A5 (with integrated GPU and SoC) is 1.9 watts. With process improvement, Apple could probably come out with a 20 core ARM chip (at least 10 times faster then the current chip) that consumes less then 15 watts at full throttle and idles at less then 1 watt.
Bingo. What Intel and some others are not seeing is that it seems very, very likely ARM can scale up much easier than Intel scaling down.
With iPad and smartphones still on fire clearly the momentum for profit growth is on the "ARM scaling up" side of things rather than "Intel scaling down".
I wish this article was released a few weeks earlier before I got my new MBA 11". I've noticed that the battery drains too quickly. So today I ran my MBA side-by-side with my wife's late 2010 model, and yes, my battery drained much quicker than hers. In 10 minutes, mine drained 10% but hers only drained 4%. This is really disappointing performance for a mobile machine. So then I checked on Apple's website in the MBA section and noticed they no longer featured long battery life for their new model, so go figure...
I'm one month into my i7 MBA to replace my late-2010 Core2Duo MBA. The i7 CPU and Thunderbolt was just too much temptation for me.
I haven't done serious comparisons but I have noticed that my i7 MBA runs out of juice much quicker than my Core2Duo MBA did with the same amount of work. I just accepted that the higher-performing CPU's just use up more juice. The Core2Duo was a power-sipper though.
People were complaining about the Core2Duo being outdated technology (I loved it), and I hope people will not start complaining about reduced battery life now that the CPU's are current. It'll be a no-win situation.
I'm hoping that Apple determines that energy-management can be improved in OSX Lion via a software patch. Perhaps Lion v1.0 wasn't tweaked enough. I'll cross my fingers.
Just a few days ago an AppleInsider blog post (via RoughlyDrafted blog) said that OSX and iOS are already the same operating system so the rumors about its merging are nonsense. Now this AppleInsider post suggests that this rumor is not nonsense after all. So what's the truth then?
Unsatisfied with the power consumption levels of Intel's Atom platform, Apple instead turned to ARM for its iPhone and iPad processors.
Is there any evidence that Apple considered using an Atom processor in the iPhone? I very much doubt they did since ARM SoC designs give so many other advantages other than power consumption (i.e. size, cost, heat output, etc.).
A5 (with integrated GPU and SoC) is 1.9 watts. With process improvement, Apple could probably come out with a 20 core ARM chip (at least 10 times faster then the current chip) that consumes less then 15 watts at full throttle and idles at less then 1 watt.
Such a chip would be potentially great in server farms, at least those that could utilize 32bit CPUs, but it would be entirely pointless in the consumer market.
We're just not at a point where consumer software can use 20cores remotely effectively.
Is there any evidence that Apple considered using an Atom processor in the iPhone? I very much doubt they did since ARM SoC designs give so many other advantages other than power consumption (i.e. size, cost, heat output, etc.).
Minor quibble, you can't list heat output as a separate issue to power consumption.
For a processor Power Consumption = Heat Output, or as close as makes no difference.
Such a chip would be potentially great in server farms, at least those that could utilize 32bit CPUs, but it would be entirely pointless in the consumer market.
We're just not at a point where consumer software can use 20cores remotely effectively.
And we probably won't be there anytime soon, parallelism is a long story.
The compromise is that ARM processors scale differently. They don't have much scheduling logic so you need to add more cores to scale (instead of scheduling hardware). The advantage to ARM is that you have a lot more raw power in the same size chip. The disadvantage is that legacy software doesn't run well on multi-core processors. Apple has been addressing this in their SDKs though. In many situations ARM will be better for both mobile and desktop. Intel has had trouble scaling a single core processor too (they could only bring the scheduling hardware so far) and are now doing multiple cores even though many programs can't take advantage of them.
The difference is much greater than one section of the chip, even if that's one of the most important sections. At least to a much greater extent than ARM designers, the Intel architecture designers spent transistors to get performance. The worst of Intel's was probably Netburst, which stayed on top of the performance heap for a while, but it was very power inefficient. The ARM designers seemed to be pretty frugal with transistors, at least in the past, to keep their efficiency up. These are very different design priorities that greatly affect what they're good at and where they're used. ARM has only been running cut-down operating systems for decades because it can't compete with Intel on the high performance, and Intel just can't sip batteries very well as a result.
