Well apart from obvious thermodynamic laws which implies that it must produce some heat, I think that photonic processors will produce much heat.
Think a little bit: photons do not interact directly, so it means that you need some matter to create interactions, and photon-matter interactions will definitely generate heat, possibly lot of heat as many useful interactions are "second order" effect ie the change of transparency of the matter is a 'byproduct' which means light must be very intense to induce the change, which means photonic processors won't be useful for many years..
One part where they may be useful is a 'fourier transform' coprocessor, as an interconnect bus but as the main processor is unchanged, this means electrical/optical conversions which release quite some heat..
> The only area where we are still significantly behind is in boot-time interrupt routing.
"Boot-time interrupt routing"?
Can someone explain what this means to me? I'm thinking about 'interrupt latency' (Linux has a low latency patch) but I'm not sure if I understand correctly..
I still remember the hype of Java for everything, everywhere.. What's fun is that even Sun is not able to make fast Java GUIs: in Solaris9, they have ported their tools to Java, which made me learn the command-line tools because the GUI was slow as hell (and was cumbersome to use).
I don't know if they've improved the tools on Solaris 10, since we've switched to Linux in the meantime..
By realizing that whatever you'll do, you're going to die so 1) you'd better profit (have fun, enjoy life) while you can. 2) you have to assess if some things are too risky or not, and avoid the 'too risky' thing: smoking, etc.. Note that the risk evaluation is highly personal: I'm a skydiver but I wouldn't drive a motorcycle because I consider it too risky..
If all else fail, you can go into religion and believe that there will be an afterlife, of course it's stupid but a huger number of people do it, so at least you wouldn't be alone:)
A better analogy: you're using an old word processor which gives you lots of trouble but is using a free file format.
Someone lends a much better wordprocessor for a while, allowing you to write much faster a lot of document and the lender also becomes famous.
Then the lender says that it's finished, on one hand you have wrote all these document so it was good, on the other hand the documents are in a specific file format and you'll have to convert all these document to the new tool you're going to have to use, possibly loosing some information in the process, which is a pain!
How can you ignore the pain, loss of time of the future conversion that will have to take place?
"PC's will take no longer then 10 seconds to boot into Windows XP on a clean install"
You're exagerating, I think that XP take at least 20s on a computer 10* more performant.
With BeOS, the computer was totally functionnal as soon as it gave you the hand. XP cheats by displaying the desktop but not giving you the hand, so that its boot time appear lower than it really is.
Also on BeOS, the system felt very responsive, more than XP running on a much more powerful hw, granted the applications which have gained weight with eye-candy improvement doesn't help. The bad part of BeOS is that there were very few applications, of course.
I disagree, do you remember the 'automatic' vehicles competition and the failure of its participants?
I bet that the software controling those vehicles worked perfectly in a simulated environement..
Even in a simple environement, the wheels are skipping, there can be dust, etc.. All things which are difficult to reproduce in a simulated environement.
Still I agree that this kind of project can only be interesting with a big numbers of robots, otherwise they will all die and the evolution won't occur.
Except of course that gay marriage is able to produce children! You know gay includes lesbians.. As for male homosexuals, they could adopt childs, AFAIK every study have shown that children raised by gay people are 'as happy' as children raised by straight people..
And currently while marriage is clearly done to encourage 'child production', it is not restricted to couples raising childs, so why should-it be different for gay couple?
>That kind of feedback is invaluable, and something a company like MicroSoft simply can't give, because they lack the necessary cross-platform experience to care.
Microsoft lack cross-platform experience? Strange, last I looked their OS have run on Alpha, MIPS (not sure about this one) for a long time and they are distributing a development kit for their next-gen X-Box console on.. PowerPC.
I don't like Microsoft either but the truth is that they have cross-platform experience.
Note that Linux works usually much better on x86 than on other architecture: much more distrib to choose from, more testers, etc..
Note that 'some believers' include popes and the like, usually those higher in the religion hierarchy are not saints..
Also I'm not American (that's why I confused father Bush and little Bush) so little Bush did nothing in my name.
As for Mother Theresa, boots@work summarized very well: she was a very good catholic, but I don't consider 'good catholic' as equivalent to good.. You can be a 'good catholic' and being good of course, but Mother Theresa wasn't..
>There are lots of religions that help other people and ask nothing from them in return.
As I said: name one religion in which believers have not killed other people in the name of the religion?
As for Mother Theresa and the orphans, sorry while helping orphans is good being against contraception in an overpopulated country is *very bad*! She is a good example of how faith can make people support stupid/ilogical behaviour..
