ARM Cortex A9 Performance Demo - Likely Chip to Power iSlate and Future iPhones

[cmaier]

I'm not so sure. They've had a hell of a time keeping secrets these days with all the vendors they need to use. Bringing manufacturing inhouse would make perfect sense for them. As for the billions of dollars, Apple's got close to $30 billion in CASH in the bank last I looked. Setting up their own fab would be chump change.

Yeah, they're not doing that. That would be a secret impossible to keep from those of us in the industry because they'd already have hired a thousand people, and they'd already have had to hire at least a few of my buddies to get it done. The buddies they've hired are all design guys, and they are clearly working with a contract fab (I believe TSMC, but not 100% sure).


It also takes years to build a fab, and Apple would have to have explained where the billions of dollars it cost to build one went in it's 10K's, etc.
Also, there is no advantage to them doing that. Even AMD doesn't make it's own chips anymore, and AMD has much more unique process demands than Apple would.

 

[halse]

CRTs?????????

where did they get the monitors?
haven't seen one, let alone two, of those in ages

couldn't they at least have gotten a pair of $200 20" flat screens?

 

[firewood]

Um, no. ARM CPUs are, from a technological perspective, quite boring. They are architecturally simple and unininteresting, ... And I say that as a guy who spent a decade designing chips for their competitor, AMD (and also designed Sparcs for Sun and PPCs for Exponential...

It makes sense that you might say that CPUs that are far from the leading edge in SPEC* are boring, given that the background mentioned above is mostly at the high performance end of the spectrum (been there, done that, burnt my fingers on GaAs prototype chips). But the ARM architecture has been a good target for a lot of interesting academic research in low power design (fully async, etc.), and has an ISA that is evolving better at allowing leading-edge low mW/"mips" implementations.

 

[cmaier]

It makes sense that you might say that CPUs that are far from the leading edge in SPEC* are boring, given that the background mentioned above is mostly at the high performance end of the spectrum (been there, done that, burnt my fingers on GaAs prototype chips). But the ARM architecture has been a good target for a lot of interesting academic research in low power design (fully async, etc.), and has an ISA that is evolving better at allowing leading-edge low mW/"mips" implementations.

Me, too re: the GaAs/AlGaAs stuff. HBT devices, CML logic.

My point was meant only to be that there's nothing inherently interesting or innovative about ARM, in response to that guy who claimed ARM was cutting edge and Intel sucked. And a lot more cutting edge stuff (microarchitecture, physical design, circuit design, process) happens in the x86 world than in the ARM world.

 

[aegisdesign]

If there is one notebook Apple wants to be extremely power efficient, it is the MacBook Air, so sign me up for a dual-processor, dual-core, 2GHz Cortex 9 model. Who needs super-fast processors when you have an OS that is focused on using what you have more efficiently?

If that was true, they'd not have given it a huge 13" screen. I don't think they're so bothered about power. It's more about looks and how thin it was.

 

[Hattig]

Web browsing is all well and good, but will the A9 decode high profile H264 1080p video?

It'll come with a video decode unit that shouldn't have any problems doing that.

Also ARM has the NEON vector instructions that aren't too shabby.

Atom has ... Intel graphics.

 

[iamkarlp]

They are LCD's!!!!!

While I have no desire to get ivolved in this conversation I would like to put to rest all the comments about the screens. They are not CRT's (although there wouldn't be anything wrong if they were). They are 15 inch viewsonc LCD's popular 4~5 years back for corporate desktop use. As a general rule they have now been replaced by larger monitors for end users and thus they are now easy to find in some companies IT shops and development labs as throw arround/ expendable monitors. There is nothing wrong with them however other than being a little small.

 

[Speedy2]

You're forgetting the fact that Flash will be gone soon. Once HTML 5.0 hits, everything will be embedded in a <video> using h.264.

Where did you get that?
Flash will (unfortunately) be around for a long, long time. There's no indication whatsoever that Google and Apple are succeeding in their endeavour to push HTML5 as Flash replacement. Heck, these two can't even agree on a common codec.

 

[wizard]

I'm surprised people still don't get it.

The device will have an ARM processor, it is pretty simple and if you read between the lines one can make good guesses about what that processor would look like.

The rumors and public actions indicate that Apple purchased PA Semi specifically to do custom ARM spins. These chips are to be implemented in Touch like devices.