Too bad Apple didn't threaten Intel with a 'wake-up call' over crappy IGP performance, while they were at it.
Tell me about it. Bundlegate, I called it, and still call it. Shutting out Nvidia then eventually settling lawsuits and what not. Now Nvidia's just a shell of it's heyday when my username meant something.
After the horrible GMA950 and X3100, Apple held off using Intel GPUs long enough with the Nvidia 9400M and 320M. But now the big deal is the iPhone and iPad, and for Macs the latest Intel CPU/GPUs are passable enough to handle tasks at 15" or below. For everything else, there's now AMD/ATI that came storming back with their Radeon 4000/5000/6000s. Now suitably back in place in Macs.
I'm sure Apple has not been pleased with Intel IGPs for a long time. Apple held on to Core 2 considerably longer than many other manufacturers in part, I'm sure, so they could stick with the 9400M and 320M which were, remember, more than just IGPs...
And we probably won't be there anytime soon, parallelism is a long story.
But we are at a stage where a 4-core 64-bit processor can be well-utilised by modern programs. Since there is inherently so much multitasking even on a smartphone, in general use for anything tablet and above at least 2-3 cores will be put to full use when apps need the power.
I really see a possibility by end of 2013 for a 4-core 64-bit ARM with a 2-core GPU to drive a 11" OSX MacBook Air. The performance-per-watt will probably kill anything Intel can scale down. By this stage AMD will probably not even matter, though it could always play some sort of intermediary role in between ARM and Intel powered Apple products.
On a side note, at least for gaming PCs quad-core is minimum, though show me a game that actually uses all the cores effectively. It's lazy, hurried, or profit-driven development and poor console porting that results in a decent 2-core PC having weird lags... My theory is because it's all been developed on and targeted for people with quad-core PCs.
I really see a possibility by end of 2013 for a 4-core 64-bit ARM with a 2-core GPU to drive a 11" OSX MacBook Air. The performance-per-watt will probably kill anything Intel can scale down. By this stage AMD will probably not even matter, though it could always play some sort of intermediary role in between ARM and Intel powered Apple products.
Not likely. Apple just switched MBA to i5/i7. I can't think of a single time when Apple went backward in performance for a product. Even a quad core ARM won't be faster than the current MBA - and certainly not faster than what's available from Intel by 2013.
It is just barely possible that they would introduce a new product category (call it MacBook Air Light) that would use ARM, but I just don't see the value in throwing a new processor into Macs for just the low end product. Might as well throw in a low power AMD chip. The switch from PPC to Intel was different - they switched the entire Mac line. In addition, they were going from the non-standard product to the standard product which simplified coding and emulation/virtual machine stuff.
Too bad Apple didn't threaten Intel with a 'wake-up call' over crappy IGP performance, while they were at it.
Well, that might be a lost cause and Apple knows it. Intel just doesn't have the accumulated technical knowhow in the graphics area like ATI and nVidia, and there aren't another graphic chips firm left to buy either.
As far as low power chip goes, I wish Apple in the future would have a combined Intel, ARM chips configuration, use ARM chip to handle background process and system upkeep, and only fire up the Intel chip when real heavy crunching is needed. Or better yet, Intel license ARM and craft a few ARM cores on their chip to handle all low power processing needs.
GCD is a wonderful technology there is no doubt there. However there is a board spectrum of potential apps out there that can make use of GCD to one degree or another. If you are lucky an App might get close to x times the single core performance where x is the number of cores available. Generally these are apps that are blindingly easy to parallelize. Some are more difficult and there is vast ocean of apps where the speed up is more variable. As a tool, GCD and the associated API's, just give developers one approach to speeding up an app.
There are other approaches to parallelize an app. However the one thing that people often miss is that few Mac these days are used to do exactly one thing at a time and in fact the user isn't always aware of what is happening in background. For example right now I'm cruising the net responding with this post while my Mac is downloading video in background. This is actually light usage for me. So we have to consider that cores make processes run much smoother on UNIX platforms.
