Here's hoping for a 7 nm SoC in the next version of the watch.
So...over time, the nanometer build process has gotten smaller by number each year. Being unfamiliar with what that is, what happens when the number is down to 1? Is anything smaller?
What happens after that is when the user boots up the iPad they immediately disappear down the nearest rabbit hole and is then able to pass through a keyhole into another world.
It looks like Apple's A series processors are going to outperform some of Intel's newest processors soon if not already. Particularly the Core M processors in the MacBook.
The scores don’t make sense. It gives a very low multicore processing efficiency compared with previous and current dual core processors from Apple. It would also mean a very high clock speed (2.8GHz) if Apple were to keep using the strategy of using the same cores in both iPad and iPhone SoC.
Id expect some variance once we have shipping processors. That is we could be seeing numbers form an engineering sample or an early pilot production chip.
As as for multi core efficiency aging I'd wait until real hardware ships but honestly I'm not that surprised. Apple would need to vastly I,prove its cache subsystem to get better multi ore performance. Caches however take lots of power and as such one has to balance performance against heat.
The high clock rate is a possibility but I could see Apple designing in capability within the cores that gets activated in a "X" variant. This wouldn't be unusual in the world of semi conductors. In apples case they might do this to manage power. On the other hand if Apple has up locked the chip and can continue to control power then I think it is fair to say TSMC is way ahead of Intel now.
Here's hoping for a 7 nm SoC in the next version of the watch.
So...over time, the nanometer build process has gotten smaller by number each year. Being unfamiliar with what that is, what happens when the number is down to 1? Is anything smaller?
I am not familiar with building process but I guess I read somewhere that 7 nm is top for now and going further will be very challenging with contemporary technologies.
Correct me whether it is not true.
So...over time, the nanometer build process has gotten smaller by number each year. Being unfamiliar with what that is, what happens when the number is down to 1? Is anything smaller?
Well you should consider that current 14nm chips could be thought of as 0.014um* chips, or 0.000014mm chips, or 0.0000000014m chips, or 0.0000000000014km chips.
So yeah, there is smaller than 1nm. Getting there will be extremely difficult. 7nm looks doable, 5nm should be possible with EUV fabrication and gate-around transistors, and TSMC has started early investigations of 3nm, which we would be look at around 2025. 2028 for 2nm, but that's got to beat all sorts of issues. 1nm? Well that will require massive investment.
More likely 3D fabrication will get popular instead, maybe 3D transistor structures, stacks of logic made in one go (rather than TSVs of separate dies)...
* Consider in the early 90s we were fabbing chips at 1um or thereabouts.
Here's hoping for a 7 nm SoC in the next version of the watch.
So...over time, the nanometer build process has gotten smaller by number each year. Being unfamiliar with what that is, what happens when the number is down to 1? Is anything smaller?
Transistors, like any switch, are made up of two conducting sides divided by a non-cunducting medium when the switch is off. To turn the switch on, you close the gap or otherwise cause the non-conducting part to conduct electrons from one end to the other.
The nanometer measurement in this case is how far apart the two conducting sides are in each transistor. Making that gap smaller allows you to cram more into the same amount of space which allows you to have a more sophisticated chip design. It also allows for a higher clock rate, but we hit the practical limits of that for other reasons a while ago.
Unfortunately, we will never reach 1nm. When the transistor gap becomes small enough, quantum tunneling starts happening. Basically, quantum tunneling allows electrons (and other particles) to jump small gaps even if that shouldn't be possible according to the potential energy field. It's been a while since I've taken a class on quantum mechanics, so I don't exactly remember why tunneling happens, but nonetheless it will start to become a problem very soon. When quantum effects start to take over, there is no longer a discernible difference between an open transistor and a closed one, so computational logic becomes impossible. At that point, we will have reached the limits of transistor based chip design, and we will have to change the fundamental way in which computers work in order to see more advancement.
After reading this gem, I am saddened by the fact that you only have 2 posts.
