It's a bit more complex than this article...
on
Pocket Wars and Cores
·
· Score: 5, Interesting
...makes out
There are many, many makers of ARM based 'application processors' and the like: Texas Instruments, Samsung, Apple, ST Microelectronics, nVidia to name but a few. In addition, some people - like Qualcomm with their Snapdragon processor - have licensed the instruction set from ARM, but then have basically built their own core around that.
The nice thing about ARM is that - if you are looking to embed processing power - you can license a core (or two), design them into your own chip and then make it. Said chip can also include a USB controller, or a wireless baseband, or whatever. Intel will not sell you an x86 core for you to design into your own chip; ARM will.
Now: before this thread descends into meaningless ARM versus Intel rivalry, can I point out that the two architectures are optimized for entirely different situations. To say ARM is better than Intel, is like saying a bicycle is better than a ship - it's not a meaningful comparison. If you want to embed processing functionality, or you want low-power (particularly low standby power), then you need ARM. If you need raw processing power, optimised to run desktop or server operating systems, then you'll be wanting x86.
And the reason why x86 is so power hungry? It's because it's on big bits of silicon. And why's it on big bits of silicon? Because it support hyper-threading, out-of-order executon, has hardware virtualisation extensions, has extensive branch prediction, and tonnes of on chip cache.
There is no reason why ARM cannot offer all of these things too (and their Eagle design goes some way to do this). But if you want to do this, then your chip is going to get bigger, and more expensive, and more power hungry.
Over the next five years, we are going to continue to see mobility become more important: and that means more and more ARM cores, and a diminution of the importance of the traditional PC market. ARM has a very bright future - but, I suspect, it will probably have a great deal of trouble getting into the traditional PC space.
Ah-ha (replying to myself...), I see the A15 'Eagle' will have virtualization support. I wonder if they can keep it lean (and power sippling), while adding features.
Oh absolutely: ARM is a long-term winner at the expense of Intel (and others).
More and more computing tasks will happen on devices like tablets and phones, and these will be powered by ARM-cores, Android and iOS. Less and less will be done by Windows on Intel.
My point is that it is surprisingly difficult for ARM to penetrate the existing Intel desktop and server markets.
Well, the data center is an interesting one: right now a lot of admins are virtualizing lightly loaded servers to achieve just those power savings.
That's going to be a tough call on ARM. The reason Intel has such a lot of 'cruff' is because it's filled with specific optimizations for its target markets. In this case, virtualization. The reason ARM is beautiful and lean is because it lacks those optimizations. To get an ARM chip that's great at virtualization, you run the risk of ending up with a big piece of silicon - and then suddenly it's a very expensive solution.
I look forward to seeing the next generation ARM chips; I knew Sophie (nee Roger) Wilson and Roger Wilson when I was at Cambridge. I really hope they do well. But I have also seen a lot of very interesting chip designs (Transmeta anyone? Or DEC Alpha) fail in the desktop market because Intel's products (and the associated compilers and ecosystem) are actually very, very good.
ARM11 has around 85-88% branch prediction accuracy; the A8 is supposed to improve this to 95%, although I haven't seen any independent verification of that.
I'm very intrigued by Project Denver, although I suspect the main benefit to developers will be DMA between GPUs and the CPU.
ARM chips are improving enormously, and there's no doubt that the ARM9 instruction set is significantly more elegant than Intel.
However, I'd be a little cautious about assuming that Intel/x86 will be threatened any time soon.
I run Debian on a TI OMAP @ 800Mhz. It started as an experiment to see if I could transition my desktop to ARM. It ended as a VPN router sitting on my network (performing the extremely useful service of fooling certain US VOD sites as to my geographic location...).
ARM chips are highly optimized for one particular feature set: extremely low-power mobile computing. Intel chips are highly optimized for another: Windows/Linux on the desktop.