Programmers also really need to change how they wrap their heads around a problem to do task based parallelization. I understand what task based parallelization like GCD it does and how it works, but it is still a real bitch to overcome years of thinking about specific problems in a thread-based paradigm and understand how to correctly slice them up task-wise.
There are very few people on the planet that have the skill at writing really excellent* task based programs, there are quite a few like me who think they understand the paradigm but still have difficulty actually doing it really well. Until we get quite a few programmers that can actually do it really well, and then have the additional talent of being able to explain the secret sauce of how they frame their problem solving process to yield excellent code rather than excremental code, we will be stuck with lots of promise and a little progress.
*Please don't bring up task based graphics as an example of programmers getting it right. The graphics domain is where the idea came from because it is a ridiculously good fit and has been improved in the GPU hardware over almost 20 years before the concept jumped widely as an option in general software. This doesn't make task-based work well for general non-graphics programs automagically. Finding and assigning parallel tasks when you have 115million+ pixels to fill per second is low hanging fruit.
Not likely. Apple just switched MBA to i5/i7. I can't think of a single time when Apple went backward in performance for a product. Even a quad core ARM won't be faster than the current MBA - and certainly not faster than what's available from Intel by 2013.
It is just barely possible that they would introduce a new product category (call it MacBook Air Light) that would use ARM, but I just don't see the value in throwing a new processor into Macs for just the low end product. Might as well throw in a low power AMD chip. The switch from PPC to Intel was different - they switched the entire Mac line. In addition, they were going from the non-standard product to the standard product which simplified coding and emulation/virtual machine stuff.
Read esummers posts. I completely agree with both of you. Why not embed an inexpensive ARM chip to handle the lightweight stuff and have a low power Intel to get the heavy weight as needed? Maybe too much trouble but where will OSX be in two years? Who knows. Maybe the last cat name is tied to a bigger, new era of Mac computing.
If Apple can make their OS efficient "instruction" wise to a processor that is designed around the OS, they will eliminate the need for normal people caring what is actually in their computer. It will be like the iPad and iPhone. You know it has a ram, but Apple doesn't market how much.
I just hope I can still virtualize XP when all this happens.
Read esummers posts. I completely agree with both of you. Why not embed an inexpensive ARM chip to handle the lightweight stuff and have a low power Intel to get the heavy weight as needed? Maybe too much trouble but where will OSX be in two years? Who knows. Maybe the last cat name is tied to a bigger, new era of Mac computing.
Heterogenous processor systems do not have a happy history, remember the Risk PC? No? Exactly! A big part of what makes Apple tick is simplicity. Simple elegant UI designs, simple elegant hardware designs and whenever possible simple elegant internals.
What you're suggesting is the very opposite of a simple or elegant machine.
Intel's marketing sucks. Core2Duo, i5, i7...what does that tell the consumer? Nothing! it used to be so simple to tell which processor was "faster" and/or had other better qualities but then Intel decided they no longer wanted to advertise the clock speed. Unfortunately, AMD didn't jump on that opportunity.
This is a big factor in why PC sales are down. The average customer can't see the need to upgrade. It's used to be "Oh, look here, this is a 1Ghz processor with hyperthreading. That's way faster than my 300mhz. Maybe I need to upgrade." Compare that to what customer with a Core2Duo thinks when they see an i7 being advertised. Heck, Core2Duo sounds like it would be the better of the two. i7 sounds like a Ping golf club!
Heterogenous processor systems do not have a happy history, remember the Risk PC? No? Exactly! A big part of what makes Apple tick is simplicity. Simple elegant UI designs, simple elegant hardware designs and whenever possible simple elegant internals.
What you're suggesting is the very opposite of a simple or elegant machine.
I remember them and that was a long time ago. Would you say that OSX is simple as well? Probably not. Very complex OS that makes it seem simple. Time will tell. I could be wrong.