I hope that you realize that there isn't much difference between a "cult" and a mainstream religion: I don't know any religion which haven't been used as some time or another to kill other people..
In a semi-related topic, I heard that Bush has said that he doesn't consider atheists as citizens and yet he has been elected, two times! How open and generous are some beleivers.. I wonder if we can infer from his two elections that his supporters share his so benevolent views against people who don't support their beliefs?
While theoretical advantages of LED are good, remember that (O)LED have not reached yet a good enough lifetime.. Plus to emit different color, the active material is different which will makes "interesting" color problems when the blue fades faster than the red..
> That being said, compiler technology isn't really taking advantage of this yet...
Are you sure? While programming a register allocator which works efficiently with 8 registers must be really hard, going from 8 to 16 is pretty easy, no?
Especially for multi-target compilers which generates also code for RISC and their big number of registers..
Now I agree that fine-tuning can be hard: if memory serves, the first 8 "legacy" register in x86-64 can be used with shorter instruction (1 byte smaller) so you can gain a little performance by ensuring that the most used registers are legacy registers..
Granted FS are used by few people but it is one type of game which is CPU-bound currently, no need to invent new types of objects: flight dynamic, bullets, etc.. are enough!
In IL2 Sturmovick, IA's opponent use a different physic engine which enables them to do figures that you can't imitate, which is a bit frustrating.. But I'm not sure that if this accelerator allowed using realistic physic for IA's opponent, the IA would be good enough to fly a plane with this complex physic.. Ok, who is going to make an IA accelerator board?:-)
On a more serious topic, I'd be interested to see what they're going to implement, while I don't really beleive that a standalone PPU will be a success, if the HW is not too different from a GPU, maybe a combined GPU-PPU would be interesting: a bunch of 'normal' FPU for doing vertex|physic process + some specialised unit dedicated to the pixel processing part.
The article give a link to an article describing a trick to implement coroutines in C, I found it quite enjoyable to read: http://www.chiark.greenend.org.uk/~sgtatham /corout ines.html
It is nice for a change to read about an "ugly trick" which is used to increase readability!
> As already mentioned games already do make use of the GPU and the CPU so we're fairly used to some multiprocessor concerns.
Correct, but of course, the higher the number of processor there is the more difficult it is to use them efficiently.. Also double core has a cost in CPU: each cores is smaller than a mono-CPU, if they are not well used, they could be slower than a single core..
As if games are CPU / GPU bound, it depends very much of the game: Prince of Persia is very different from a flight simulator and of course of the resolution..
Also you're right of course that games have several thread, but if your rendering thread use 100% of one core and the other threads use 10% of the other core, it is likely that the multi-core will be slower than a more powerfull single-core CPU..
I disagree for the latency part: multi-core, SMT CPU can somehow hide the latency: it can schedule another thread during the wait for the information.
Now of course, for this to work the implementation must allow the scheduling of another thread very fast and there must be threads schedulable with their workload present in the cache otherwise this just increase the pressure on memory.
And? That's just the CPU, dammit! Who really cares about the instruction set which the CPU runs?
As for being glorified PCs, have you read the article? Good LOM, no cable, Hot-swappable fan and power modules: it looks to me as a real server not a glorified PC! OK, it is not a Stratus but still it is as good as previous Sparc servers from Sun..
You're right the SPEs are double precision capable as the last article showed it. Now the SPUs don't really have double precision unit, they reuse the single precision unit: this saves a lot of gates as a poster said it below but also it reduces speed: DP maximum theoretical FLOPS are 1/10 of SP FLOPS maximum theoretical..
Now granted DP maximum theoretical FLOPS should still be very high for a chip which will go into PlayStations.
I doubt that it is possible to do this: dual licensing *increase* the user possibilities not limit it. As long as the GPL allows you to use the program without any consequence if you don't redistribute it, the other license doesn't matter..
While you're right about the difference, the truth is: when you boot a PC, 99.99% of the time the hardware is identical to what it was the last time.
So it should be possible to store an HW description into a small flash and boot *fast*, without having to discover hardware, and if the user want to add new critical "cold-plug" hardware (such as boot disk), he would just have to hit Del at the boot, and then a HW discover would be made..
Slashdot Mirror
Well apart from obvious thermodynamic laws which implies that it must produce some heat, I think that photonic processors will produce much heat.
Think a little bit: photons do not interact directly, so it means that you need some matter to create interactions, and photon-matter interactions will definitely generate heat, possibly lot of heat as many useful interactions are "second order" effect ie the change of transparency of the matter is a 'byproduct' which means light must be very intense to induce the change, which means photonic processors won't be useful for many years..