Almost all x86 has extremely sophisticated branch prediction to minimize calls to (slow) DRAM. ARM9 has pretty simple branch prediction. You will have far, far more cache misses on an ARM9 chip than on x86. So, to maintain performance for a given clock speed, you'll need to add on-die cache. Which starts getting pretty expensive. And the branch prediction on Intel is specifically geared around the way Windows (and to a lesser extent Linix) works. Unless Windows is completely re-written, or ARM ceases to be a low-cost chip, it's hard to see how ARM can offer equivilent performance at the same clock-speed as x86. For this reason (as well as the fact that ARM Windows will lack any kind of compatibility layer), I am pretty pessimistic about Windows on ARM - or indeed, ARM penetrating the desktop market.
And I am equally pessimistic about Intel succesfully getting into phones and tablets. When running at low loads, ARM chips are extraordinarily efficient. Intel has made a big fuss about its HUGI ("Hurry Up and Get Idle") efforts. But, of course, this is incredibly misleading. Most of the time an ARM core is doing something... just not very much. Will consumers accept a phone or tablet with 50% less battery life (or worse) for an Intel Inside logo? I think not.
Of course, ARM has another advantage (which is also, tangentially, a disadvantage). ARM does not make its own processors - it licenses its core designs to nVidia/Samsung/TI/Qualcomm/etc. This means that we can see an incredibly diversity of ARM-based products. Qualcomm can offer ARM cores with integrated 3G baseband. nVidia can add a couple of graphics processors, and call it Tegra 2. This means that ARM cores can be used in more applications, and more flexibly.
But it also means that ARM cores will be at least one line-width generation behind Intel. Intel has a very efficient design and *internally* build structure, with the best process technology in the industry. Which means 32nm Intel chips battle 42nm ARM ones. It was this process disadvantage that did for AMD, and it means that ARM will struggle against Intel in desktop. It is tough to compete on cost when someone else has a 50% higher transistor density for the same cost.
Wrapping up: ARM is fantastically well positioned for the fast growing tablet and smartphone markets; and Intel has a surprisingly defensible position in desktop/server chips.
Unfortunately, your data is wrong. The biggest Android maker is HTC (who account for 30-40% of the total Android market, although Samsung is catching up). Their bestselling phone in 2010 was the Desire. (Hardly a low-end product.)
In fact, HTC's low(er) end devices like the Wildfire, have sold relatively poorly. That doesn't mean there isn't a substantial low-end Android user-base: there obviously is, and ZTE and co. are keen to benefit from it. But currently, the high end is pretty much a wash (in terms of total unit numbers) between iPhone 4 on the one hand, and HTC Desire/Desire HD/Samsung Galaxy S/Droid 2.
And nuclear is perfect for baseload electricity generation. What it's not great for is peaking power. And normally - in the US - we use gas for that.
But here's the thing. PG&E's peak electricity load correlates almost exactly with when it is sunniest. (Because all that solar irradiation leads to aircons being turned on across Northern California.) There's actually a chart on it on some DoE paper; PG&E needs *twice* the generating capacity when the temperature is 110 degrees as when it is 60.
This is what solar is great for in sunny places. It provides expensive peaking power. (And if you think all electricity costs the same, well you don't know anything about the electricity industry.)
In my ideal world, we'd use nuclear and wind for baseload, with solar, biomass and gas for peaking power. But, hey, I'm not in charge.
The question is not "are there substantial untapped hydrocarbon reserves of the size of Ghawar?" but "are there substantial, accessible, economic, etc."
The Brazilian find is in the sub-salt region. It will cost tens, perhaps hundreds of billions before oil starts to flow. In Russia, Sakhalin is expensive and late. We've used up all the easy to get to oil, and now we've moved onto the much harder, often much lower quality oil.
Here, the oil seems to be good quality, but is trapped in long, shallow, relative unporous rock many miles below the surface. It will cost rather more than the $6/barrel to extract that we see in the Middle East. We will find oil to power our cars.
Yes: there are certain goods where demand is greater at higher prices (think perfume in the luxury area); similarly, you may need to spend more on necessities (such as basic starches) if their prices rises.
