Yeah, but you forget between these 2 computers lie nearly 2 years of moores law, and if i remember correctly ES was more like 360 mio$.
And 5m Bucks for the cluster? 2000 machines a 2.500$. SOmewhere something doesnt fit there. Sure, they have gigabit ethernet on board, but the switching architectuer alone should cost at least a few millions. Not to mention the building.
Yeah, sure. How do you thnik that they did implement that magic software correction in the linpack benchmark? Simply run the matrix multiplies twice and check if the result is the same? (no, and i dont need to explain why)
Superconductivity was observed before there was a mathematical background. Not understood.
Who do you think really discoverd gravity: Newton after realizing the 1/r^2 nature or the fist sapient man dropping a rock on his foot?
Usage without understanding isnt worth very much. And certainly not a nobel price. Onnes got his not for his "discovery" of superconductivity, but because he created a process to liquify helium, thus revolutioning deep temperature physics.
Von Klitzing is another example: He didnt just notice the bumbs in the hall-effect, he created a mathematical theory by which nor the unit Ohm can be derived from e, h, and a constant R_K. And he got his price...
Hey, i wasnt joking. I use mathematica 4 and have a hp 48ex calculator, but both cant compare to the parser of google.
The unit conversion alone makes this thing invaluable. No need to use the right constants if you want eV or Joule, or defect ions per m^3 or cm^3, just type "in " after the formula....
What kind of processor boundness has your program? If it is fpu-limited, than i totaly accept you speedup. Integer would be a little trickier, because of the pipeline lenght of the g5 which makes such a speedup a bit unlikely.
People seem to forget why apple never submitted spec-scores: A 1GZ G4 has a SpecFloat of around 240 using the best compiled Apple has. Compare this to 1000+ of Amd and intel, you ll see the reason. And yes, spec is biased blalbalbla, the fact is that before the g5, crap compilers and abyssmal memory performance destroyed any kind of performance UNLESS you spend the time and hand-optimized the code using alti-vec ect. That the whole reason MAc-fans believed they had better/same performance back then: Because a few programms, mostly photoshop, were optimized so much they were as fast or faster than on the pc. But why did they need so much optimisation? Because it would have looked really bad if the ppc had lost by a factor of 5 against pcs, a thing that most likely would have happened if they were just compiling c-code.
a joke between a few friends and me years ago was: How did apple manage to create a risc-cpu with 32 registers that SUCKS THAT MUCH?!? You would expect less memory pressure and higher efficiency because of the register set, but somehow that never happened...
Well, there is the opteron, so think again about "Front side bus". And then you should realize the MEMORY BANDWITH. dual channel ddr400 can theoretically deliver 6.4 GB/s. If you have mixed read/writes, say 4GB effective. At 2GHz, 4load per cycle and 2 Stores per cyle using 64bit floating point numbers means 6*8Byte*2^9=48GB/s. But more: You have 2 Cpus attached to the memory bus. So you would need 96 GB/s, but you have no more than 6.4 GB/s.
See what i mean? You can optimize memory location for burst access, but you cant get faster then the memory itself. (ok, with blocking algorithms you can most likly cut bandwith in half, but the g5 is still big-brained and little-legged, so to say)
considering the fact that you will need a machine of this power to run hl2 or any other high profile game of the next year, you should whine about this super duper powerful ubermacs wasting their time with word. Btw: i have a magazine from 1990 in which 486s are tested. they costed between 12000-20000$, and the reviewer was eager to tell the reader that it would be a waste to use that much computing power for anything else than high end cad or database work...
sorry, if you only need performance, stay in the pc area. With the arival of the g5, the performance of macs has finally catched up with x86. But while on the paper the chip looks like a killer, it looses to the a64. and most likely prescott,too, but thats speculation.
Yes, the g5 has dual fpus capable of doing a mac each per cyle. But people should realize that even with 32 registers you need 2 loads and one store per MAC. Thus whenever you could really use the power of the 2 fpus, you will be so hopelessly memory-bound that it wont matter if you have 1 or 2 mac units...
I think the problem like in every system is that you have to adapt to become succesfull. Back in the old days, when news took days or weeks to reach the people, things like public identity, pr,ect were irrelevant. Today, if a scientist changes to politics, he has to change his way of thinking from "rational scientist" to ">how to cheat the prople into voting me politician". It may not happen at one, but it will happen or he will fail. And when it happened, he could have come from a totally diffrent background and it wouldnt matter.