While this is in part due to the lower watt per MIPS capability of ARM hardware, I believe there is a significant issue with being able to easily do custom SoC spins. Something Intel has yet to demonstrate skills with.

In essence the ability to do a progressive SoC is today the equivalent of designing a PC board in the 70s. In otherwords the Woz'es today would not be laying out a PC board but rather the System on Chips (SoC). To be competitive today a company has to be able to design at this level.

Of course that is describing glueing together different IP to make a whole. PA Semi could and is likely doing far more. I could see them deleting the ARM vector and multimedia units for a vector unit of their own design for example. Frankly what this final iteration of the tablets processor is is more interesting to me than the tablet itself.

You see contrary to statements made else where in this thread ARM and it's processors are very interesting indeed. They have reached a level of industry acceptance that has generated a lot of attention from other companies. The IP for ARM businesses are well known, but we are now seeing the EDA houses agressive applying low power tech against ARM designs. To me this is an indication that ARM hardware (IP) is going to be implemented more widely than ever before. The only bad thing there is can Apple keep up with it's ARM spins.

There is a real question to be asked here. Can Apple, as a single sink for the designs, keep up with the industry as a whole. It is going to be tuff as I've seen some pretty impressive A8 designs announced.

In any event there are a lot of variables with respect to what could be in Apples SoC. Everything from the process feature size to the low power tech implemented. That won't keep me from guessing though. So here are some guesses:

1.
The chip will likely come with an ARM A8 core. Well at least one of the chips as there are rumors of multiple designs on the boards for various devices. Why A8, time to market for one, more so A9 can be saved for rev2. Plus the same SoC would be usable on an array of devices.
2.
A9 with SMP support is possible also. This would then become a tablet specific chip. My concern here is that ARM IP might not be ready or Apples modifications. Demos aside I'd be really impressed if Apple had this series ready. The only reason to bring it up is the rumored OS features that might really make use of the SMP hardware.
3.
The GPU will be the latest from Imagination. That IP is adjustable in the sense that you can implement various number of cores. It will be interesting to see what Apple implements as the trade offs for performance and power usage will be interesting. I'm going to say 8 cores just for the hell of it.
4.
There is an interesting possibility that Apple will implement a frame buffer for the GPU right on the SoC. At the right process node there should be plenty of transistors available. That could be shared memory or it could be GPU dedicated. In any event it does a couple of things for them, it keeps the GPU from having to go off chip to memory saving power and it should end up being very fast. Yeah in one sense it is a lot of space on the chip, but should be doable for a buffer in the 32 to 64 MB range. For a tablet that would be a lot of RAM for video.
5.
Carefully tailored I/O will be in the game plan. This to eliminate as much unused electronics as possible to save power and die space. Some of that save die space can be used for the video buffer above. In any event we are talking two USB ports built In, WiFi hardware, Ethernet, the video drives/DACs, Touch interface hardware, some serial I/O for what can't go on the SOC, maybe Blutooth too and that is about it. Sounds like a lot but it really isn't when compared against the competition. Remember too this is to be integrated into the SoC. Apple may have to move one or two of those features off the chip, in that case PCI Express lanes would be implemented. But in the end I expect as much of the high speed stuff as is possible to be on the SoC. That to save power more than anything.
6.
The mystery meat in the sauce is unknown right now but I have to believe there is something to be had in this case. How else can Apple justify the expense of PA Semi? I alluded to the idea of an Apple inspired vector unit/multi media unit but there is opportunity for other ideas. For example Apple extends the ARM architecture to 64 bits. Another possibility is that Apple adds instructions or co-processors that enhance the execution of Objective C and Cocoa. Anybody have anyother ideas in this respect?

Oh one more thing, putting a desktop OS on a tablet is just stupid as is expecting desktop apps to run well on such a platform. All of this has happened before… Now is the time to break the cycle.



Dave

 

[cmaier]

PA Semi was cheap, and EDA means "electronic design automation" and refers to vendors who license software to other to do design work. The low power techniques used by PA Semiconductor stem from their own circuit design work and their own tools, as is generally the case amongst low power design houses, so not sure why you refer to EDA.

You dispute my statement that ARM is technologically uninteresting, but your reason seems to be that it has achieved market acceptance? What do apples have to do with oranges?