I think that could be great, but only if Apple really took the opportunity to push the *entire* Mac lineup well beyond what could have been done with Intel chips.
In particular, I'd love to see a Mac Pro that demolishes (in terms of performance) anything that Intel offers at the same price. The prices Intel charges for Xeons with more than 4 cores are so absurdly high that Apple really could do this. I'll bet they could produce an A# chip with 8 cores (no GPU, no ISP) that's smaller than an A5X and offers better multicore performance than a 6 core Kaby Lake, but for much less money. Put two of those in a Mac Pro and they'd have a very compelling unix workstation. Put one of them in an iMac.
The scores don’t make sense. It gives a very low multicore processing efficiency compared with previous and current dual core processors from Apple. It would also mean a very high clock speed (2.8GHz) if Apple were to keep using the strategy of using the same cores in both iPad and iPhone SoC.
Id expect some variance once we have shipping processors. That is we could be seeing numbers form an engineering sample or an early pilot production chip.
As as for multi core efficiency aging I'd wait until real hardware ships but honestly I'm not that surprised. Apple would need to vastly I,prove its cache subsystem to get better multi ore performance. Caches however take lots of power and as such one has to balance performance against heat.
The high clock rate is a possibility but I could see Apple designing in capability within the cores that gets activated in a "X" variant. This wouldn't be unusual in the world of semi conductors. In apples case they might do this to manage power. On the other hand if Apple has up locked the chip and can continue to control power then I think it is fair to say TSMC is way ahead of Intel now.
Engineering samples wouldn’t be able to run Geekbench, unless it was tested inside Apple, since preproduction machines don’t run iOS outside of Apple. And that is another reason why I always doubt when these numbers come out too early.
The scores don’t make sense. It gives a very low multicore processing efficiency compared with previous and current dual core processors from Apple. It would also mean a very high clock speed (2.8GHz) if Apple were to keep using the strategy of using the same cores in both iPad and iPhone SoC.
Keep in mind that this is geekbench 4, not geekbench 3. Geekbench 4 has much more demanding workloads that can trigger thermal throttling when multiple cores are active. So if your intuition about scaling with more cores is based on geekbench 3, your intuition could be off here.
Here's hoping for a 7 nm SoC in the next version of the watch.
So...over time, the nanometer build process has gotten smaller by number each year. Being unfamiliar with what that is, what happens when the number is down to 1? Is anything smaller?
It gets smaller every two to three years, by about a third in linear terms, which gives about a doubling of elements on the chip.
unfortunately, it doesn't look as though getting below 7 is going to be easy, if possible at all. The average size of the atoms used in chips is about .3 to .5 nanometer. We're getting awfully close to the limit on what we can do here.
Thank you jbdragon, iaeen2, and rs1919 for explaining how size matters in the function of the transistors in a CPU with respect to power consumption and performance.
I myself am amazed at the progress that has been made. So much in so short time that the designers and engineers are running into trouble with the laws of physics.
That being said, TSMC is ahead of all others in chip fabrication. As Intel and Samsung catch up as they will because the transistors are hitting the limit in terms of size and shrinkage, TSMC will be able to continue to "tweak" their facilities for greater yields and lower cost. Apple will have ongoing access to TSMC's most advanced developments and should be able to stay ahead of everyone else. They have essentially passed Intel.
I do see 12 nm FDSOI making a big splash as the power consumption is much lower than the latest FinFet designs with very good performance. The Chinese are betting big on the technology and companies that rely on software only are going to be in for a rude awakening as the Chinese leverage a hardware edge to take over the market. It is very likely the reason that Google is so keen to get into hardware. They better assemble a chip design team fast, because a day is coming when Huawei and Xiaomi sell phones that do not come with Google Play.
Apple will still be in a good spot and could always take advantage of Global Foundries 12 nm FDSOI process for the S series SIP in the watch.
In any case, Apple's mobile processors are the best in the business. No one else is even close.
