The problem with this approach is that the energy used for computation is a relatively small part of the whole. Much more energy is spent on fetching instructions, decoding instructions, fetching data, predicting branches, managing caches and many other processes. And the addition of approximate arithmetic increases the area and leakage of the processor which increases engergy consumption for all programs.
Approximate computation is already widely used in media and numerical applications, but it is far from clear that it is a good idea to put approximate arithmetic circuits in a standard processor.
Android I think is only scheduled to get 64-bit support in late 2014 as well.
I suspect that Android will get 64-bit support when there is a phone that needs it, but it feels like that will be closer to the middle of 2014 rather than the end.
I'm sure Intel's worried - the latest Bay Trail Atoms are basically even with the A7 in performance.
And Bay Trail is on a newer 22nm process compared to A7's 28nm. I don't know the mind of Intel, but they have to be concerned that their "process advantage" is still not delivering concrete benefits.
Far too much is published already, and much of it just isn't very good.
But who decides what is "good"? The temptation is for scientists to publish the results that support their theory and reject the rest of the results as "bad" data, leading to massive selection bias.
Surely it is better to publish all the data so that others can check the conclusions that the author has drawn from it? And in the case of publicly-funded research it seems right that all the data should be made publicly available (with the obvious exception of sensitive information such as personal medical details).
The NHS works at base cost, how can base cost + profit margin be a cheaper alternative and more sustainable?
Because there is no such thing as "base cost", there is just "cost". What the NHS considers "cost" can still be more than what the alternative calls "cost + profit margin".
First, their Mac line will still run Intel CPUs, while the iOS line runs ARM based processors. You can't really merge the two for various reasons.
This is the weakest part of your argument. I agree that iOS and OSX will not be unified any time soon, but it is not really about the processor architecture. Apple have already created a 64-bit processor that matches Intel's equivalent design on a less favourable process, and they will always prefer a processor that they have made over one that they have to buy in. Running Intel apps on other processors with good performance is technical very feasible. There is no good reason why there won't be an OSX device containing an Apple processor within the next few years, in addition to a line of machines with Intel processors.
ENIAC was arguably the first general purpose electronic computer, and it was built for the US military (a wing of their government).
You are right, there was a lot of government funding at the start of the computing era. But most of the work that took us from the Antikythera Mechanism to the modern 28nm processor was privately funded.
My point is that you can't compare the development of modern computers, which was mostly a commercial engineering process, with the advancement of particle physics, which is mostly a tax-funded scientific process. The money that might have been "wasted" on colliders is a different kind of money from that used to create the i7.
64-bit ARM is [] not necessarily the best thought out instruction set The difference between ARM 32 and ARM 64 is far greater than the difference between X86 and X86_64.
Which parts of 64-bit ARM do you think are badly designed? I'm sure a lot of thought went in to it!!
But you are right about the differences. There are things in 32-bit ARM that seemed like a good idea in the 90s but now just make life complicated, so they are missing from 64-bit ARM.
Are you seriously suggesting that Apple migrates their desktop machines to hardware that's about 10 years behind the curve in terms of performance when compared to x86?
The processor in the iPhone 5 was designed by Apple. So the question is whether Apple think they can design a better processor than Intel or ARM. And since this is Apple we are talking about, they almost certainly think that they can.
This is not a transition from one third-party processor to another: It is a transition from a third-party processor to a processor that is 100% controlled by Apple.
I seriously doubt that ARM itself is going to develop the 32-bit platform much beyond where it is today.
Do you seriously think your washing machine needs a 64 bit processor?
So they'll just use MIPS instead? Or older ARM chips?
ARM sells M-class processors for embedded applications and R-class processors for real-time applications like disc drives and automotive. These are active product lines, not older designs. These compete with MIPS, ARC, Intel Quark and a host of 8-bit and 16-bit devices.
64-bit processor will consume more power than an equivalent 32-bit version, so if the extra abilities aren't being used then it's a waste. And the extra abilities are almost all about memory addressing ranges.
The extra abilities include >2x integer registers, 2x FP registers, 2x SIMD registers, new SIMD instructions including AES (one area where Intel smashes ARM performance) and a number of archicture improvements that make it significantly more efficient to implement.
Similar issues apply to AMD64 vs x86. 64-bit code is generally faster than 32-bit code and faster code saves energy, as does keeping more data in registers.
Power numbers depend on implementation, but 64-bit code is likely to use less power than 32-bit code on a 32-bit-only device. And since much of the power is spent in the OS it is likely that adding 64-bit reduces overall power rather than increasing it even with 32-bit apps.
ARM will sputter out while trying to move up the CPU segment long before Intel will sputter out while trying to move down it.
To pretend it can just easily do what Intel has been doing for decades is pure hubris.
No more hubris than pretending that Intel can just easily do what ARM's partners have been doing for a long time. But I don't think either Intel or the ARM partners think this is going to be easy.
Like saying a compiler has a bug if it can't optimise away "if (1 != 2)".