I would be surprised if Apple moved completely away from x86 chips that supports Windows. Let's face it, that alone helped them become more main stream by virtual support of Windows at near native speeds.
Comments
A5 (with integrated GPU and SoC) is 1.9 watts. With process improvement, Apple could probably come out with a 20 core ARM chip (at least 10 times faster then the current chip) that consumes less then 15 watts at full throttle and idles at less then 1 watt.
Bingo. What Intel and some others are not seeing is that it seems very, very likely ARM can scale up much easier than Intel scaling down.
With iPad and smartphones still on fire clearly the momentum for profit growth is on the "ARM scaling up" side of things rather than "Intel scaling down".
Maybe Apple did not learn from the PowerPC fiasco.
not comparable.
I wish this article was released a few weeks earlier before I got my new MBA 11". I've noticed that the battery drains too quickly. So today I ran my MBA side-by-side with my wife's late 2010 model, and yes, my battery drained much quicker than hers. In 10 minutes, mine drained 10% but hers only drained 4%. This is really disappointing performance for a mobile machine. So then I checked on Apple's website in the MBA section and noticed they no longer featured long battery life for their new model, so go figure...
I'm one month into my i7 MBA to replace my late-2010 Core2Duo MBA. The i7 CPU and Thunderbolt was just too much temptation for me.
I haven't done serious comparisons but I have noticed that my i7 MBA runs out of juice much quicker than my Core2Duo MBA did with the same amount of work. I just accepted that the higher-performing CPU's just use up more juice. The Core2Duo was a power-sipper though.
People were complaining about the Core2Duo being outdated technology (I loved it), and I hope people will not start complaining about reduced battery life now that the CPU's are current. It'll be a no-win situation.
I'm hoping that Apple determines that energy-management can be improved in OSX Lion via a software patch. Perhaps Lion v1.0 wasn't tweaked enough. I'll cross my fingers.
Unsatisfied with the power consumption levels of Intel's Atom platform, Apple instead turned to ARM for its iPhone and iPad processors.
Is there any evidence that Apple considered using an Atom processor in the iPhone? I very much doubt they did since ARM SoC designs give so many other advantages other than power consumption (i.e. size, cost, heat output, etc.).
A5 (with integrated GPU and SoC) is 1.9 watts. With process improvement, Apple could probably come out with a 20 core ARM chip (at least 10 times faster then the current chip) that consumes less then 15 watts at full throttle and idles at less then 1 watt.
Such a chip would be potentially great in server farms, at least those that could utilize 32bit CPUs, but it would be entirely pointless in the consumer market.
We're just not at a point where consumer software can use 20cores remotely effectively.
Is there any evidence that Apple considered using an Atom processor in the iPhone? I very much doubt they did since ARM SoC designs give so many other advantages other than power consumption (i.e. size, cost, heat output, etc.).
Minor quibble, you can't list heat output as a separate issue to power consumption.
For a processor Power Consumption = Heat Output, or as close as makes no difference.
I'm at at my computer amazed that the proof reading isn't a little better. C'mon guys!
UPDATE: You fixed it. Quick!
proofreading here isn't that good, but the site is bar none, the best apple hang out, to me it's become charming the occasional errors.
Such a chip would be potentially great in server farms, at least those that could utilize 32bit CPUs, but it would be entirely pointless in the consumer market.
We're just not at a point where consumer software can use 20cores remotely effectively.
And we probably won't be there anytime soon, parallelism is a long story.
The compromise is that ARM processors scale differently. They don't have much scheduling logic so you need to add more cores to scale (instead of scheduling hardware). The advantage to ARM is that you have a lot more raw power in the same size chip. The disadvantage is that legacy software doesn't run well on multi-core processors. Apple has been addressing this in their SDKs though. In many situations ARM will be better for both mobile and desktop. Intel has had trouble scaling a single core processor too (they could only bring the scheduling hardware so far) and are now doing multiple cores even though many programs can't take advantage of them.