Well, actually, Intel is ahead of everyone else. TSMC may be number two. Intel has the densest process, meaning that they can actually fit more transistors on the same size chip than anyone else. In the coming years, as we reach 7nm, and gain experience with it, that advantage for Intel will disappear. But that won't be for at least three years, and more likely four or five.
The scores don’t make sense. It gives a very low multicore processing efficiency compared with previous and current dual core processors from Apple. It would also mean a very high clock speed (2.8GHz) if Apple were to keep using the strategy of using the same cores in both iPad and iPhone SoC.
Id expect some variance once we have shipping processors. That is we could be seeing numbers form an engineering sample or an early pilot production chip.
As as for multi core efficiency aging I'd wait until real hardware ships but honestly I'm not that surprised. Apple would need to vastly I,prove its cache subsystem to get better multi ore performance. Caches however take lots of power and as such one has to balance performance against heat.
The high clock rate is a possibility but I could see Apple designing in capability within the cores that gets activated in a "X" variant. This wouldn't be unusual in the world of semi conductors. In apples case they might do this to manage power. On the other hand if Apple has up locked the chip and can continue to control power then I think it is fair to say TSMC is way ahead of Intel now.
What I believe we're seeing this year is that Apple, with its highly optimized low power cores, is able to optimize its high power cores for performance better than before, as now they don't have the concerns about serious power gating in those cores. As we see from reports on this, Apple is talking a very different path to this than ARM itself has taken, but using a sp3cialized portion on the chip for core management which doesn't allow the chip to have mixed usage as Samsung first did with the Big/Little cores on their ARM designs, which other makers ended up in doing as well.
the problem with that approach is that the Little cores lack instructions the Big cores offer, as well as running at much lower speeds, making the use of all cores on high performance computing problematic. I've seen tests on Anandtech that show no actual improvement in computation when using all cores, and even some backstepping, as the Big cores wait for the Little cores.
apple has decided to not go that route. I'd love to know just what the low power cores can do. It's possible that they're significantly restricted as to function. Of course, we don't yet know how the GPU is handled when the low power cores are in use.
The scores don’t make sense. It gives a very low multicore processing efficiency compared with previous and current dual core processors from Apple. It would also mean a very high clock speed (2.8GHz) if Apple were to keep using the strategy of using the same cores in both iPad and iPhone SoC.
Id expect some variance once we have shipping processors. That is we could be seeing numbers form an engineering sample or an early pilot production chip.
As as for multi core efficiency aging I'd wait until real hardware ships but honestly I'm not that surprised. Apple would need to vastly I,prove its cache subsystem to get better multi ore performance. Caches however take lots of power and as such one has to balance performance against heat.
The high clock rate is a possibility but I could see Apple designing in capability within the cores that gets activated in a "X" variant. This wouldn't be unusual in the world of semi conductors. In apples case they might do this to manage power. On the other hand if Apple has up locked the chip and can continue to control power then I think it is fair to say TSMC is way ahead of Intel now.
Engineering samples wouldn’t be able to run Geekbench, unless it was tested inside Apple, since preproduction machines don’t run iOS outside of Apple. And that is another reason why I always doubt when these numbers come out too early.
And yet, as has been pointed out in several places, this guy has a good record with his numbers. So I assume that they're at least fairly close to the final results.
Thank you jbdragon, iaeen2, and rs1919 for explaining how size matters in the function of the transistors in a CPU with respect to power consumption and performance.
I myself am amazed at the progress that has been made. So much in so short time that the designers and engineers are running into trouble with the laws of physics.
That being said, TSMC is ahead of all others in chip fabrication. As Intel and Samsung catch up as they will because the transistors are hitting the limit in terms of size and shrinkage, TSMC will be able to continue to "tweak" their facilities for greater yields and lower cost. Apple will have ongoing access to TSMC's most advanced developments and should be able to stay ahead of everyone else. They have essentially passed Intel.