A compiler does have a bug if it can't optimise away "if (1 != 2)". The following code contains exactly that expression and any decent compiler will optimise away the test and the code in the conditional block.
It seems that Graphene has even better thermal conductivity than diamond and there is a more realistic chance of getting that into the right shape for a heatsink.
There is an idea that popularising particle physics and astronomy encourages young people to be interested in careers in science. What it actually does is encourage young people to be interested in careers in particle physics and astronomy. The result is a glut of specialists in those particular areas while other disciplines are starved of good people. Those offering posts in other areas of physics find it hard to get good candidates in their field and sometimes have to hire and re-train specialists from these popular areas (if they are prepared to take the job in the first place).
Tracking is better the better that the AR marker can be identified.
In this case the marker is a tiny part of the image once you have moved to the other side of the room, so I imagine that they are tracking other parts of the scene as well, using the catalogue as an initial reference point.
We will know better once we see some real footage.
Not the same feature at all. The 2013 version just shows 3D models above the pages of the catalogue so that you can see what products look like or browse directly to web pages for more information on the product. The 2014 version shows the products "in situ" so that you can see what they would look like if you had them in your room. This requires significantly better 3D tracking to work properly, though the promotional video doesn't really show how well it works in practice.
The video is at least partially faked (the images do not move in sync with the devices, especially in the last shot with multiple devices) so I wonder how well it actually works?
I am extremely skeptical when the US government tries to shovel propaganda in my direction.
You should be skeptical of all propaganda, but it does tell you a lot about the country that produces it. Try reading some North Korean propaganda: http://www.kcna.co.jp/index-e.htm and tell me what sort of country produces something like that.
So while I am certain that life in North Korea is nothing like we are told, I douubt that it is significantly better than it appears to be from the outside.
Slashdot Mirror
The problem with this approach is that the energy used for computation is a relatively small part of the whole. Much more energy is spent on fetching instructions, decoding instructions, fetching data, predicting branches, managing caches and many other processes. And the addition of approximate arithmetic increases the area and leakage of the processor which increases engergy consumption for all programs.
Approximate computation is already widely used in media and numerical applications, but it is far from clear that it is a good idea to put approximate arithmetic circuits in a standard processor.
Android I think is only scheduled to get 64-bit support in late 2014 as well.
I suspect that Android will get 64-bit support when there is a phone that needs it, but it feels like that will be closer to the middle of 2014 rather than the end.
I'm sure Intel's worried - the latest Bay Trail Atoms are basically even with the A7 in performance.
And Bay Trail is on a newer 22nm process compared to A7's 28nm. I don't know the mind of Intel, but they have to be concerned that their "process advantage" is still not delivering concrete benefits.
Far too much is published already, and much of it just isn't very good.
But who decides what is "good"? The temptation is for scientists to publish the results that support their theory and reject the rest of the results as "bad" data, leading to massive selection bias.
Surely it is better to publish all the data so that others can check the conclusions that the author has drawn from it? And in the case of publicly-funded research it seems right that all the data should be made publicly available (with the obvious exception of sensitive information such as personal medical details).
The NHS works at base cost, how can base cost + profit margin be a cheaper alternative and more sustainable?
Because there is no such thing as "base cost", there is just "cost". What the NHS considers "cost" can still be more than what the alternative calls "cost + profit margin".
Imagination will continue to thrive as long as Apple thrives
I wouldn't bet on that, Apple have been building a GPU team for a while now.
First, their Mac line will still run Intel CPUs, while the iOS line runs ARM based processors. You can't really merge the two for various reasons.
This is the weakest part of your argument. I agree that iOS and OSX will not be unified any time soon, but it is not really about the processor architecture. Apple have already created a 64-bit processor that matches Intel's equivalent design on a less favourable process, and they will always prefer a processor that they have made over one that they have to buy in. Running Intel apps on other processors with good performance is technical very feasible. There is no good reason why there won't be an OSX device containing an Apple processor within the next few years, in addition to a line of machines with Intel processors.
Get your thoughts into paper and contact/join a party like UKIP
So UKIP is opposed to the European Convention on Human Rights?
Interesting...
ENIAC was arguably the first general purpose electronic computer, and it was built for the US military (a wing of their government).
You are right, there was a lot of government funding at the start of the computing era. But most of the work that took us from the Antikythera Mechanism to the modern 28nm processor was privately funded.
My point is that you can't compare the development of modern computers, which was mostly a commercial engineering process, with the advancement of particle physics, which is mostly a tax-funded scientific process. The money that might have been "wasted" on colliders is a different kind of money from that used to create the i7.
We clearly shouldn't have wasted all the money on mechanical and tube computers and just waited until we got i7s.
We didn't. Early computers were funded from commercial sources not taxation, and they had practical applications right from the start.
64-bit ARM is [] not necessarily the best thought out instruction set
The difference between ARM 32 and ARM 64 is far greater than the difference between X86 and X86_64.
Which parts of 64-bit ARM do you think are badly designed? I'm sure a lot of thought went in to it!!