Neither of these is applicable to Microsoft Vista. This is a simple case of a monopolist forgetting that - even if you are the only supplier of a good - there is still a price elasticity curve. And at $400 a copy of Vista, Microsoft is leaving a lot of money on the table. I suspect that, at $100, MSFT would persuade a great many more users to 'upgrade'.
It's like talking to a broken record: I repeat; Open sourcing is the equivalent of selling the source for zero dollars. Whether you can open source and/or whether you can on-sell all depend on the wording of the parent licenses. A blanket pronouncement like "can't open source it" is nonsense equivalent to a blanket pronouncement saying "can't sell it".
Well, I suspect that Microsoft has a clause in the agreement which allows resale, but does not allow opening it up to the community. Selling for zero dollars != Open-sourcing.
"Actually most of that 100bn is worthless because it is mostly in mortgages"...
Whoah there!
Worthless? Let's take some assumptions. Assume 20% of Northern Rock borrowers default (10x historic levels).
Then assume they have loans at an average loan-to-value of 90% (which it isn't). And assume that house prices have fallen 30% (which they haven't). So - approximately - Northern Rock would lose one-in-five pounds lent to those who defaulted. Or, a 4% total loss on assets. (Which would be 5x the greatest loss a Swedish bank managed when their house prices dropped 25% in the early 90s.)
Fertilizer is mostly nitrogen - which makes up a good three quarters of the air we breathe.
You 'make' nitrogen with energy, using the Haber-Bosch process. When oil is the cheapest way to acquire energy (as used to be the case) people will make it with oil.
Nowadays, most nitrogen-based fertilizer uses natural gas as its power source. Still a fossil fuel, but not in quite such short supply as oil. If the price of energy continues to rise, then (a) fertilizer and agricultural products will become more expensive, and (b) nuclear, solar and wind will become more common sources of energy for the Haber-Bosch process.
It's surely a simple one to work out: are food prices in first world countries without subsidies higher or lower? Are they more volatile? And do people starve in years when food is less plentiful?
Now, the best examples I can think if are countries with basically no agricultural production at all (Hong Kong, Singapore), but in both these cases the populations are well fed - and on cheap food too.
You could make the case that the US is not like HK or Singapore. But it is like Australia - which is a first world country with a similar % of GDP in agriculture to the US. And Australia has, to all intents and purposes, no agricultural subsidies. Are food prices higher than the US? No. Do people starve when the rains fail? No. Are prices more volatile? I don't know.
Now, if the US wants to be a superpower, maybe there is a case for subsiding a minimal level of food production. But the GP makes a good point - why should farmers be treated any differently to any other business or industry? Why should they receive your cash to the tune of $16bn a year?
"There is actually no empirical evidence that the carbon we emit is doing any damage"
May I turn the question round? The government needs to spend money on schools, prisons, hospitals, wars in the Middle East, etc. It needs to get money from somewhere through tax. If you charge somebody for their behaviour (i.e. if you tax it) - whether emitting carbon, smoking or working - then you discourage it.
Surely it is better that we discourage the emission of carbon into the atmosphere than we discourage working?
Well, there is Guido's definition of Python 2.4, and Guido's definition of Python 2.6. But there is no definitive "This is Python". Broadly, Guido leads, and everyone else follows.
"Why are IronPython, PyPy, Jython etc trying to be compatible with cPython?" - errr, because the more libraries, etc., are interoperable the more likely people are to use these products when they are appropriate. (Obviously, IronPython is great if you want to use Silverlight, and Jython is wonderful if you want to run Python on your JVM.)
Slashdot Mirror
...makes out
There are many, many makers of ARM based 'application processors' and the like: Texas Instruments, Samsung, Apple, ST Microelectronics, nVidia to name but a few. In addition, some people - like Qualcomm with their Snapdragon processor - have licensed the instruction set from ARM, but then have basically built their own core around that.
The nice thing about ARM is that - if you are looking to embed processing power - you can license a core (or two), design them into your own chip and then make it. Said chip can also include a USB controller, or a wireless baseband, or whatever. Intel will not sell you an x86 core for you to design into your own chip; ARM will.