I do see 12 nm FDSOI making a big splash as the power consumption is much lower than the latest FinFet designs with very good performance. The Chinese are betting big on the technology and companies that rely on software only are going to be in for a rude awakening as the Chinese leverage a hardware edge to take over the market. It is very likely the reason that Google is so keen to get into hardware. They better assemble a chip design team fast, because a day is coming when Huawei and Xiaomi sell phones that do not come with Google Play.
Apple will still be in a good spot and could always take advantage of Global Foundries 12 nm FDSOI process for the S series SIP in the watch.
In any case, Apple's mobile processors are the best in the business. No one else is even close.
Well, actually, Intel is ahead of everyone else. TSMC may be number two. Intel has the densest process, meaning that they can actually fit more transistors on the same size chip than anyone else. In the coming years, as we reach 7nm, and gain experience with it, that advantage for Intel will disappear. But that won't be for at least three years, and more likely four or five.
Intel is not ahead. Not by a long shot. Read the following article and ponder the issues raised.
In the semiconductor business, TSMC is acknowledged as the leader by those who are experts.
Intel just went to a less dense process in manufacturing Kaby Lake also.
Intel does not have a 3 to 5 year lead. They don't have the same yields, Their manufacturing process is more costly and they do not have an advanced packaging process in the form of InFO.
TSMC is ahead of Intel and they are moving far faster. They just entered HVM on 10 nm and will be rapidly transitioning to 7 nm. TSMC's roadmap goes to 3 nm. Whether they get there remains to be seen. However, they are ahead of Intel. The A10 fusion is performing at i7 levels with a far narrower TDP envelope.
The scores don’t make sense. It gives a very low multicore processing efficiency compared with previous and current dual core processors from Apple. It would also mean a very high clock speed (2.8GHz) if Apple were to keep using the strategy of using the same cores in both iPad and iPhone SoC.
Id expect some variance once we have shipping processors. That is we could be seeing numbers form an engineering sample or an early pilot production chip.
As as for multi core efficiency aging I'd wait until real hardware ships but honestly I'm not that surprised. Apple would need to vastly I,prove its cache subsystem to get better multi ore performance. Caches however take lots of power and as such one has to balance performance against heat.
The high clock rate is a possibility but I could see Apple designing in capability within the cores that gets activated in a "X" variant. This wouldn't be unusual in the world of semi conductors. In apples case they might do this to manage power. On the other hand if Apple has up locked the chip and can continue to control power then I think it is fair to say TSMC is way ahead of Intel now.
Engineering samples wouldn’t be able to run Geekbench, unless it was tested inside Apple, since preproduction machines don’t run iOS outside of Apple. And that is another reason why I always doubt when these numbers come out too early.
And yet, as has been pointed out in several places, this guy has a good record with his numbers. So I assume that they're at least fairly close to the final results.
Actually "he" hasn’t a good track record and the article points that out " the testing results that were presented were nearly 15 percent too low for the production chip found in the iPhone 7, casting significant doubt on the veracity of the source." I have never seen any results coming out so soon that were proven to be true, for unreleased iPads or iPhones. It has only happened, at best, 1-2 days before the official presentation, at a time where there are already fully functioning devices being packaged.
Thank you jbdragon, iaeen2, and rs1919 for explaining how size matters in the function of the transistors in a CPU with respect to power consumption and performance.
I myself am amazed at the progress that has been made. So much in so short time that the designers and engineers are running into trouble with the laws of physics.
That being said, TSMC is ahead of all others in chip fabrication. As Intel and Samsung catch up as they will because the transistors are hitting the limit in terms of size and shrinkage, TSMC will be able to continue to "tweak" their facilities for greater yields and lower cost. Apple will have ongoing access to TSMC's most advanced developments and should be able to stay ahead of everyone else. They have essentially passed Intel.