But you are right about the differences. There are things in 32-bit ARM that seemed like a good idea in the 90s but now just make life complicated, so they are missing from 64-bit ARM.
Are you seriously suggesting that Apple migrates their desktop machines to hardware that's about 10 years behind the curve in terms of performance when compared to x86?
The processor in the iPhone 5 was designed by Apple. So the question is whether Apple think they can design a better processor than Intel or ARM. And since this is Apple we are talking about, they almost certainly think that they can.
This is not a transition from one third-party processor to another: It is a transition from a third-party processor to a processor that is 100% controlled by Apple.
It's more like they didn't have much else for the iPhone 5S, just the fingerprint sensor.
Designing a new 64-bit processor is a strategic investment, not something you do for a minor marketing advantage on a single product.
I seriously doubt that ARM itself is going to develop the 32-bit platform much beyond where it is today.
Do you seriously think your washing machine needs a 64 bit processor?
So they'll just use MIPS instead? Or older ARM chips?
ARM sells M-class processors for embedded applications and R-class processors for real-time applications like disc drives and automotive. These are active product lines, not older designs. These compete with MIPS, ARC, Intel Quark and a host of 8-bit and 16-bit devices.
64-bit processor will consume more power than an equivalent 32-bit version, so if the extra abilities aren't being used then it's a waste. And the extra abilities are almost all about memory addressing ranges.
The extra abilities include >2x integer registers, 2x FP registers, 2x SIMD registers, new SIMD instructions including AES (one area where Intel smashes ARM performance) and a number of archicture improvements that make it significantly more efficient to implement.
Similar issues apply to AMD64 vs x86. 64-bit code is generally faster than 32-bit code and faster code saves energy, as does keeping more data in registers.
Power numbers depend on implementation, but 64-bit code is likely to use less power than 32-bit code on a 32-bit-only device. And since much of the power is spent in the OS it is likely that adding 64-bit reduces overall power rather than increasing it even with 32-bit apps.
ARM [...] will face very touch competition from things like Moonshot
Moonshot supports both x86 and ARM processors.
ARM will sputter out while trying to move up the CPU segment long before Intel will sputter out while trying to move down it.
To pretend it can just easily do what Intel has been doing for decades is pure hubris.
No more hubris than pretending that Intel can just easily do what ARM's partners have been doing for a long time. But I don't think either Intel or the ARM partners think this is going to be easy.
Like saying a compiler has a bug if it can't optimise away "if (1 != 2)".
A compiler does have a bug if it can't optimise away "if (1 != 2)". The following code contains exactly that expression and any decent compiler will optimise away the test and the code in the conditional block.
#define DEBUG_SUMMARY 1
#define DEBUG_VERBOSE 2
.
.
#define DEBUG_MODE DEBUG_SUMMARY
.
.
if (DEBUG_MODE != DEBUG_VERBOSE) {
}
It seems that Graphene has even better thermal conductivity than diamond and there is a more realistic chance of getting that into the right shape for a heatsink.
http://en.wikipedia.org/wiki/Graphene#Thermal
There is an idea that popularising particle physics and astronomy encourages young people to be interested in careers in science. What it actually does is encourage young people to be interested in careers in particle physics and astronomy. The result is a glut of specialists in those particular areas while other disciplines are starved of good people. Those offering posts in other areas of physics find it hard to get good candidates in their field and sometimes have to hire and re-train specialists from these popular areas (if they are prepared to take the job in the first place).
Tracking is better the better that the AR marker can be identified.
In this case the marker is a tiny part of the image once you have moved to the other side of the room, so I imagine that they are tracking other parts of the scene as well, using the catalogue as an initial reference point.
We will know better once we see some real footage.
This year's catalog also had the same feature
Not the same feature at all. The 2013 version just shows 3D models above the pages of the catalogue so that you can see what products look like or browse directly to web pages for more information on the product. The 2014 version shows the products "in situ" so that you can see what they would look like if you had them in your room. This requires significantly better 3D tracking to work properly, though the promotional video doesn't really show how well it works in practice.
The video is at least partially faked (the images do not move in sync with the devices, especially in the last shot with multiple devices) so I wonder how well it actually works?
I doubt that the tracking is really that smooth.
Of course these applications are FPU based. Last I checked the latest ARM had the performance of a 1997 pentium pro in that area.
You need to check again. ARM cores have up to 50X the floating point performance of a Pentium Pro.
Try reading some North Korean propaganda
I can't because I don't understand Korean.
You make a valid point about translations provided by adversaries, but the link I gave was in English and provided directly by North Korea.
I am extremely skeptical when the US government tries to shovel propaganda in my direction.
You should be skeptical of all propaganda, but it does tell you a lot about the country that produces it. Try reading some North Korean propaganda: http://www.kcna.co.jp/index-e.htm and tell me what sort of country produces something like that.
So while I am certain that life in North Korea is nothing like we are told, I douubt that it is significantly better than it appears to be from the outside.