Finally, the guys at PA Semi are not SoC guys - they are custom guys. They aren't integrating hard IP like you seem to be implying. They are doing a full-on design, starting with ARM RTL (and possibly merely starting with ARM ISA). They will have no trouble beating any other vendor, all of which use ASIC design flows. The problem will be spinning iterations fast enough to keep up, as the PA Semi design flow is far more laborious than the crappy ASIC flows used by everyone else (other than the Intel's and AMD's of the world).

The device will have an ARM processor, it is pretty simple and if you read between the lines one can make good guesses about what that processor would look like.

The rumors and public actions indicate that Apple purchased PA Semi specifically to do custom ARM spins. These chips are to be implemented in Touch like devices.

While this is in part due to the lower watt per MIPS capability of ARM hardware, I believe there is a significant issue with being able to easily do custom SoC spins. Something Intel has yet to demonstrate skills with.

In essence the ability to do a progressive SoC is today the equivalent of designing a PC board in the 70s. In otherwords the Woz'es today would not be laying out a PC board but rather the System on Chips (SoC). To be competitive today a company has to be able to design at this level.

Of course that is describing glueing together different IP to make a whole. PA Semi could and is likely doing far more. I could see them deleting the ARM vector and multimedia units for a vector unit of their own design for example. Frankly what this final iteration of the tablets processor is is more interesting to me than the tablet itself.

You see contrary to statements made else where in this thread ARM and it's processors are very interesting indeed. They have reached a level of industry acceptance that has generated a lot of attention from other companies. The IP for ARM businesses are well known, but we are now seeing the EDA houses agressive applying low power tech against ARM designs. To me this is an indication that ARM hardware (IP) is going to be implemented more widely than ever before. The only bad thing there is can Apple keep up with it's ARM spins.

There is a real question to be asked here. Can Apple, as a single sink for the designs, keep up with the industry as a whole. It is going to be tuff as I've seen some pretty impressive A8 designs announced.

In any event there are a lot of variables with respect to what could be in Apples SoC. Everything from the process feature size to the low power tech implemented. That won't keep me from guessing though. So here are some guesses:

1.
The chip will likely come with an ARM A8 core. Well at least one of the chips as there are rumors of multiple designs on the boards for various devices. Why A8, time to market for one, more so A9 can be saved for rev2. Plus the same SoC would be usable on an array of devices.
2.
A9 with SMP support is possible also. This would then become a tablet specific chip. My concern here is that ARM IP might not be ready or Apples modifications. Demos aside I'd be really impressed if Apple had this series ready. The only reason to bring it up is the rumored OS features that might really make use of the SMP hardware.
3.
The GPU will be the latest from Imagination. That IP is adjustable in the sense that you can implement various number of cores. It will be interesting to see what Apple implements as the trade offs for performance and power usage will be interesting. I'm going to say 8 cores just for the hell of it.
4.
There is an interesting possibility that Apple will implement a frame buffer for the GPU right on the SoC. At the right process node there should be plenty of transistors available. That could be shared memory or it could be GPU dedicated. In any event it does a couple of things for them, it keeps the GPU from having to go off chip to memory saving power and it should end up being very fast. Yeah in one sense it is a lot of space on the chip, but should be doable for a buffer in the 32 to 64 MB range. For a tablet that would be a lot of RAM for video.
5.
Carefully tailored I/O will be in the game plan. This to eliminate as much unused electronics as possible to save power and die space. Some of that save die space can be used for the video buffer above. In any event we are talking two USB ports built In, WiFi hardware, Ethernet, the video drives/DACs, Touch interface hardware, some serial I/O for what can't go on the SOC, maybe Blutooth too and that is about it. Sounds like a lot but it really isn't when compared against the competition. Remember too this is to be integrated into the SoC. Apple may have to move one or two of those features off the chip, in that case PCI Express lanes would be implemented. But in the end I expect as much of the high speed stuff as is possible to be on the SoC. That to save power more than anything.
6.
The mystery meat in the sauce is unknown right now but I have to believe there is something to be had in this case. How else can Apple justify the expense of PA Semi? I alluded to the idea of an Apple inspired vector unit/multi media unit but there is opportunity for other ideas. For example Apple extends the ARM architecture to 64 bits. Another possibility is that Apple adds instructions or co-processors that enhance the execution of Objective C and Cocoa. Anybody have anyother ideas in this respect?

Oh one more thing, putting a desktop OS on a tablet is just stupid as is expecting desktop apps to run well on such a platform. All of this has happened before… Now is the time to break the cycle.



Dave

 

[wizard]

I'm very surprised at your statements here.

