I usually consider human multitasking as "begin able to perform two precise tasks simultaneously in less time than it would take to do one and then the other." Or for the 'continuous' tasks it would be without loss of accuracy.
Clearly, I can not talk and watch a movie. And clealry my wife can not browse the web and watch tv.
There are traces of single core cortex A9 on arm's website. But I don't know if there is one FPU or 2 FPUs. But I think there are two.
The scenario here is not a practical one, it is LinPack which is a quite dense computation. Many architecture reaches 90% of peak performance on linpack.
Also you are talking about double precision computation, I am not sure linpack is double precision, it might be single precision. (actually wikipedia confirms it is double precision http://en.wikipedia.org/wiki/LINPACK_benchmarks)
Some of these architectures (not sure about cortex a9) have a separate pipeline for the FPU and the ALU, allowing to perform integer operations and jumping logic in parallel to the floating point operations.
You are right caches are important. I quoted theoretical peak. But here we are talking about dense linear algebra kernels. Matrix vector multiply for instance is a O(n^3) algorithm on O(n^2) memory. So overall you do more computation than memory transfer. With good caching techniques, tiling, software prefetching, you might be able to correct for it.
That's what these benchmarking techniques are designed to do. find the optimal blocking at every cache level and register blocking and software prefetching policy.
GP said: "A Cray 2 can pump out 1.9GFLOPS, an ipad2 can put out about 300-400Mflops. About as much as a early PentiumPro"
I am not sure a cray 2 can pump 1.9GFlop/s but an ipad 2 can do more than 300MFlop/s. The processor is clocked a 1 Ghz. It is dual core and it supports NEON vectorial instruction that can do multiple single floating point instruction at a time. (I am not sure exactly how many, I'd need to go to the actual spec, there might be fused multipy-add.) So I am pretty sure the theoretical peak performance is over 2GFlop/s. Probably around 4GFlop/s.
After that, I am not sure you can run linpackwith that speed. But 300MFlop/s seems quite low.
Moreover, I know Jack Dongarra: I trust his experiments.
Anybody can learn the basics of all domain (except obvious physical impairment). Everybody can learn how to draw, everybody can learn how to dance, everybody can learn how to speak an other language, everybody can learn maths, everybody can learn how to program. The amount of time required to learn depends on motivation and natural skills.
Not everybody can become exceptional in a particular field. But everything can be taught at a college level to anybody.
Provided the linux kernel is used in android and that intel tries so much to enter the android market, I highly doubt that their new power efficient chip wont work with linux. It would mean they won't ship it in android as well.
of course there is a lower bound to the energy you need. But low energy enough that body heat or ambient temperature powers it is close to zero enough for me. That is what they are talking about: so low that the actual value does not matter.
I am sure there is bottom limit that we will not be able to pass. But how low is that limit actually? We are pushing the efficiency of all our technology way down. in recent screen technologies: LCD, LED, e-ink. Recent storage technology: flash, SSD, NVRAM.
Making smartglasses flash with a raspberry pi, the glasses are tainted with a e-ink type of display (is that possible on glasses? I don't know the technology enough), the storage is on a SD card, and the input control is a microphone. That could in total cost less than 10W. Ok there is no networking included in that, I don't know much about lower power network interfaces, but bluetooth is cheap and you have a phone in your pocket (if you have fancy glasses, you most likely do have a fancy phone in your pocket).
We are not that at "no measurable power footprint", but we are getting pretty low.
I reply to this comment because I want you to get the message and I am not sure how to reach you otherwise.
Let me tell you that I am really fond of what you are doing there. I am really interested in seeing "open source" computing hardware becoming a reality. I am a software guy (on the theory side) so I am not sure I can help you. But I really like what you do. Please keep us (slashdot) posted on the development of this hardware.
I'll probably get one when it is released just for fun. (I just got a raspberry pi today for this reason.)These are good toys to make fun code porting/testing projects. Depending on the performance and network capabilities, I might be interested in studying a small cluster of these. It could be fun!
In brief, keep up the good work, that sounds really good!
It means you could include a pi in every single screen and keyboard in the world without too much of overhead. Computing power is really cheap nowadays!
See my other post on the topic. I think my main point is that, if in a couple you do not try to understand what the other like and why s/he does. Then what the hell are you doing together?
There are multiple aspects to "fashion". The most important one is personal. Selecting the right clothe is part of a form of discipline: I will shave even if I'll stay home. Similarly, she will pick the right clothe. Some people are not ready to go out/do something before they mediate, before they have coffee. It is somewhat the same. Part of it is artistic and it is difficult for me to fully understand it because I see clothe as "tools" and not as "art". Part of it is social, both "inward" and "outward". Proper clothing places you in a given group because of what you wear. It is from your perspective (inward) and from the perspective of other people (outward). You can see that in a similar way to people that read the news not because they are interested, but because it is considered socially important to keep in touch with the outside world.