Most people SERIOUSLY underestimate the power of gunpowder and explosives. I guess this comes from too many movies showing the hero surviving handgranates detonatin 5 feet away act.
"a few ounces" is around the payload of a normal offensive hand granate. A AIM-9J missile has less then a pound of explosives in the warhead and can destroy a jet fighter.
I once tried what happened if you take about 5 grams of nitrocellusoses and fire it closely enough confined that a deflagration to detonation transition happened. I couldnt really hear the initial blast (dont know why, perhaps ear overload), but 5 seconds later the sound returned from the other side of the valley i live in like a thunderclap.
At that moment i was REALLY happy that i was smart enough to burry a hole 2 feet deep and insert a bigger metal tube to stop fragmentation.
From what i see of shares in winmx, kazaa and edonkey, i would say by volume:
45% Movies 20% Porn 20% Warez 10% Music 5% Series, books, ect.
Music doesnt matter anymore. In times of dsl and cable, 5mb for a song dont matter. even if you download 200. Movies are a great part. 2cd releases are common, weighting 1.5GB per shot and most people download them, watch them, and at once download the next. Porn is very common, but not as much as you would believe. It isnt shared very much (compared to warez and movies), and there are less files. Porn just isnt very suitable for p2p. Who wants to download for a day to get a porn movie to jack off? Just get a crack for a porn-site and watch the video stream in realtime...:)
but the differences are bluring. The earth simulater uses vector processors, but NEC produces even workstations with that cpu. Thats not too different from the other clusters. Even if itanium,opteron and alpha may me "of the shelf" cpus, a rack with >myrinet connectivity or even switched HT links isnt very different from a "highly specialied, pupose.build machine"
I can give you your wanted "review" A64 fx51:=Opteron A64 a little slower G5: Slow crap, Half to 75% of A64 performance (not countin photoshop gaussian blur and the 3 other altivec optimiced algorithms in exitance)
But i guess you smartass dont understand that 64bit x86 on the desktop means x86 desktop chips should be used in the comparison. And yes, p4 is one.
I just wonder how well gouvernment would handle things like child care, social welfare, building of highways (NOT information), waste removal or foreign politics if only slashdot approved geeks who spend their time thinking about whether GPL or the bsd-license is better for free software were politicians.
Sucks. Is slow as a Geforce MX and expensive like a radeon 9800pro. Doesnt have any drives providing the (limited) DX9 features the card has, like 10Bit/channel. And it wont get any, because they said so and they fired the driver team this spring.
Well, it does have good 2D and triplehead, which would make sense if the card was 5 times as fast as it is (of course you can play Q3 or any other 4 year old game, no probs, but try something newer..)
how much money do you believe could be made in this cutthoat market where the whole device has to be cheaper than a high end gfx-card and handhelds with GPU shouldnt be more expensive than those without?
on the other hand, if you really look you will find PowerVR MBX and the new Bitboys chip.
But the last 2 years, PowerVR, SIS, S3 and Tritend produced little more than hot Air. The specs might look good on paper, but in the end the chips still sucked.
Prime example: Parhelia. On release 256bit memory interface,8 texel per clock -> everybody thought it would rock. reality: Horrible drivers, DX9 drivers "will not be made", abysmal memory performance because of lack of bandwith saving gimmicks, ect.
S3 in particular hasnt got a very good track record. The last time they released a product that was supposed to reach nvidea&atis performance, they ended up with a chip chose T&L never worked and was emulated in a driver that sucked in every aspect except producing render errors...
Slashdot Mirror
a spinning HD uses 3-5 watt.
A modern cpu at full power needs 60-100W, idling 10-40.
Do the math
They run nativ apps. Both processors can process 32bit code natively, and they do it here.
And AMD should profit MUCH more from 64 bit than g5:
G5 runs the same, only in 64 bit (more memory/cache bw required)
K8 gets twice as much registers.
Yeah, but you forget between these 2 computers lie nearly 2 years of moores law, and if i remember correctly ES was more like 360 mio$.
And 5m Bucks for the cluster? 2000 machines a 2.500$.