Now: before this thread descends into meaningless ARM versus Intel rivalry, can I point out that the two architectures are optimized for entirely different situations. To say ARM is better than Intel, is like saying a bicycle is better than a ship - it's not a meaningful comparison. If you want to embed processing functionality, or you want low-power (particularly low standby power), then you need ARM. If you need raw processing power, optimised to run desktop or server operating systems, then you'll be wanting x86.
And the reason why x86 is so power hungry? It's because it's on big bits of silicon. And why's it on big bits of silicon? Because it support hyper-threading, out-of-order executon, has hardware virtualisation extensions, has extensive branch prediction, and tonnes of on chip cache.
There is no reason why ARM cannot offer all of these things too (and their Eagle design goes some way to do this). But if you want to do this, then your chip is going to get bigger, and more expensive, and more power hungry.
Over the next five years, we are going to continue to see mobility become more important: and that means more and more ARM cores, and a diminution of the importance of the traditional PC market. ARM has a very bright future - but, I suspect, it will probably have a great deal of trouble getting into the traditional PC space.
Ah-ha (replying to myself...), I see the A15 'Eagle' will have virtualization support. I wonder if they can keep it lean (and power sippling), while adding features.
If they can, things could get very interesting.
Oh absolutely: ARM is a long-term winner at the expense of Intel (and others).
More and more computing tasks will happen on devices like tablets and phones, and these will be powered by ARM-cores, Android and iOS. Less and less will be done by Windows on Intel.
My point is that it is surprisingly difficult for ARM to penetrate the existing Intel desktop and server markets.
Well, the data center is an interesting one: right now a lot of admins are virtualizing lightly loaded servers to achieve just those power savings.
That's going to be a tough call on ARM. The reason Intel has such a lot of 'cruff' is because it's filled with specific optimizations for its target markets. In this case, virtualization. The reason ARM is beautiful and lean is because it lacks those optimizations. To get an ARM chip that's great at virtualization, you run the risk of ending up with a big piece of silicon - and then suddenly it's a very expensive solution.
I look forward to seeing the next generation ARM chips; I knew Sophie (nee Roger) Wilson and Roger Wilson when I was at Cambridge. I really hope they do well. But I have also seen a lot of very interesting chip designs (Transmeta anyone? Or DEC Alpha) fail in the desktop market because Intel's products (and the associated compilers and ecosystem) are actually very, very good.
ARM11 has around 85-88% branch prediction accuracy; the A8 is supposed to improve this to 95%, although I haven't seen any independent verification of that.
I'm very intrigued by Project Denver, although I suspect the main benefit to developers will be DMA between GPUs and the CPU.
ARM chips are improving enormously, and there's no doubt that the ARM9 instruction set is significantly more elegant than Intel.
However, I'd be a little cautious about assuming that Intel/x86 will be threatened any time soon.
I run Debian on a TI OMAP @ 800Mhz. It started as an experiment to see if I could transition my desktop to ARM. It ended as a VPN router sitting on my network (performing the extremely useful service of fooling certain US VOD sites as to my geographic location...).
ARM chips are highly optimized for one particular feature set: extremely low-power mobile computing.
Intel chips are highly optimized for another: Windows/Linux on the desktop.
Almost all x86 has extremely sophisticated branch prediction to minimize calls to (slow) DRAM. ARM9 has pretty simple branch prediction. You will have far, far more cache misses on an ARM9 chip than on x86. So, to maintain performance for a given clock speed, you'll need to add on-die cache. Which starts getting pretty expensive. And the branch prediction on Intel is specifically geared around the way Windows (and to a lesser extent Linix) works. Unless Windows is completely re-written, or ARM ceases to be a low-cost chip, it's hard to see how ARM can offer equivilent performance at the same clock-speed as x86. For this reason (as well as the fact that ARM Windows will lack any kind of compatibility layer), I am pretty pessimistic about Windows on ARM - or indeed, ARM penetrating the desktop market.