Me, too re: the GaAs/AlGaAs stuff. HBT devices, CML logic.

You seem to define interesting as bleeding edge technology. That is certainly one perspective but there are others to consider or at least acknowledge.

My point was meant only to be that there's nothing inherently interesting or innovative about ARM, in response to that guy who claimed ARM was cutting edge and Intel sucked.

I can't say that Intel sucks at everything they do. They are hitting on all cylinders right now with respect to most of their CPU line up.

However I don't believe they even come close to competeing with ARM when it comes to low power devices, IP or an industry accepted platfom for the embedded world. Sure this is a different focus than what Intel has but ARM still innovates.

And a lot more cutting edge stuff (microarchitecture, physical design, circuit design, process) happens in the x86 world than in the ARM world.

Unfortunately you lost a lot of credibility with that statement above as frankly you are so far off as to be laughable. With i86 you have Intel and AMD making most of the processors and a small number of companies doing support chips. Frankly there aren't even that many motherboard manufactures when looked at in comparison to the ARM market.

In fact if you take one sub category of ARM business, say cell phones, I would have to say it far outclasses what is being done with the entire I86 industry. It is not impossible for a modern cell phone to contain 6 ARM cores in various configurations.

ARM is everwhere but because there are no ARM inside stickers you don't notice the hardware nor the innovation. It isn't just the cell phone industry as ARM hardware shows up in just about everything. The embeeded world may not interest you but you have to consider that an ARM running at substantially lower power levels than Atom but yet offering as good or better performance is innovation. At least I would say that is impressive for a company moving away from its embeeded roots


Dave

 

[cmaier]

You seem to define interesting as bleeding edge technology. That is certainly one perspective but there are others to consider or at least acknowledge.

The only discussion in this thread of "interesting" had to do with "technically interesting." It's not my definition, it's just that you're changing the subject.

However I don't believe they even come close to competeing with ARM when it comes to low power devices, IP or an industry accepted platfom for the embedded world. Sure this is a different focus than what Intel has but ARM still innovates.

Of course they don't. They don't try to, either. At least not yet. When Intel can sell every high-end CPU it can make for $100, why would it spend a lot of energy on chips that sell for $20 unless they have to? The time will come when they have no choice, but for now Intel's focus is on where it makes the most profit.

Unfortunately you lost a lot of credibility with that statement above as frankly you are so far off as to be laughable. With i86 you have Intel and AMD making most of the processors and a small number of companies doing support chips. Frankly there aren't even that many motherboard manufactures when looked at in comparison to the ARM market.

I agree with the above, but not sure how my statement lost me credibility? I never claimed there were more x86 manufacturers than ARM manufacturers. I simply claim that x86's are more technologically innovative and interesting. ARM chips built today use technology that x86 guys mastered years ago.

In fact if you take one sub category of ARM business, say cell phones, I would have to say it far outclasses what is being done with the entire I86 industry. It is not impossible for a modern cell phone to contain 6 ARM cores in various configurations.

Outclasses how? In number of cores? No. AMD has 6 core chips already in production, and each x86 core is significantly more powerful than any ARM core. AMD's architecture is actually designed to allow up to 8 cores per die without having to redesign the crossbar.

ARM is everwhere but because there are no ARM inside stickers you don't notice the hardware nor the innovation. It isn't just the cell phone industry as ARM hardware shows up in just about everything. The embeeded world may not interest you but you have to consider that an ARM running at substantially lower power levels than Atom but yet offering as good or better performance is innovation. At least I would say that is impressive for a company moving away from its embeeded roots

So what? MIPS chips are in even more devices than ARM chips. Ubiquity doesn't equal innovation. Do you think 8-bit microcontrollers are innovative because they outsell ARM chips 1000 to 1?

x86 chips contain architectural features like traceback caches, out-of-order execution and scheduling units, register renaming, variable length instructions, instruction alignment, on-the-fly opcode translation, etc. They use advanced fabrication techniques and smaller lithographies. They use more advanced transistors with double gates. They have advanced floating point units. They have far more addressing modes. They use silicon-on-insulator technology, taller metal stacks, more advanced packages. They are hand-designed transistor by transistor for maximum optimization, unlike ARM chips which are run through a Synopsis and puked out as a tape on the other end. They have on-chip at-speed memory controllers and point-to-point interconnect buses like hypertransport.

