeons, which doesn't seem that good considering what is advertized, and considering the 4x difference in the peak performance specs. And CBE does much worse where double precision is necessary, with 6 cycle stalls being unaviodable on every instruction. It seems overblown. Comments?
Slashdot Mirror
The sun is about 30% brighter than it was long ago, and it will continue getting brighter. The other poster just has his time-scales all wrong - it won't be a critical problem for hundreds of millions of years.
And the sun is getting bigger, as hydrogen turns to helium. But again, the time scale is long.
When I read about information technology on /., I learn stuff. When I read about science, I'm dismayed. Most of this global warming thread is ridiculous.
The Vista machine it is on is still booting.
I am bored almost out of my mind. And I get to do algorithm development too, not just coding.
Its largely a kind of alienation....the illusion of being a part of a team is long, long gone. Even feeling like a cog in a big wheel would be a significant improvement. Work is more like being a part of a food chain - a tool for people above you who want to climb the salary scale.
Maybe this is nobody's fault, I don't know. It seems like it has happened to nearly everyone.
I'm hoping there are still niches I can find later where the magic isn't all gone, but this is probably wishful thinking.
Who stole it from who?
the way it has been going.
Paypal claimed the seller had the money, and told the seller we (the buyer) had the money, and lied and stonewalled for months. Finally they just gave the money back without explanation after a state Attourney General inquired on our behalf.
/., I was modded down as a liar. But unbelievable as the stody is, it is the truth.
The previous time I posted this on
In fairness to Paypal, our experience was shortly after it was purchased by EBay, so probably EBay cleaned it up some since then.
The South China Sea all the way to Malaysia and Philippines as being a part of China. Needless to say, the map treats all other seas, gulfs, and bays on the globe as being international waters.
Chinese culture seems to me to still be in the 19th century in many regards, and unable to engage in self-criticism. I've never met a Chinese person who could admit opposing points in relation to Tibet for example, notwithstanding that these people are all intelligent and decent in other regards.
I'd be able to feel more sorry for Tibet if the exiled government wasn't stuck in the middle ages though.
Megatest Q2/52 testers are still being used. As I recall, the host computer is a PDP-11. In some contexts there's no compelling reason to upgrade to more expensive equipment if the old equipment is getting the job done. The same company may use ancient equipment alongside latest-greatest, depending on requirements and cost tradeoffs for different products.
Aerial imagery is impersonal, and like ink blots in the sense that its possible for law enforcement to fit it to plausible seeming yet completely inaccurate crime scenarios. And even if it doesn't stand up in court, just being accused costs an innocent defendent a lot of time, money, and other problems. This stuff is seriously bad news. And we need to get past this being a partisan issue - its not coming from the top, from the 'Bush administration'. Granted that the Bush administration bears much responsibility, a lot of the people who are developing and pushing this technology are Democrats, and the issue isn't going to just go away when there's new leadership at the very top.
Hasbro sues Idaho over infrignement on Mr. Potato Head brand
You can't get out of the "gravity well" with a "spaceplane". Where does the energy come from? That's what the big booster rockets and gigantic hydrogen tank on the shuttle are for.
If anyone is inclined to mod me a troll for contradicting facts well established in "Star Trek", please at least have integrity to respond with a counter-argument.
whoops, that's almost as annoying as someone leaving the caps-lock on, sorry
Something like the manned moon mission was cool for its own sake as a demonstration of engineering prowess.
But the idea that any of this has been or can be steps towards a space-faring future is utter BS. From energy considerations, it doesn't work. If you think of planets as barren or far away, like say, a far away, barren continent, then it seems to work metaphorically. But if you calculate how far away they really are, and how hot or cold they really are, it never works, not even close. Not suprisingly, space-travel components always avoid those calculations. For example...people talk of extracting hydrogen from the Martian atmosphere for a return trip. How long would this take, using optimistic, back-of-the-envelope calculations? Serious thinking about space exploration should start with such estimates. But they are always mysteriously vague or absent.
While I'm at it, "private" space flight involving low-cost airplane-like vehicles is also a con. Yes, such craft can get into "space". But to reach a useful orbit requires something like a huge rocket - just calculate the potential energy difference. The press releases never mention that.