The difference is much greater than one section of the chip, even if that's one of the most important sections. At least to a much greater extent than ARM designers, the Intel architecture designers spent transistors to get performance. The worst of Intel's was probably Netburst, which stayed on top of the performance heap for a while, but it was very power inefficient. The ARM designers seemed to be pretty frugal with transistors, at least in the past, to keep their efficiency up. These are very different design priorities that greatly affect what they're good at and where they're used. ARM has only been running cut-down operating systems for decades because it can't compete with Intel on the high performance, and Intel just can't sip batteries very well as a result.
Too bad Apple didn't threaten Intel with a 'wake-up call' over crappy IGP performance, while they were at it.
Tell me about it. Bundlegate, I called it, and still call it. Shutting out Nvidia then eventually settling lawsuits and what not. Now Nvidia's just a shell of it's heyday when my username meant something.
After the horrible GMA950 and X3100, Apple held off using Intel GPUs long enough with the Nvidia 9400M and 320M. But now the big deal is the iPhone and iPad, and for Macs the latest Intel CPU/GPUs are passable enough to handle tasks at 15" or below. For everything else, there's now AMD/ATI that came storming back with their Radeon 4000/5000/6000s. Now suitably back in place in Macs.
I'm sure Apple has not been pleased with Intel IGPs for a long time. Apple held on to Core 2 considerably longer than many other manufacturers in part, I'm sure, so they could stick with the 9400M and 320M which were, remember, more than just IGPs...
And we probably won't be there anytime soon, parallelism is a long story.
But we are at a stage where a 4-core 64-bit processor can be well-utilised by modern programs. Since there is inherently so much multitasking even on a smartphone, in general use for anything tablet and above at least 2-3 cores will be put to full use when apps need the power.
I really see a possibility by end of 2013 for a 4-core 64-bit ARM with a 2-core GPU to drive a 11" OSX MacBook Air. The performance-per-watt will probably kill anything Intel can scale down. By this stage AMD will probably not even matter, though it could always play some sort of intermediary role in between ARM and Intel powered Apple products.
On a side note, at least for gaming PCs quad-core is minimum, though show me a game that actually uses all the cores effectively. It's lazy, hurried, or profit-driven development and poor console porting that results in a decent 2-core PC having weird lags... My theory is because it's all been developed on and targeted for people with quad-core PCs.
I really see a possibility by end of 2013 for a 4-core 64-bit ARM with a 2-core GPU to drive a 11" OSX MacBook Air. The performance-per-watt will probably kill anything Intel can scale down. By this stage AMD will probably not even matter, though it could always play some sort of intermediary role in between ARM and Intel powered Apple products.
Not likely. Apple just switched MBA to i5/i7. I can't think of a single time when Apple went backward in performance for a product. Even a quad core ARM won't be faster than the current MBA - and certainly not faster than what's available from Intel by 2013.
It is just barely possible that they would introduce a new product category (call it MacBook Air Light) that would use ARM, but I just don't see the value in throwing a new processor into Macs for just the low end product. Might as well throw in a low power AMD chip. The switch from PPC to Intel was different - they switched the entire Mac line. In addition, they were going from the non-standard product to the standard product which simplified coding and emulation/virtual machine stuff.
Too bad Apple didn't threaten Intel with a 'wake-up call' over crappy IGP performance, while they were at it.
Well, that might be a lost cause and Apple knows it. Intel just doesn't have the accumulated technical knowhow in the graphics area like ATI and nVidia, and there aren't another graphic chips firm left to buy either.
As far as low power chip goes, I wish Apple in the future would have a combined Intel, ARM chips configuration, use ARM chip to handle background process and system upkeep, and only fire up the Intel chip when real heavy crunching is needed. Or better yet, Intel license ARM and craft a few ARM cores on their chip to handle all low power processing needs.
GCD is a wonderful technology there is no doubt there. However there is a board spectrum of potential apps out there that can make use of GCD to one degree or another. If you are lucky an App might get close to x times the single core performance where x is the number of cores available. Generally these are apps that are blindingly easy to parallelize. Some are more difficult and there is vast ocean of apps where the speed up is more variable. As a tool, GCD and the associated API's, just give developers one approach to speeding up an app.