And I am equally pessimistic about Intel succesfully getting into phones and tablets. When running at low loads, ARM chips are extraordinarily efficient. Intel has made a big fuss about its HUGI ("Hurry Up and Get Idle") efforts. But, of course, this is incredibly misleading. Most of the time an ARM core is doing something... just not very much. Will consumers accept a phone or tablet with 50% less battery life (or worse) for an Intel Inside logo? I think not.
Of course, ARM has another advantage (which is also, tangentially, a disadvantage). ARM does not make its own processors - it licenses its core designs to nVidia/Samsung/TI/Qualcomm/etc. This means that we can see an incredibly diversity of ARM-based products. Qualcomm can offer ARM cores with integrated 3G baseband. nVidia can add a couple of graphics processors, and call it Tegra 2. This means that ARM cores can be used in more applications, and more flexibly.
But it also means that ARM cores will be at least one line-width generation behind Intel. Intel has a very efficient design and *internally* build structure, with the best process technology in the industry. Which means 32nm Intel chips battle 42nm ARM ones. It was this process disadvantage that did for AMD, and it means that ARM will struggle against Intel in desktop. It is tough to compete on cost when someone else has a 50% higher transistor density for the same cost.
Wrapping up: ARM is fantastically well positioned for the fast growing tablet and smartphone markets; and Intel has a surprisingly defensible position in desktop/server chips.
Unfortunately, your data is wrong. The biggest Android maker is HTC (who account for 30-40% of the total Android market, although Samsung is catching up). Their bestselling phone in 2010 was the Desire. (Hardly a low-end product.)
In fact, HTC's low(er) end devices like the Wildfire, have sold relatively poorly. That doesn't mean there isn't a substantial low-end Android user-base: there obviously is, and ZTE and co. are keen to benefit from it. But currently, the high end is pretty much a wash (in terms of total unit numbers) between iPhone 4 on the one hand, and HTC Desire/Desire HD/Samsung Galaxy S/Droid 2.
I've googled and can't find this car.
Source please?
And it would originally be a Napoleon quote.
The Criminal Justice Act.
You know, that thing that repealed Habeus Corpus, and abolished the Right to Silence.
That's true.
And nuclear is perfect for baseload electricity generation. What it's not great for is peaking power. And normally - in the US - we use gas for that.
But here's the thing. PG&E's peak electricity load correlates almost exactly with when it is sunniest. (Because all that solar irradiation leads to aircons being turned on across Northern California.) There's actually a chart on it on some DoE paper; PG&E needs *twice* the generating capacity when the temperature is 110 degrees as when it is 60.
This is what solar is great for in sunny places. It provides expensive peaking power. (And if you think all electricity costs the same, well you don't know anything about the electricity industry.)
In my ideal world, we'd use nuclear and wind for baseload, with solar, biomass and gas for peaking power. But, hey, I'm not in charge.
The question is not "are there substantial untapped hydrocarbon reserves of the size of Ghawar?" but "are there substantial, accessible, economic, etc."
The Brazilian find is in the sub-salt region. It will cost tens, perhaps hundreds of billions before oil starts to flow. In Russia, Sakhalin is expensive and late. We've used up all the easy to get to oil, and now we've moved onto the much harder, often much lower quality oil.
Here, the oil seems to be good quality, but is trapped in long, shallow, relative unporous rock many miles below the surface. It will cost rather more than the $6/barrel to extract that we see in the Middle East. We will find oil to power our cars.
But it may be at $10/barrel or more.
Yes: there are certain goods where demand is greater at higher prices (think perfume in the luxury area); similarly, you may need to spend more on necessities (such as basic starches) if their prices rises.
Neither of these is applicable to Microsoft Vista. This is a simple case of a monopolist forgetting that - even if you are the only supplier of a good - there is still a price elasticity curve. And at $400 a copy of Vista, Microsoft is leaving a lot of money on the table. I suspect that, at $100, MSFT would persuade a great many more users to 'upgrade'.
Whoosh!
(It was a great comment tho')
It's like talking to a broken record: I repeat; Open sourcing is the equivalent of selling the source for zero dollars. Whether you can open source and/or whether you can on-sell all depend on the wording of the parent licenses. A blanket pronouncement like "can't open source it" is nonsense equivalent to a blanket pronouncement saying "can't sell it".