One part where they may be useful is a 'fourier transform' coprocessor, as an interconnect bus but as the main processor is unchanged, this means electrical/optical conversions which release quite some heat..
> The only area where we are still significantly behind is in boot-time interrupt routing.
"Boot-time interrupt routing"?
Can someone explain what this means to me?
I'm thinking about 'interrupt latency' (Linux has a low latency patch) but I'm not sure if I understand correctly..
You're not the only one..
I still remember the hype of Java for everything, everywhere..
What's fun is that even Sun is not able to make fast Java GUIs: in Solaris9, they have ported their tools to Java, which made me learn the command-line tools because the GUI was slow as hell (and was cumbersome to use).
I don't know if they've improved the tools on Solaris 10, since we've switched to Linux in the meantime..
By realizing that whatever you'll do, you're going to die so
:)
1) you'd better profit (have fun, enjoy life) while you can.
2) you have to assess if some things are too risky or not, and avoid the 'too risky' thing: smoking, etc..
Note that the risk evaluation is highly personal: I'm a skydiver but I wouldn't drive a motorcycle because I consider it too risky..
If all else fail, you can go into religion and believe that there will be an afterlife, of course it's stupid but a huger number of people do it, so at least you wouldn't be alone
A better analogy: you're using an old word processor which gives you lots of trouble but is using a free file format.
Someone lends a much better wordprocessor for a while, allowing you to write much faster a lot of document and the lender also becomes famous.
Then the lender says that it's finished, on one hand you have wrote all these document so it was good, on the other hand the documents are in a specific file format and you'll have to convert all these document to the new tool you're going to have to use, possibly loosing some information in the process, which is a pain!
How can you ignore the pain, loss of time of the future conversion that will have to take place?
"PC's will take no longer then 10 seconds to boot into Windows XP on a clean install"
You're exagerating, I think that XP take at least 20s on a computer 10* more performant.
With BeOS, the computer was totally functionnal as soon as it gave you the hand.
XP cheats by displaying the desktop but not giving you the hand, so that its boot time appear lower than it really is.
Also on BeOS, the system felt very responsive, more than XP running on a much more powerful hw, granted the applications which have gained weight with eye-candy improvement doesn't help.
The bad part of BeOS is that there were very few applications, of course.
I disagree, do you remember the 'automatic' vehicles competition and the failure of its participants?
I bet that the software controling those vehicles worked perfectly in a simulated environement..
Even in a simple environement, the wheels are skipping, there can be dust, etc..
All things which are difficult to reproduce in a simulated environement.
Still I agree that this kind of project can only be interesting with a big numbers of robots, otherwise they will all die and the evolution won't occur.
Except of course that gay marriage is able to produce children!
You know gay includes lesbians..
As for male homosexuals, they could adopt childs, AFAIK every study have shown that children raised by gay people are 'as happy' as children raised by straight people..
And currently while marriage is clearly done to encourage 'child production', it is not restricted to couples raising childs, so why should-it be different for gay couple?
[ And no, I'm not gay ]
That's funny but when I read L4 design papers it looks very similar to your description of QNX behaviour..
Care to describe where L4 did a mistake?
Other made a similar remark but you didn't answer, please backup your criticism with facts, otherwise you loose credibility..
>That kind of feedback is invaluable, and something a company like MicroSoft simply can't give, because they lack the necessary cross-platform experience to care.
.. PowerPC.
Microsoft lack cross-platform experience?
Strange, last I looked their OS have run on Alpha, MIPS (not sure about this one) for a long time and they are distributing a development kit for their next-gen X-Box console on
I don't like Microsoft either but the truth is that they have cross-platform experience.
Note that Linux works usually much better on x86 than on other architecture: much more distrib to choose from, more testers, etc..
Note that 'some believers' include popes and the like, usually those higher in the religion hierarchy are not saints..
Also I'm not American (that's why I confused father Bush and little Bush) so little Bush did nothing in my name.
As for Mother Theresa, boots@work summarized very well: she was a very good catholic, but I don't consider 'good catholic' as equivalent to good..
You can be a 'good catholic' and being good of course, but Mother Theresa wasn't..
>There are lots of religions that help other people and ask nothing from them in return.
As I said: name one religion in which believers have not killed other people in the name of the religion?
As for Mother Theresa and the orphans, sorry while helping orphans is good being against contraception in an overpopulated country is *very bad*!