ARM chips aren''t much more complicated than AMD's old K6 chips, but are built on semi-modern processes with clock gating and other features to reduce power (and, by the way, all of those features have already made it into use in x86, as well).

 

[wizard]

Still missing my points.

PA Semi was cheap, and EDA means "electronic design automation" and refers to vendors who license software to other to do design work.

I know what EDA means in any event I will repeat there are EDA vendors target ARM and the low power market with their tool sets. The point is there is a lot of innovation, with respect to the ARM market that is interesting to some of us. It may not mean much to you that a company can effective lower the power usage on an ARM core but it does to many. Especially to people in a forum thread focused on portable devices. Granted this is not "interesting" to many.

The low power techniques used by PA Semiconductor stem from their own circuit design work and their own tools, as is generally the case amongst low power design houses, so not sure why you refer to EDA.

Because there are EDA houses offering up tools to improve the power usage of of pruducts with respect to the same design built with other tools. You can't make an assumption that they are using in house tools because there are tools available that they could use.

As to what PA Semi is using I'd say that is more of an open question than you may think. Mostly due to time to market issues. Plus there is a strong likely hood that they have multiple chips in progress so they literally could be doing both.

You dispute my statement that ARM is technologically uninteresting, but your reason seems to be that it has achieved market acceptance? What do apples have to do with oranges?

No it is the fact that ARMs stradegy has achieved market acceptance that make the hardware interesting technologically. What you have is a core(s) that are being implemented in an enormous number of ways often with very interesting hardware attached.

As to the ARM cores you can think of them as technologically un interesting and that is a perspective you are free to have. On the other hand ARM continues to improve the core to make it better while keeping size growth in check.

This is not the same as Intel dropping one i86 design for another newer one but it is maintaining and improving the product line. Further ARM has something Intel doesn't have which is the partners that add to the ARM ecosystem. Maybe all that additional IP is also un insteresting but I see it in a different light, ARM provides a place to implement technology which is very interesting to me.

Finally, the guys at PA Semi are not SoC guys - they are custom guys.

Yes this is very true. But they where also a company with economic issues it is really hard to say what they contracted with Apple for and then what lead to the buy out. I'd love to know for sure though.

They aren't integrating hard IP like you seem to be implying. They are doing a full-on design, starting with ARM RTL (and possibly merely starting with ARM ISA).

You seem to be implying that they never have implemented somebody elses IP. That would surprise me but it is possible.

As to a full on design I don't discount that that is possible what I'm bothered with is it possible that design would be read for a tablet by now. Especially considering that some of that chip would be new functionality for PA. The other thing that bothers me is that some of Apples products simply don't have the lifetimes and volumes to justify a custom design. Lately the iPod Touch has been getting a new processor every year for example.

In any event you sound awfully sure about your position with respect to a full on design. Do you have inside information.

They will have no trouble beating any other vendor, all of which use ASIC design flows. The problem will be spinning iterations fast enough to keep up, as the PA Semi design flow is far more laborious than the crappy ASIC flows used by everyone else (other than the Intel's and AMD's of the world).

Well this to me is a serious issue. That is keeping up with all those ASIC building vendors out there. It is one of the reasons I'm kinda expecting one of this stich and glue crappy ASIC as you call them. Especially to get the rev one devices out. Apple could easily go the ASIC route, collect the required IP blocks and have PA create the missing functionality that started Apple down this road. In effect a hybrid approach where a portion of the chips IP is Apples. A side effect is that Apple could find IP for WiFi and some of the other wireless tech that PA doesn't have experience with.

Now the full on custom idea isn't a bad idea at all. In fact it would likely be the only practical way for Apple to implement some things I believe they are interested in. One thing I suspect would be a custom vector/multi media engine to replace Neon and other ARM technologies. Make that unit a programmable processor and it would work with Apples OpenCL technologies.

Basically I'm looking for a reason for Apple to go full custom. Especially when the volume isn't assured. A capability that no one else can offer is one possibility but that would be fleeting in this day and age. The synic in me says that the whole process is for lock in


Dave

 

[cmaier]

I know what EDA means in any event I will repeat there are EDA vendors target ARM and the low power market with their tool sets.

Name such a product.

The point is there is a lot of innovation, with respect to the ARM market that is interesting to some of us. It may not mean much to you that a company can effective lower the power usage on an ARM core but it does to many. Especially to people in a forum thread focused on portable devices. Granted this is not "interesting" to many.

I'm just asking what the innovation is - you said it's not technological. So what is it?