SOmewhere something doesnt fit there. Sure, they have gigabit ethernet on board, but the switching architectuer alone should cost at least a few millions.
Not to mention the building.
Yeah, sure. How do you thnik that they did implement that magic software correction in the linpack benchmark?
Simply run the matrix multiplies twice and check if the result is the same? (no, and i dont need to explain why)
Superconductivity was observed before there was a mathematical background. Not understood.
Who do you think really discoverd gravity: Newton after realizing the 1/r^2 nature or the fist sapient man dropping a rock on his foot?
Usage without understanding isnt worth very much. And certainly not a nobel price.
Onnes got his not for his "discovery" of superconductivity, but because he created a process to liquify helium, thus revolutioning deep temperature physics.
Von Klitzing is another example: He didnt just notice the bumbs in the hall-effect, he created a mathematical theory by which nor the unit Ohm can be derived from e, h, and a constant R_K. And he got his price...
Hey, i wasnt joking. I use mathematica 4 and have a hp 48ex calculator, but both cant compare to the parser of google.
The unit conversion alone makes this thing invaluable. No need to use the right constants if you want eV or Joule, or defect ions per m^3 or cm^3, just type "in " after the formula....
or add round brackets (123-867-5309)
/x eV))
or add an = 123-867-5309=
Btw: google calculater is SO COOOOOL.
Need a bolzmann distribution but dont want to look up k_b?
just type in google (e^(k*300 K
What kind of processor boundness has your program?
If it is fpu-limited, than i totaly accept you speedup. Integer would be a little trickier, because of the pipeline lenght of the g5 which makes such a speedup a bit unlikely.
People seem to forget why apple never submitted spec-scores: A 1GZ G4 has a SpecFloat of around 240 using the best compiled Apple has. Compare this to 1000+ of Amd and intel, you ll see the reason. And yes, spec is biased blalbalbla, the fact is that before the g5, crap compilers and abyssmal memory performance destroyed any kind of performance UNLESS you spend the time and hand-optimized the code using alti-vec ect.
That the whole reason MAc-fans believed they had better/same performance back then: Because a few programms, mostly photoshop, were optimized so much they were as fast or faster than on the pc. But why did they need so much optimisation?
Because it would have looked really bad if the ppc had lost by a factor of 5 against pcs, a thing that most likely would have happened if they were just compiling c-code.
a joke between a few friends and me years ago was: How did apple manage to create a risc-cpu with 32 registers that SUCKS THAT MUCH?!?
You would expect less memory pressure and higher efficiency because of the register set, but somehow that never happened...
Well, there is the opteron, so think again about "Front side bus".
And then you should realize the MEMORY BANDWITH.
dual channel ddr400 can theoretically deliver 6.4 GB/s. If you have mixed read/writes, say 4GB effective.
At 2GHz, 4load per cycle and 2 Stores per cyle using 64bit floating point numbers means 6*8Byte*2^9=48GB/s.
But more: You have 2 Cpus attached to the memory bus. So you would need 96 GB/s, but you have no more than 6.4 GB/s.
See what i mean?
You can optimize memory location for burst access, but you cant get faster then the memory itself.
(ok, with blocking algorithms you can most likly cut bandwith in half, but the g5 is still big-brained and little-legged, so to say)
considering the fact that you will need a machine of this power to run hl2 or any other high profile game of the next year, you should whine about this super duper powerful ubermacs wasting their time with word.
Btw: i have a magazine from 1990 in which 486s are tested. they costed between 12000-20000$, and the reviewer was eager to tell the reader that it would be a waste to use that much computing power for anything else than high end cad or database work...
times change
sorry, if you only need performance, stay in the pc area.
With the arival of the g5, the performance of macs has finally catched up with x86. But while on the paper the chip looks like a killer, it looses to the a64. and most likely prescott,too, but thats speculation.
Yes, the g5 has dual fpus capable of doing a mac each per cyle. But people should realize that even with 32 registers you need 2 loads and one store per MAC. Thus whenever you could really use the power of the 2 fpus, you will be so hopelessly memory-bound that it wont matter if you have 1 or 2 mac units...
well i could say it is spelled Handgranate. Dort wo beim bund war hat man einen Arsch wie dich zum Kloputzen mit der Zunge eingeteilt.
I think the problem like in every system is that you have to adapt to become succesfull.