I do see 12 nm FDSOI making a big splash as the power consumption is much lower than the latest FinFet designs with very good performance. The Chinese are betting big on the technology and companies that rely on software only are going to be in for a rude awakening as the Chinese leverage a hardware edge to take over the market. It is very likely the reason that Google is so keen to get into hardware. They better assemble a chip design team fast, because a day is coming when Huawei and Xiaomi sell phones that do not come with Google Play.
Apple will still be in a good spot and could always take advantage of Global Foundries 12 nm FDSOI process for the S series SIP in the watch.
In any case, Apple's mobile processors are the best in the business. No one else is even close.
Well, actually, Intel is ahead of everyone else. TSMC may be number two. Intel has the densest process, meaning that they can actually fit more transistors on the same size chip than anyone else. In the coming years, as we reach 7nm, and gain experience with it, that advantage for Intel will disappear. But that won't be for at least three years, and more likely four or five.
Intel is not ahead. Not by a long shot. Read the following article and ponder the issues raised.
In the semiconductor business, TSMC is acknowledged as the leader by those who are experts.
Intel just went to a less dense process in manufacturing Kaby Lake also.
Intel does not have a 3 to 5 year lead. They don't have the same yields, Their manufacturing process is more costly and they do not have an advanced packaging process in the form of InFO.
TSMC is ahead of Intel and they are moving far faster. They just entered HVM on 10 nm and will be rapidly transitioning to 7 nm. TSMC's roadmap goes to 3 nm. Whether they get there remains to be seen. However, they are ahead of Intel. The A10 fusion is performing at i7 levels with a far narrower TDP envelope.
Actually, no. That article is full of half truths and errors. Nobody's real roadmap goes to 3nm. Most chip experts still aren't sure of 5nm can be achieved.
The scores don’t make sense. It gives a very low multicore processing efficiency compared with previous and current dual core processors from Apple. It would also mean a very high clock speed (2.8GHz) if Apple were to keep using the strategy of using the same cores in both iPad and iPhone SoC.
Id expect some variance once we have shipping processors. That is we could be seeing numbers form an engineering sample or an early pilot production chip.
As as for multi core efficiency aging I'd wait until real hardware ships but honestly I'm not that surprised. Apple would need to vastly I,prove its cache subsystem to get better multi ore performance. Caches however take lots of power and as such one has to balance performance against heat.
The high clock rate is a possibility but I could see Apple designing in capability within the cores that gets activated in a "X" variant. This wouldn't be unusual in the world of semi conductors. In apples case they might do this to manage power. On the other hand if Apple has up locked the chip and can continue to control power then I think it is fair to say TSMC is way ahead of Intel now.
Engineering samples wouldn’t be able to run Geekbench, unless it was tested inside Apple, since preproduction machines don’t run iOS outside of Apple. And that is another reason why I always doubt when these numbers come out too early.
And yet, as has been pointed out in several places, this guy has a good record with his numbers. So I assume that they're at least fairly close to the final results.
Actually "he" hasn’t a good track record and the article points that out " the testing results that were presented were nearly 15 percent too low for the production chip found in the iPhone 7, casting significant doubt on the veracity of the source." I have never seen any results coming out so soon that were proven to be true, for unreleased iPads or iPhones. It has only happened, at best, 1-2 days before the official presentation, at a time where there are already fully functioning devices being packaged.
Actually that makes complete sense. I have no doubt has testing devices that aren't yet optimized. Even with the beta's of iOS that I run, Apple states that they aren't optimized, and that actually production could be faster, and usually is. That's doubly true for hardware. We've seen results come out pretty early for years now. But even with production results, on various websites, I'm seeing the A7 coming out as anywhere from 3,050 to 3,600 for single core. Depends on the software present on the device, as well as the OS version. I've noticed that my own iPhones sometimes get somewhat faster over the two years I keep them.
Thank you jbdragon, iaeen2, and rs1919 for explaining how size matters in the function of the transistors in a CPU with respect to power consumption and performance.
I myself am amazed at the progress that has been made. So much in so short time that the designers and engineers are running into trouble with the laws of physics.