Because there are EDA houses offering up tools to improve the power usage of of pruducts with respect to the same design built with other tools. You can't make an assumption that they are using in house tools because there are tools available that they could use. As to what PA Semi is using I'd say that is more of an open question than you may think.

It's not an assumption. I know what they are using. I was responsible for EDA at AMD's california design team for nearly a decade. I know what's going on in the industry, and I know the people in the industry and what tools they are using.

No it is the fact that ARMs stradegy has achieved market acceptance that make the hardware interesting technologically. What you have is a core(s) that are being implemented in an enormous number of ways often with very interesting hardware attached.

What interesting hardware attached?

As to the ARM cores you can think of them as technologically un interesting and that is a perspective you are free to have. On the other hand ARM continues to improve the core to make it better while keeping size growth in check.

But to be innovative, doesn't the core have to have something other cores don't have? Or at least arrange things in a novel combination? If so, what is the new thing or combination that makes ARM so innovative?

Further ARM has something Intel doesn't have which is the partners that add to the ARM ecosystem. Maybe all that additional IP is also un insteresting but I see it in a different light, ARM provides a place to implement technology which is very interesting to me.

Surely you are joking. Intel doesn't have partners and an ecosystem? It's ecosystem is gargantuan compared to ARM.

In any event you sound awfully sure about your position with respect to a full on design. Do you have inside information.

Yes. I don't know what the people are working on, but I know who the people are - in many cases the people have been to my house and shared beer with me. I know what they know how to do, and I know that if they weren't doing what they wanted to be doing, they wouldn't still work there.

.

Basically I'm looking for a reason for Apple to go full custom. Especially when the volume isn't assured.

Dave

The design cost is low compared to the fab cost, and they don't have to pay the fab cost until they start the wafers. The advantage of full custom is that without even trying hard you immediately reduce the power per clock by 25-50%. Alternatively you can reduce area by x% (and hence cost) and power per clock by some other percent. Or you can increase clock and leave power constant. It is extraordinarily easy for trained people to beat synthesis tools. At AMD we did an experiment, and just by hand instantiating and placing standard cells (i.e.: letting a tool do routing, and not designing at the transistor level) we were able to beat the EDA companies by 20%. And we let them do the design with their own tools, so that the results would be unbiased. If instead you do transistor level design (which is what the PA Semi guys are trained to do from their DEC days), you easily do much better than that.

 

[wizard]

Innovation isn't just what you want to see.

The only discussion in this thread of "interesting" had to do with "technically interesting." It's not my definition, it's just that you're changing the subject.

Nope, you just have a narrow definition of what is technology.

Of course they don't. They don't try to, either. At least not yet. When Intel can sell every high-end CPU it can make for $100, why would it spend a lot of energy on chips that sell for $20 unless they have to? The time will come when they have no choice, but for now Intel's focus is on where it makes the most profit.

The point is it is technology! Maybe the ARM ecosystem isn't interesting to you but it is to many. That due to having the right technology to do the job in a way few others can. It is not wiz bang tech but that doesn't dismiss that it is tech.

I agree with the above, but not sure how my statement lost me credibility? I never claimed there were more x86 manufacturers than ARM manufacturers. I simply claim that x86's are more technologically innovative and interesting. ARM chips built today use technology that x86 guys mastered years ago.

This "… more technologically innovative and interesting" is exactly what I'm responding to. Many people would disagree with that.

Outclasses how? In number of cores? No. AMD has 6 core chips already in production, and each x86 core is significantly more powerful than any ARM core. AMD's architecture is actually designed to allow up to 8 cores per die without having to redesign the crossbar.

No you missed my point totally. The point is those ARM processors can be implementing many functions in a phone thus the many cores. The implementation of these specialized chips is very interesting. How many i86 chips get integrated into a chip to become a baseband processor.

Look I know that is a silly question but the point is it is interesting tech.

So what? MIPS chips are in even more devices than ARM chips. Ubiquity doesn't equal innovation. Do you think 8-bit microcontrollers are innovative because they outsell ARM chips 1000 to 1?