Well, I suspect that Microsoft has a clause in the agreement which allows resale, but does not allow opening it up to the community. Selling for zero dollars != Open-sourcing.
"Actually most of that 100bn is worthless because it is mostly in mortgages"...
Whoah there!
Worthless? Let's take some assumptions. Assume 20% of Northern Rock borrowers default (10x historic levels).
Then assume they have loans at an average loan-to-value of 90% (which it isn't). And assume that house prices have fallen 30% (which they haven't). So - approximately - Northern Rock would lose one-in-five pounds lent to those who defaulted. Or, a 4% total loss on assets. (Which would be 5x the greatest loss a Swedish bank managed when their house prices dropped 25% in the early 90s.)
So, the 100bn would be worth 96bn.
Hardly worthless...
Fertilizer is mostly nitrogen - which makes up a good three quarters of the air we breathe.
You 'make' nitrogen with energy, using the Haber-Bosch process. When oil is the cheapest way to acquire energy (as used to be the case) people will make it with oil.
Nowadays, most nitrogen-based fertilizer uses natural gas as its power source. Still a fossil fuel, but not in quite such short supply as oil. If the price of energy continues to rise, then (a) fertilizer and agricultural products will become more expensive, and (b) nuclear, solar and wind will become more common sources of energy for the Haber-Bosch process.
So, fertilizer is not mostly petroleum based.
Can we have a fact based argument please?
It's surely a simple one to work out: are food prices in first world countries without subsidies higher or lower? Are they more volatile? And do people starve in years when food is less plentiful?
Now, the best examples I can think if are countries with basically no agricultural production at all (Hong Kong, Singapore), but in both these cases the populations are well fed - and on cheap food too.
You could make the case that the US is not like HK or Singapore. But it is like Australia - which is a first world country with a similar % of GDP in agriculture to the US. And Australia has, to all intents and purposes, no agricultural subsidies. Are food prices higher than the US? No. Do people starve when the rains fail? No. Are prices more volatile? I don't know.
Now, if the US wants to be a superpower, maybe there is a case for subsiding a minimal level of food production. But the GP makes a good point - why should farmers be treated any differently to any other business or industry? Why should they receive your cash to the tune of $16bn a year?
To quote Churchill at you: "democracy is the worst form of government except all the others that have been tried."
Just my 2c
So maintenance "average $126 million per reactor per year"?
BANG! Made up number alert. Source please?
Actually, trivially false. Lets Google shall we...
EDF in France has 59 nuclear power plants, some of which have more than one reactor. But lets just take the 59 number for now.
59 reactors * $126m = c. $7.5bn
EDF annual maintenance bill - slightly more than E1bn. And that includes all its non-nuclear and distribution assets too.
So, you're out by somewhere between five and ten times.
"There is actually no empirical evidence that the carbon we emit is doing any damage"
May I turn the question round? The government needs to spend money on schools, prisons, hospitals, wars in the Middle East, etc. It needs to get money from somewhere through tax. If you charge somebody for their behaviour (i.e. if you tax it) - whether emitting carbon, smoking or working - then you discourage it.
Surely it is better that we discourage the emission of carbon into the atmosphere than we discourage working?
Frankly, this looks like the nearest thing to Lotus Improv...
You don't sign anything when you buy the iPhone. Seriously: just the credit card slip. And you could always pay cash if you so desired.
Well, there is Guido's definition of Python 2.4, and Guido's definition of Python 2.6. But there is no definitive "This is Python". Broadly, Guido leads, and everyone else follows.
"Why are IronPython, PyPy, Jython etc trying to be compatible with cPython?" - errr, because the more libraries, etc., are interoperable the more likely people are to use these products when they are appropriate. (Obviously, IronPython is great if you want to use Silverlight, and Jython is wonderful if you want to run Python on your JVM.)
"Without carrier support, which it will never get, it will never be able to use any faster data connection than GPRS."
Errr, no. There's no reason it can't have a standard 3G chipset, as - in fact - most of the new phones from HTC have.