That being said, TSMC is ahead of all others in chip fabrication. As Intel and Samsung catch up as they will because the transistors are hitting the limit in terms of size and shrinkage, TSMC will be able to continue to "tweak" their facilities for greater yields and lower cost. Apple will have ongoing access to TSMC's most advanced developments and should be able to stay ahead of everyone else. They have essentially passed Intel.
I do see 12 nm FDSOI making a big splash as the power consumption is much lower than the latest FinFet designs with very good performance. The Chinese are betting big on the technology and companies that rely on software only are going to be in for a rude awakening as the Chinese leverage a hardware edge to take over the market. It is very likely the reason that Google is so keen to get into hardware. They better assemble a chip design team fast, because a day is coming when Huawei and Xiaomi sell phones that do not come with Google Play.
Apple will still be in a good spot and could always take advantage of Global Foundries 12 nm FDSOI process for the S series SIP in the watch.
In any case, Apple's mobile processors are the best in the business. No one else is even close.
Well, actually, Intel is ahead of everyone else. TSMC may be number two. Intel has the densest process, meaning that they can actually fit more transistors on the same size chip than anyone else. In the coming years, as we reach 7nm, and gain experience with it, that advantage for Intel will disappear. But that won't be for at least three years, and more likely four or five.
Intel is not ahead. Not by a long shot. Read the following article and ponder the issues raised.
In the semiconductor business, TSMC is acknowledged as the leader by those who are experts.
Intel just went to a less dense process in manufacturing Kaby Lake also.
Intel does not have a 3 to 5 year lead. They don't have the same yields, Their manufacturing process is more costly and they do not have an advanced packaging process in the form of InFO.
TSMC is ahead of Intel and they are moving far faster. They just entered HVM on 10 nm and will be rapidly transitioning to 7 nm. TSMC's roadmap goes to 3 nm. Whether they get there remains to be seen. However, they are ahead of Intel. The A10 fusion is performing at i7 levels with a far narrower TDP envelope.
Actually, no. That article is full of half truths and errors. Nobody's real roadmap goes to 3nm. Most chip experts still aren't sure of 5nm can be achieved.
I had given a response but it apparently never posted.
The Intel core i7 5650u that went into the 2015 MacBook Air is on a die size of 133 mm2 and contains 1.9 billion transistors. The A10 fusion in the 2016 iPhone 7 is on a die size of 125 mm2 and contains 3.3 billion transistors. The GPU performance of the A10 fusion substantially exceeds the GPU in the core i7 5650u in performance. The A10 fusion doesn't exceed 3 watts with the i7 consuming 15 watts.
By every measure, including transistor density, the A10 fusion beats the broadwell based i7. And the A10 fusion nearly matches it in single core performance. Multi core is a different story, but the achievement Apple and TSMC have achieved is astounding.
Intel isn't competitive here. The A10X is being readied for release and is going to put the iPad in a class entirely of its own. At 5 watts, none of the latest core m CPUs will match it. The A10X will definitely exceed the best portable core i7 chip in GPU performance and very likely in single core performance also. Released on TSMC's 10 nm FF process, the transistors will be smaller than intel's 14 nm FF product. And manufactured at high volumes. Intel had been having yield problems at 14 nm. And Kaby Lake is less dense than the previous generation Skylake chips. Although with better performance. Small size and high density do not always translate to highest performance. Even Intel essentially admitted to that with their latest CPUs.
Both TSMC and Global Foundries wafer requirements exceeded Intel last year. Global Foundries planar 12 nm fully depleted silicon on insulator process is nearly complete also. While the process does not offer the ultimate in performance, it does provide superior power consumption levels.