You are equating variety with sales. To your question yes some 8bit microprocessors are innovative. Just look at what comes on die and look at some of the CPU implementations. Are they comparable to and i86, nope but the do innovate even in 8bit land.
[/quote]

x86 chips contain architectural features like traceback caches, out-of-order execution and scheduling units, register renaming, variable length instructions, instruction alignment, on-the-fly opcode translation, etc. They use advanced fabrication techniques and smaller lithographies. They use more advanced transistors with double gates. They have advanced floating point units. They have far more addressing modes. They use silicon-on-insulator technology, taller metal stacks, more advanced packages. They are hand-designed transistor by transistor for maximum optimization, unlike ARM chips which are run through a Synopsis and puked out as a tape on the other end. They have on-chip at-speed memory controllers and point-to-point interconnect buses like hypertransport.
[/quote]
So!

Really are they anymore innovative than a 8 bit CPU that implements a precision timing co-processor. Now you may say that is not innovating the CPU itself as that is I/O. To an engineer trying to implement a dirt cheap lawnmower engine controller it is beyound innovation it is a solution to a problem. A fancy Cache is like wise a solution to a problem and also innovation. Which is interesting is a matter of perspective.

ARM chips aren''t much more complicated than AMD's old K6 chips, but are built on semi-modern processes with clock gating and other features to reduce power (and, by the way, all of those features have already made it into use in x86, as well).

What do you mean by already? It is only recently that i86 manufactures have even cared about power. As to some of those other i86 features they didn't break ground on i86. Many have filtered down from mainframes, mini computers and RISC chips. Sure it is great that the whole thing is on one piece of Silicon but they are not all new ideas.

That is not to discount the efforts required to get all those fancy features working on an i86 chip. It's just that one could say that the features aren't all innovation. For you they may be interesting for a compiler writer they may be a pain in the butt. Not to mention the grief they cause an embedded engineer. There are a number of ways to look at innovation and not everybody sees the same thing. As to interesting that is up to the individual and his tastes.


Dave

 

[cmaier]

What do you mean by already? It is only recently that i86 manufactures have even cared about power.

Nonsense. At AMD our prime design constraint on Opteron/Athlon 64, starting in 2000, was the power envelope. It was our number one, immutable design constraint. Sure, the TDP is much higher than ARMs, but the chip also does more. We had to use all the same tricks that PA Semi will use (gated clocks, elimination of dynamic logic, buffering strategies, minimum edge rates to remove crowbar current, special transistor optimizations) and more (SOI, etc.) to meet the power requirements. The point is you can always choose to dedicate the design to throughput or to low power, and the fact that ARM chooses a different point, and does it using technology that is years behind that used by Intel and AMD in making x86 processors doesn't make their technology innovative. I still haven't seen you explain a single innovative feature of any ARM core. (Yes, x86 chips have been glued to baseband processors, as well as all sorts of other SoC blocks. For many years AMD, National Semiconductor, and others sold these kinds of chips.)

 

[wizard]

This is interesting.

Name such a product.

I will have to look it up if I can. The info came out of one of my electronic design magazines. Basically they where touting a tool that took an ARM IP and generated a chip will a significant power drop over the same design run through conventional tools. The firm wasn't a mainstream EDA outfit.

I'm just asking what the innovation is - you said it's not technological. So what is it?

Well is all innovation technology? That is a good question. In the case if ARM I would call their ability to gain industry acceptance of microprocessors as IP to be used as ingredients in your own designs as innovation. If you efectively change the nature of the game that is innovation.

It's not an assumption. I know what they are using. I was responsible for EDA at AMD's california design team for nearly a decade. I know what's going on in the industry, and I know the people in the industry and what tools they are using.

I can accept your back ground, that isn't the problem. What I wonder about is getting a fully custom design out the door in the time since PA was purchased.

That may not be a concern though as I've heard that PA was doing work for Apple well before the buy out. If true then they would have had more time to produce something.

What interesting hardware attached?

I find it hard to believe you asked that question as some of the most interesting stuff out there is built on ARM SoC. There is hardly a month (week) that goes by without an ARM related announcement. Many of those announcements have the companis involved touting the latest hardware that sits next to an ARM core. It could be the latest GPU, a media engine or some other Piece of hardware to make the chip stand out.

But to be innovative, doesn't the core have to have something other cores don't have? Or at least arrange things in a novel combination? If so, what is the new thing or combination that makes ARM so innovative?

You can't just look at ARM but rather it is the ecosystem that allows one to build your one SoC reasonable quick. It is not the core that is the innovation but the ability for many to build what isn't economically possible otherwise or to bring technology to market in a far more salable way.

ARM doesn't need rapid innovation in and of themselves as they have many companies doing that for them. ARM just needs to incrementally improve their designs, which they do, to keep all these third parties interested in their solution.