NASA scientists must of course make a living. It would be nice to see more earth science and big telescopes. Give the honest scientists more money, and let the rest find something else to do for a living.
would seem to be a much bigger problem for most of us.
And if you're vegetarian, the metal-laden mining tailings that are commonly used as fertilizer can't be a good thing.
I'm guessing that the typical top 6% clicker uses a different metric than brands when clicking.
Right, though as I recollect equation editor only spams you when it thinks you're frustrated. Years back equation editor was significantly more efficient to use than it is now (at least as I use it), then they crippled it and added those messages.
If, by quadword instruction, you mean an instruction that operate
on 128-bit registers, this is incorrect.
First of all, 95% of the SPU instructions operate on 128-bit registers.
More than half of them have a latency of only 2 or 4 cycles. And the
vast majority of them have a throughput of 1 per cycle (ie. 0-cycle stall).
The only instructions with a 6-cycle stall are double-precision instructions
and 'fscrrd' (floating-point status control register read). See table B-1
in the Cell Broadband Engine Programming Handbook:
http://www.ibm.com/chips/techlib/techlib.nsf/techdocs/9F820A5FFA3ECE8C8725716A0062585F?Open&S_TACT=105AGX16&S_CMP=LP
Yes, mostly you just repeated what I said, though perhaps I was insufficiently clear. Almost all of the single precision floating point operations have 6 cycle latencies and 1 cycle throughputs, which is why IBM's CBE Tutorial lists the "single-precision floating-point class" as having a latency of 6 cycles (Table 3.2). Most of the integer related operations relevant to my signal processing operations are also 6 cycles. A few are 7 cycles, but it doesn't matter much, since I was able to get 1 cycle throughput with loop unrolling, as I indicated. For the sake of brevity I didn't go into more detail. Saying that only double precision operations have unavoidable stalls is also just what I indicated.This means for example that you can execute 1 'fm' per cycle (4-way
SIMD single-precision floating-point multiply), or 1 'and' per cycle
(4-way SIMD word add).
Another point of clarification....most of the important parts of my applications can be coded entirely and efficiently using the SIMD instructions, without selects, shifts, or shuffles. However, in some places approximations can be made which are actually faster overall, even though they involve what are essentially scalar operations. I made these improvements where uncertainty analysis showed that they did not significantly degrade the alorithm accuracy, and where they improved performance on CBE. Similar improvements were much easier and more efficient on other processors however, namely on Xeon (5310) and Opteron. Without a dozen page post I can't completely spell everything out, but I think these comments are generally applicable to other scientific and signal processing applications, because they involve common features like interpolations and complex arithmetic.
I did invite feedback, and there was no way to tell from my original post how careful I was in reaching my conclusions, so I appreciate the response.
Thanks, I've seen that. But IBM is pitching the cell processor as having a future in signal processing and high performance computing. Developing those kinds of applications on PS3 and current (very expensive) blade servers might make sense if there are going to be more power and cost effective solutions available later. So my interest was in knowing what the figure is for the processor itself, not for PS3.
Relevance of CBE beyond PS3 of course depends in large degree on its computing performance. For the applications I've looked at, I haven't been very impressed. They say it does 204GFLOPS, but approaching that requires being able to use all multiply-add instructions, which count as two operations. (Some sources say the two operations per clock cycle per SPU is due to there being two pipelines, however, only one of the pipelines handles arithmentic operations and the other is exclusively for load, store, control, and a few shift operations.) Also, it seems to take a lot of select, shift, and shuffle instructions to make efficient use of the quadword (SIMD) instructions. With Xeon and Opteron, use of the quadword instructions seems to require far fewer other additional cycles. And this is with floats, with instruction related stalls completely eliminated on CBE through careful loop unrolling and other methods. (The quadword instructions have 6 cycle latencies.) I can only get performance comparable to 2 quad-core Xeons, which doesn't seem that good considering what is advertized, and considering the 4x difference in the peak performance specs. And CBE does much worse where double precision is necessary, with 6 cycle stalls being unaviodable on every instruction. It seems overblown. Comments?