There are other approaches to parallelize an app. However the one thing that people often miss is that few Mac these days are used to do exactly one thing at a time and in fact the user isn't always aware of what is happening in background. For example right now I'm cruising the net responding with this post while my Mac is downloading video in background. This is actually light usage for me. So we have to consider that cores make processes run much smoother on UNIX platforms.
Programmers also really need to change how they wrap their heads around a problem to do task based parallelization. I understand what task based parallelization like GCD it does and how it works, but it is still a real bitch to overcome years of thinking about specific problems in a thread-based paradigm and understand how to correctly slice them up task-wise.
There are very few people on the planet that have the skill at writing really excellent* task based programs, there are quite a few like me who think they understand the paradigm but still have difficulty actually doing it really well. Until we get quite a few programmers that can actually do it really well, and then have the additional talent of being able to explain the secret sauce of how they frame their problem solving process to yield excellent code rather than excremental code, we will be stuck with lots of promise and a little progress.
*Please don't bring up task based graphics as an example of programmers getting it right. The graphics domain is where the idea came from because it is a ridiculously good fit and has been improved in the GPU hardware over almost 20 years before the concept jumped widely as an option in general software. This doesn't make task-based work well for general non-graphics programs automagically. Finding and assigning parallel tasks when you have 115million+ pixels to fill per second is low hanging fruit.
Not likely. Apple just switched MBA to i5/i7. I can't think of a single time when Apple went backward in performance for a product. Even a quad core ARM won't be faster than the current MBA - and certainly not faster than what's available from Intel by 2013.
It is just barely possible that they would introduce a new product category (call it MacBook Air Light) that would use ARM, but I just don't see the value in throwing a new processor into Macs for just the low end product. Might as well throw in a low power AMD chip. The switch from PPC to Intel was different - they switched the entire Mac line. In addition, they were going from the non-standard product to the standard product which simplified coding and emulation/virtual machine stuff.
Read esummers posts. I completely agree with both of you. Why not embed an inexpensive ARM chip to handle the lightweight stuff and have a low power Intel to get the heavy weight as needed? Maybe too much trouble but where will OSX be in two years? Who knows. Maybe the last cat name is tied to a bigger, new era of Mac computing.
If Apple can make their OS efficient "instruction" wise to a processor that is designed around the OS, they will eliminate the need for normal people caring what is actually in their computer. It will be like the iPad and iPhone. You know it has a ram, but Apple doesn't market how much.
I just hope I can still virtualize XP when all this happens.
Read esummers posts. I completely agree with both of you. Why not embed an inexpensive ARM chip to handle the lightweight stuff and have a low power Intel to get the heavy weight as needed? Maybe too much trouble but where will OSX be in two years? Who knows. Maybe the last cat name is tied to a bigger, new era of Mac computing.
Heterogenous processor systems do not have a happy history, remember the Risk PC? No? Exactly! A big part of what makes Apple tick is simplicity. Simple elegant UI designs, simple elegant hardware designs and whenever possible simple elegant internals.
What you're suggesting is the very opposite of a simple or elegant machine.
This is a big factor in why PC sales are down. The average customer can't see the need to upgrade. It's used to be "Oh, look here, this is a 1Ghz processor with hyperthreading. That's way faster than my 300mhz. Maybe I need to upgrade." Compare that to what customer with a Core2Duo thinks when they see an i7 being advertised. Heck, Core2Duo sounds like it would be the better of the two. i7 sounds like a Ping golf club!
Heterogenous processor systems do not have a happy history, remember the Risk PC? No? Exactly! A big part of what makes Apple tick is simplicity. Simple elegant UI designs, simple elegant hardware designs and whenever possible simple elegant internals.
What you're suggesting is the very opposite of a simple or elegant machine.
I remember them and that was a long time ago. Would you say that OSX is simple as well? Probably not. Very complex OS that makes it seem simple. Time will tell. I could be wrong.
I would be surprised if Apple moved completely away from x86 chips that supports Windows. Let's face it, that alone helped them become more main stream by virtual support of Windows at near native speeds.