Surely you are joking. Intel doesn't have partners and an ecosystem? It's ecosystem is gargantuan compared to ARM.

Intel has an ecosystem, however it is no where near as broad as ARMs. Surely you can agree with that? As to sheer size of that ecosystem I really don't have numbers right now but it is not insignificant. It's composition is radically different too, largely due to servicing different markets.

Yes. I don't know what the people are working on, but I know who the people are - in many cases the people have been to my house and shared beer with me. I know what they know how to do, and I know that if they weren't doing what they wanted to be doing, they wouldn't still work there.

Not to distract from the conversation but I just realized these people are in California. I'm sitting here in cold NY after three days of Snow. It is hot coffee or chocolate weather.

.


The design cost is low compared to the fab cost, and they don't have to pay the fab cost until they start the wafers. The advantage of full custom is that without even trying hard you immediately reduce the power per clock by 25-50%. Alternatively you can reduce area by x% (and hence cost) and power per clock by some other percent. Or you can increase clock and leave power constant. It is extraordinarily easy for trained people to beat synthesis tools. At AMD we did an experiment, and just by hand instantiating and placing standard cells (i.e.: letting a tool do routing, and not designing at the transistor level) we were able to beat the EDA companies by 20%. And we let them do the design with their own tools, so that the results would be unbiased. If instead you do transistor level design (which is what the PA Semi guys are trained to do from their DEC days), you easily do much better than that.

Wasn't one of the problems with the so called G5 Power PC a reflection of what you said above. That is the synthesized logic was very power hungery for what it was doing?

In any event we should know what Apple/ARM/PA Semi have been up to in a couple of weeks. In a way I'm actually hoping you are right and that Apple has had the time to do a full custom spin. More so I'm hoping that they do add a little extra to the sauce to address handling of multi media.

The concern about multi media is do to some rather impressive ARM based chips recently announced. Likewise I still have this concern that Apple will have a very hard time keeping up with all the other innovation in the ARM marketplace.

I will get back to this thread latter as the conversation has been most interesting.



Dave

 

[JimNoble]

MIPS chips are in even more devices than ARM chips.

Are you sure?

MIPS appear to sell about 100M units/qtr at the moment (source: http://www.mips.com/news-events/newsroom/index.cfm?i=43321).

Where as ARM shipped "over 1.0 billion" units in the most recently reported quarter (source: http://ir.arm.com/phoenix.zhtml?c=197211&p=irol-newsArticle&ID=1346312), which must surely make it the most popular microprocessor architecture on the planet? (They hit 14 billion cumulative units early in 2009 (source: http://ericschorn.com/2009/04/16/arm-basics-1/) - has any single microprocessor architecture sold more than that previously?)

Do you think 8-bit microcontrollers are innovative because they outsell ARM chips 1000 to 1?

A quick google suggests the 2009 market for 8-bit microcontrollers was estimated at 4.8 billion units (source: http://chipdesignmag.com/lpd/blog/2009/05/13/lines-blur-between-processor-and-microcontroller/), so I think maybe you're out by about a factor of 1000 there.

Jim

 

[cmaier]

Are you sure?

MIPS appear to sell about 100M units/qtr at the moment (source: http://www.mips.com/news-events/newsroom/index.cfm?i=43321).

Quite sure. You have 20 years of MIPS shipments to take into account. And I'm talking design wins, not units shipped. SKUs not sales. Same logic applies for both numbers - every car, traffic light, and fancy toaster made for a couple of decades had an 8-bit microcontroller in it.

 

[JimNoble]

Quite sure. You have 20 years of MIPS shipments to take into account. And I'm talking design wins, not units shipped. SKUs not sales.

The first ARM based products shipped in 1987

Same logic applies for both numbers - every car, traffic light, and fancy toaster made for a couple of decades had an 8-bit microcontroller in it.

No argument there. 8-bit mcus have been around a long time. Although you did say "outsells 1000 to 1" rather than "outsold", but that's just me being picky...


Jim

 

[cmaier]

The first ARM based products shipped in 1987



No argument there. 8-bit mcus have been around a long time. Although you did say "outsells 1000 to 1" rather than "outsold", but that's just me being picky...


Jim

Fair enough.

 

[MikhailT]

Oh man, this looks very juicy for the Apple Slate.

http://www.anandtech.com/gadgets/showdoc.aspx?i=3714