TSMC is rapidly moving to 7 nm which will significantly exceed Intel's 14 nm process. And it's coming in less than a year. Not 3 to 5 years, less than a year. While arguments can be made that Intel's 14 nm process is equivalent to TSMC 10 nm process, TSMC's 7 nm will be denser and significantly so. No matter how Intel or their apologists like Mel Gross wish otherwise
I give up. My posts are just aren't going through. I typed up a detailed explanation regarding the feasibility and near certainty of 5 nm processors and included links. With 10 nm Cannon Lake pushed out to 2018 and 7 nm A11(X) SOCs just around the corner, TSMC actually is ahead of Intel. The A10X release is imminent and built on 10 nm.
I actually have no idea where Mel comes up with the notion that Intel is 3 to 5 years ahead in process technology. It is just not true. While I respect his opinions, I have to respectfully but emphatically disagree.
Global Foundries has essentially caught Intel and Samsung is close. SMIC has plans to move to 10 nm in the near future. TSMC and Global Foundries both move more wafers these days than Intel.
In fact, Global Foundries may enable AMD to produce a "fusion" processor with x86-64 and ARM cores. Such a chip would be ideal for a surface book or surface pro and provide MSFT a path forward for migrating off of x86.
Intel is inside the iPhone 7 with their modem. However the future of x86 computing is seriously in doubt. Intel no longer leads and if AMD can capture the Microsoft contract, Intel will need to refocus again on memory as high performance CPUs move over to the ARM platform.
The real issue in moving to 5 nm is cost. And Apple is willing to spend capital dollars to bring such a chip to fruition. Even a 3 nm chip is theoretically possible. The issue again is cost of building the technology into the fabs to build such chips. Quantum tunneling effects only become a serious issue at sub 3 nm.
Samsung apparently is putting some serious work into nanotube processors. One thing is very clear, Intel no longer is the undisputed leader. Apple's chips are the state of the art and built on TSMC fabs. Let's put the issue to rest now. I will no longer make any further posts on this issue. But in less than a year, Apple's iPhones will be significantly outperforming Intel's best portable CPUs. Never mind the iPads.
The scores don’t make sense. It gives a very low multicore processing efficiency compared with previous and current dual core processors from Apple. It would also mean a very high clock speed (2.8GHz) if Apple were to keep using the strategy of using the same cores in both iPad and iPhone SoC.
Id expect some variance once we have shipping processors. That is we could be seeing numbers form an engineering sample or an early pilot production chip.
As as for multi core efficiency aging I'd wait until real hardware ships but honestly I'm not that surprised. Apple would need to vastly I,prove its cache subsystem to get better multi ore performance. Caches however take lots of power and as such one has to balance performance against heat.
The high clock rate is a possibility but I could see Apple designing in capability within the cores that gets activated in a "X" variant. This wouldn't be unusual in the world of semi conductors. In apples case they might do this to manage power. On the other hand if Apple has up locked the chip and can continue to control power then I think it is fair to say TSMC is way ahead of Intel now.
Engineering samples wouldn’t be able to run Geekbench, unless it was tested inside Apple, since preproduction machines don’t run iOS outside of Apple. And that is another reason why I always doubt when these numbers come out too early.
And yet, as has been pointed out in several places, this guy has a good record with his numbers. So I assume that they're at least fairly close to the final results.
Actually "he" hasn’t a good track record and the article points that out " the testing results that were presented were nearly 15 percent too low for the production chip found in the iPhone 7, casting significant doubt on the veracity of the source." I have never seen any results coming out so soon that were proven to be true, for unreleased iPads or iPhones. It has only happened, at best, 1-2 days before the official presentation, at a time where there are already fully functioning devices being packaged.
Actually that makes complete sense. I have no doubt has testing devices that aren't yet optimized. Even with the beta's of iOS that I run, Apple states that they aren't optimized, and that actually production could be faster, and usually is. That's doubly true for hardware. We've seen results come out pretty early for years now. But even with production results, on various websites, I'm seeing the A7 coming out as anywhere from 3,050 to 3,600 for single core. Depends on the software present on the device, as well as the OS version. I've noticed that my own iPhones sometimes get somewhat faster over the two years I keep them.