That is completely orthogonal to the question. I do not give shit about dress code, pretty purses and fine jewelery. But I do understand why my significant cares about it and thinks it is important. She explained to me why she cares about it. It is important in a couple to be able to understand why she likes something. And it is important that she understand why I like something. I do not expect her to care about RPGs, programming and video games. It would be fine if she did not understand, but it is so much better that she actually does.
Actually, KAAPI (I distributed memory workstealing programming framework) was ported to Iphone/Ipod touch a few years ago. It still appears on their website http://moais.imag.fr/membres/thierry.gautier/TG/Kaapi.html Not sure if it still run...
Any metric will be good for me. If you like better number of HTTP request per watt, I am fine with that. The performance will highly depend on the application anyway. Without actual numbers it is difficult to know if it is interesting or not.
I can understand that. But do you ACTUALLY get 50% of the computing power for 25% of the electric power? You still need disk running, memory. Less computing power means you might need to increase the number of nodes. Which means more network equipement, fan,...
Is it really worth it? Note that it is a real question, it is not rhetorical.
The main question is how much GFlop per watt you get out of it, or the number of transactions per watt. Saying it is ARM so it is energy efficient is as stupid as saying it is pink so it is pretty.
Some application are best processed (energy wise) by using a kick ass power hungry GPU. Who cares if you consume a lot of electricity if you have a tremendous throughput?
OTOH, I live in columbus OH. I have been using tmobile for years and I have perfect coverage and 3g data speed. Oh, I also tether from time to time without any issues. They are the cheapest provider that can give me that type of coverage. They don't carry iPhone. So what? With the price difference on the contract I could buy one if I was interested.
Slashdot Mirror
I usually consider human multitasking as "begin able to perform two precise tasks simultaneously in less time than it would take to do one and then the other." Or for the 'continuous' tasks it would be without loss of accuracy.
Clearly, I can not talk and watch a movie. And clealry my wife can not browse the web and watch tv.
There are traces of single core cortex A9 on arm's website. But I don't know if there is one FPU or 2 FPUs. But I think there are two.
The scenario here is not a practical one, it is LinPack which is a quite dense computation. Many architecture reaches 90% of peak performance on linpack.
Also you are talking about double precision computation, I am not sure linpack is double precision, it might be single precision. (actually wikipedia confirms it is double precision http://en.wikipedia.org/wiki/LINPACK_benchmarks)
Some of these architectures (not sure about cortex a9) have a separate pipeline for the FPU and the ALU, allowing to perform integer operations and jumping logic in parallel to the floating point operations.
Finally, as you said, cortex A9 can perform one 64 bit FMAC in a cycle. FMAC is fused add multiply, that is to say out = in1 + in2 * in3, so there are 2 Floating point operations performed. ( http://infocenter.arm.com/help/topic/com.arm.doc.dui0068b/Bcfbbjgf.html )
In brief, if there is a single FPU, thanks to fused add multiply, peak performance is 2GFlop/s. If there are 2 FPUs, it is 4GFlop/s.
You are right caches are important. I quoted theoretical peak. But here we are talking about dense linear algebra kernels. Matrix vector multiply for instance is a O(n^3) algorithm on O(n^2) memory. So overall you do more computation than memory transfer. With good caching techniques, tiling, software prefetching, you might be able to correct for it.
That's what these benchmarking techniques are designed to do. find the optimal blocking at every cache level and register blocking and software prefetching policy.
GP said: "A Cray 2 can pump out 1.9GFLOPS, an ipad2 can put out about 300-400Mflops. About as much as a early PentiumPro"
I am not sure a cray 2 can pump 1.9GFlop/s but an ipad 2 can do more than 300MFlop/s. The processor is clocked a 1 Ghz. It is dual core and it supports NEON vectorial instruction that can do multiple single floating point instruction at a time. (I am not sure exactly how many, I'd need to go to the actual spec, there might be fused multipy-add.) So I am pretty sure the theoretical peak performance is over 2GFlop/s. Probably around 4GFlop/s.
After that, I am not sure you can run linpackwith that speed. But 300MFlop/s seems quite low.
Moreover, I know Jack Dongarra: I trust his experiments.
Anybody can learn the basics of all domain (except obvious physical impairment). Everybody can learn how to draw, everybody can learn how to dance, everybody can learn how to speak an other language, everybody can learn maths, everybody can learn how to program. The amount of time required to learn depends on motivation and natural skills.
Not everybody can become exceptional in a particular field. But everything can be taught at a college level to anybody.
last time I checked, the main reason was the compiler. They are not compiling the FF release for windows with gcc. But most distro use gcc.