Back in the old days, when news took days or weeks to reach the people, things like public identity, pr,ect were irrelevant.
Today, if a scientist changes to politics, he has to change his way of thinking from "rational scientist" to ">how to cheat the prople into voting me politician". It may not happen at one, but it will happen or he will fail.
And when it happened, he could have come from a totally diffrent background and it wouldnt matter.
Most people SERIOUSLY underestimate the power of gunpowder and explosives. I guess this comes from too many movies showing the hero surviving handgranates detonatin 5 feet away act.
"a few ounces" is around the payload of a normal offensive hand granate. A AIM-9J missile has less then a pound of explosives in the warhead and can destroy a jet fighter.
I once tried what happened if you take about 5 grams of nitrocellusoses and fire it closely enough confined that a deflagration to detonation transition happened. I couldnt really hear the initial blast (dont know why, perhaps ear overload), but 5 seconds later the sound returned from the other side of the valley i live in like a thunderclap.
At that moment i was REALLY happy that i was smart enough to burry a hole 2 feet deep and insert a bigger metal tube to stop fragmentation.
From what i see of shares in winmx, kazaa and edonkey, i would say by volume:
... :)
45% Movies
20% Porn
20% Warez
10% Music
5% Series, books, ect.
Music doesnt matter anymore. In times of dsl and cable, 5mb for a song dont matter. even if you download 200.
Movies are a great part. 2cd releases are common, weighting 1.5GB per shot and most people download them, watch them, and at once download the next.
Porn is very common, but not as much as you would believe. It isnt shared very much (compared to warez and movies), and there are less files.
Porn just isnt very suitable for p2p. Who wants to download for a day to get a porn movie to jack off? Just get a crack for a porn-site and watch the video stream in realtime
but the differences are bluring. The earth simulater uses vector processors, but NEC produces even workstations with that cpu. Thats not too different from the other clusters. Even if itanium,opteron and alpha may me "of the shelf" cpus, a rack with >myrinet connectivity or even switched HT links isnt very different from a "highly specialied, pupose.build machine"
actually, the 10km are not hight. A modern simulation uses 30-70 layers, spread across the 15-25km height they simulate
actually, we know the speed of light exactly, because the meter was redefined to make c==2,99792458*10^8m/s
And how do you get the tousands of tons of material to construct the space elevator in orbit?
I can give you your wanted "review"
A64 fx51:=Opteron
A64 a little slower
G5: Slow crap, Half to 75% of A64 performance (not countin photoshop gaussian blur and the 3 other altivec optimiced algorithms in exitance)
But i guess you smartass dont understand that 64bit x86 on the desktop means x86 desktop chips should be used in the comparison. And yes, p4 is one.
I just wonder how well gouvernment would handle things like child care, social welfare, building of highways (NOT information), waste removal or foreign politics if only slashdot approved geeks who spend their time thinking about whether GPL or the bsd-license is better for free software were politicians.
/. by me)
longest sentence ever (on
Sucks. Is slow as a Geforce MX and expensive like a radeon 9800pro.
Doesnt have any drives providing the (limited) DX9 features the card has, like 10Bit/channel.
And it wont get any, because they said so and they fired the driver team this spring.
Well, it does have good 2D and triplehead, which would make sense if the card was 5 times as fast as it is (of course you can play Q3 or any other 4 year old game, no probs, but try something newer..)
how much money do you believe could be made in this cutthoat market where the whole device has to be cheaper than a high end gfx-card and handhelds with GPU shouldnt be more expensive than those without?
on the other hand, if you really look you will find PowerVR MBX and the new Bitboys chip.
But the last 2 years, PowerVR, SIS, S3 and Tritend produced little more than hot Air. The specs might look good on paper, but in the end the chips still sucked.
Prime example: Parhelia.
On release 256bit memory interface,8 texel per clock -> everybody thought it would rock.
reality: Horrible drivers, DX9 drivers "will not be made", abysmal memory performance because of lack of bandwith saving gimmicks, ect.
S3 in particular hasnt got a very good track record. The last time they released a product that was supposed to reach nvidea&atis performance, they ended up with a chip chose T&L never worked and was emulated in a driver that sucked in every aspect except producing render errors...
Galileo is multinational operation, and it is not under military direction.
and are not the us-citizens those who claim the eu doenst do anything without talking for month?