Really? Can you show me any early results (by weeks, months) that were proven to be true? If you dismiss errors as lack of optimization then any value is true, even if it isn’t. And didn’t you notice that I said that early hardware won’t run iOS outside of Apple? Only Apple could run those kinds of tests. Geekbench results usually change very little with new iOS beta versions.
Comments
You, Sir, are as mad as a hatter.
As as for multi core efficiency aging I'd wait until real hardware ships but honestly I'm not that surprised. Apple would need to vastly I,prove its cache subsystem to get better multi ore performance. Caches however take lots of power and as such one has to balance performance against heat.
The high clock rate is a possibility but I could see Apple designing in capability within the cores that gets activated in a "X" variant. This wouldn't be unusual in the world of semi conductors. In apples case they might do this to manage power. On the other hand if Apple has up locked the chip and can continue to control power then I think it is fair to say TSMC is way ahead of Intel now.
More likely 3D fabrication will get popular instead, maybe 3D transistor structures, stacks of logic made in one go (rather than TSVs of separate dies)...
* Consider in the early 90s we were fabbing chips at 1um or thereabouts.
In particular, I'd love to see a Mac Pro that demolishes (in terms of performance) anything that Intel offers at the same price. The prices Intel charges for Xeons with more than 4 cores are so absurdly high that Apple really could do this. I'll bet they could produce an A# chip with 8 cores (no GPU, no ISP) that's smaller than an A5X and offers better multicore performance than a 6 core Kaby Lake, but for much less money. Put two of those in a Mac Pro and they'd have a very compelling unix workstation. Put one of them in an iMac.
And that is another reason why I always doubt when these numbers come out too early.
unfortunately, it doesn't look as though getting below 7 is going to be easy, if possible at all. The average size of the atoms used in chips is about .3 to .5 nanometer. We're getting awfully close to the limit on what we can do here.
Well, actually, Intel is ahead of everyone else. TSMC may be number two. Intel has the densest process, meaning that they can actually fit more transistors on the same size chip than anyone else. In the coming years, as we reach 7nm, and gain experience with it, that advantage for Intel will disappear. But that won't be for at least three years, and more likely four or five.
the problem with that approach is that the Little cores lack instructions the Big cores offer, as well as running at much lower speeds, making the use of all cores on high performance computing problematic. I've seen tests on Anandtech that show no actual improvement in computation when using all cores, and even some backstepping, as the Big cores wait for the Little cores.
apple has decided to not go that route. I'd love to know just what the low power cores can do. It's possible that they're significantly restricted as to function. Of course, we don't yet know how the GPU is handled when the low power cores are in use.
And yet, as has been pointed out in several places, this guy has a good record with his numbers. So I assume that they're at least fairly close to the final results.
https://www.semiwiki.com/forum/content/6240-will-tsmc-alone-10nm-7nm.html
In the semiconductor business, TSMC is acknowledged as the leader by those who are experts.
Intel just went to a less dense process in manufacturing Kaby Lake also.
Intel does not have a 3 to 5 year lead. They don't have the same yields, Their manufacturing process is more costly and they do not have an advanced packaging process in the form of InFO.
TSMC is ahead of Intel and they are moving far faster. They just entered HVM on 10 nm and will be rapidly transitioning to 7 nm. TSMC's roadmap goes to 3 nm. Whether they get there remains to be seen. However, they are ahead of Intel. The A10 fusion is performing at i7 levels with a far narrower TDP envelope.
" the testing results that were presented were nearly 15 percent too low for the production chip found in the iPhone 7, casting significant doubt on the veracity of the source."
I have never seen any results coming out so soon that were proven to be true, for unreleased iPads or iPhones. It has only happened, at best, 1-2 days before the official presentation, at a time where there are already fully functioning devices being packaged.
And didn’t you notice that I said that early hardware won’t run iOS outside of Apple? Only Apple could run those kinds of tests.
Geekbench results usually change very little with new iOS beta versions.