Provided the linux kernel is used in android and that intel tries so much to enter the android market, I highly doubt that their new power efficient chip wont work with linux. It would mean they won't ship it in android as well.
of course there is a lower bound to the energy you need. But low energy enough that body heat or ambient temperature powers it is close to zero enough for me. That is what they are talking about: so low that the actual value does not matter.
I am sure there is bottom limit that we will not be able to pass. But how low is that limit actually? We are pushing the efficiency of all our technology way down. in recent screen technologies: LCD, LED, e-ink. Recent storage technology: flash, SSD, NVRAM.
Making smartglasses flash with a raspberry pi, the glasses are tainted with a e-ink type of display (is that possible on glasses? I don't know the technology enough), the storage is on a SD card, and the input control is a microphone. That could in total cost less than 10W. Ok there is no networking included in that, I don't know much about lower power network interfaces, but bluetooth is cheap and you have a phone in your pocket (if you have fancy glasses, you most likely do have a fancy phone in your pocket).
We are not that at "no measurable power footprint", but we are getting pretty low.
I just read the story in a french newspaper. They are in divorce.
Also they rewrote a linked list data structure out of need (or what they thought was a need) and not out of "hey let's write our own!"
Hi,
I reply to this comment because I want you to get the message and I am not sure how to reach you otherwise.
Let me tell you that I am really fond of what you are doing there. I am really interested in seeing "open source" computing hardware becoming a reality. I am a software guy (on the theory side) so I am not sure I can help you. But I really like what you do. Please keep us (slashdot) posted on the development of this hardware.
I'll probably get one when it is released just for fun. (I just got a raspberry pi today for this reason.)These are good toys to make fun code porting/testing projects. Depending on the performance and network capabilities, I might be interested in studying a small cluster of these. It could be fun!
In brief, keep up the good work, that sounds really good!
It means you could include a pi in every single screen and keyboard in the world without too much of overhead. Computing power is really cheap nowadays!
See my other post on the topic. I think my main point is that, if in a couple you do not try to understand what the other like and why s/he does. Then what the hell are you doing together?
Thank you. That was not a typo. As a non native speaker I make consistent silly mistakes such as this one.
There are multiple aspects to "fashion".
The most important one is personal. Selecting the right clothe is part of a form of discipline: I will shave even if I'll stay home. Similarly, she will pick the right clothe. Some people are not ready to go out/do something before they mediate, before they have coffee. It is somewhat the same.
Part of it is artistic and it is difficult for me to fully understand it because I see clothe as "tools" and not as "art".
Part of it is social, both "inward" and "outward". Proper clothing places you in a given group because of what you wear. It is from your perspective (inward) and from the perspective of other people (outward). You can see that in a similar way to people that read the news not because they are interested, but because it is considered socially important to keep in touch with the outside world.
That is completely orthogonal to the question. I do not give shit about dress code, pretty purses and fine jewelery. But I do understand why my significant cares about it and thinks it is important. She explained to me why she cares about it. It is important in a couple to be able to understand why she likes something. And it is important that she understand why I like something. I do not expect her to care about RPGs, programming and video games. It would be fine if she did not understand, but it is so much better that she actually does.
When I drink corona, I do not crunch data too well. Unless it involves a deck of cards!
Actually, KAAPI (I distributed memory workstealing programming framework) was ported to Iphone/Ipod touch a few years ago. It still appears on their website http://moais.imag.fr/membres/thierry.gautier/TG/Kaapi.html
Not sure if it still run...
Any metric will be good for me. If you like better number of HTTP request per watt, I am fine with that. The performance will highly depend on the application anyway. Without actual numbers it is difficult to know if it is interesting or not.
I can understand that. But do you ACTUALLY get 50% of the computing power for 25% of the electric power? You still need disk running, memory. Less computing power means you might need to increase the number of nodes. Which means more network equipement, fan, ...
Is it really worth it? Note that it is a real question, it is not rhetorical.
The main question is how much GFlop per watt you get out of it, or the number of transactions per watt. Saying it is ARM so it is energy efficient is as stupid as saying it is pink so it is pretty.
Some application are best processed (energy wise) by using a kick ass power hungry GPU. Who cares if you consume a lot of electricity if you have a tremendous throughput?
There was a good deal on wall street, so we had sold it. Strangely, we can not buy it anymore. Where is the factory that produce those?
which gives incentives to the company not to do shit that will get them involved.
OTOH, I live in columbus OH. I have been using tmobile for years and I have perfect coverage and 3g data speed. Oh, I also tether from time to time without any issues. They are the cheapest provider that can give me that type of coverage. They don't carry iPhone. So what? With the price difference on the contract I could buy one if I was interested.