She is a good example of how faith can make people support stupid/ilogical behaviour..
I hope that you realize that there isn't much difference between a "cult" and a mainstream religion: I don't know any religion which haven't been used as some time or another to kill other people..
In a semi-related topic, I heard that Bush has said that he doesn't consider atheists as citizens and yet he has been elected, two times!
How open and generous are some beleivers.. I wonder if we can infer from his two elections that his supporters share his so benevolent views against people who don't support their beliefs?
While theoretical advantages of LED are good, remember that (O)LED have not reached yet a good enough lifetime..
Plus to emit different color, the active material is different which will makes "interesting" color problems when the blue fades faster than the red..
> That being said, compiler technology isn't really taking advantage of this yet...
Are you sure? While programming a register allocator which works efficiently with 8 registers must be really hard, going from 8 to 16 is pretty easy, no?
Especially for multi-target compilers which generates also code for RISC and their big number of registers..
Now I agree that fine-tuning can be hard: if memory serves, the first 8 "legacy" register in x86-64 can be used with shorter instruction (1 byte smaller) so you can gain a little performance by ensuring that the most used registers are legacy registers..
Link?
The only part were I read him criticize PowerPC is MMU's handling which I don't really consider that it is part of the assembly language..
Granted FS are used by few people but it is one type of game which is CPU-bound currently, no need to invent new types of objects: flight dynamic, bullets, etc.. are enough!
:-)
In IL2 Sturmovick, IA's opponent use a different physic engine which enables them to do figures that you can't imitate, which is a bit frustrating..
But I'm not sure that if this accelerator allowed using realistic physic for IA's opponent, the IA would be good enough to fly a plane with this complex physic..
Ok, who is going to make an IA accelerator board?
On a more serious topic, I'd be interested to see what they're going to implement, while I don't really beleive that a standalone PPU will be a success, if the HW is not too different from a GPU, maybe a combined GPU-PPU would be interesting: a bunch of 'normal' FPU for doing vertex|physic process + some specialised unit dedicated to the pixel processing part.
The article give a link to an article describing a trick to implement coroutines in C, I found it quite enjoyable to read:m /corout ines.html
http://www.chiark.greenend.org.uk/~sgtatha
It is nice for a change to read about an "ugly trick" which is used to increase readability!
> As already mentioned games already do make use of the GPU and the CPU so we're fairly used to some multiprocessor concerns.
Correct, but of course, the higher the number of processor there is the more difficult it is to use them efficiently..
Also double core has a cost in CPU: each cores is smaller than a mono-CPU, if they are not well used, they could be slower than a single core..
As if games are CPU / GPU bound, it depends very much of the game: Prince of Persia is very different from a flight simulator and of course of the resolution..
Also you're right of course that games have several thread, but if your rendering thread use 100% of one core and the other threads use 10% of the other core, it is likely that the multi-core will be slower than a more powerfull single-core CPU..
I disagree for the latency part: multi-core, SMT CPU can somehow hide the latency: it can schedule another thread during the wait for the information.
Now of course, for this to work the implementation must allow the scheduling of another thread very fast and there must be threads schedulable with their workload present in the cache otherwise this just increase the pressure on memory.
> Just x86 forever and ever, Amen.
And? That's just the CPU, dammit! Who really cares about the instruction set which the CPU runs?
As for being glorified PCs, have you read the article?
Good LOM, no cable, Hot-swappable fan and power modules: it looks to me as a real server not a glorified PC!
OK, it is not a Stratus but still it is as good as previous Sparc servers from Sun..
You're right the SPEs are double precision capable as the last article showed it.
Now the SPUs don't really have double precision unit, they reuse the single precision unit: this saves a lot of gates as a poster said it below but also it reduces speed:
DP maximum theoretical FLOPS are 1/10 of SP FLOPS maximum theoretical..
Now granted DP maximum theoretical FLOPS should still be very high for a chip which will go into PlayStations.
I doubt that it is possible to do this: dual licensing *increase* the user possibilities not limit it.
As long as the GPL allows you to use the program without any consequence if you don't redistribute it, the other license doesn't matter..
While you're right about the difference, the truth is: when you boot a PC, 99.99% of the time the hardware is identical to what it was the last time.
So it should be possible to store an HW description into a small flash and boot *fast*, without having to discover hardware, and if the user want to add new critical "cold-plug" hardware (such as boot disk), he would just have to hit Del at the boot, and then a HW discover would be made..
Uh?
As long as you don't redistribute it, doing whatever you want with a GPL program is safe..