Share This Entry:
In my first blog, we examined gave the historical context of the instruction set battles of ARM and x86, covering the RISC-CISC Wars in the PrePC Era and the PC Era. This blog covers Round 3, the PostPC Era [1].Round 3: RISC vs. CISC in the PostPC Era
The importance of maintaining the sequential programming model combined with the increasingly abundant number of transistors from Moore’s Law led, in my view, to wretched excess in computer design. Measured by performance per transistor or by performance per watt, the designs of the late 1990s and early 2000s were some of the least efficient microprocessors ever built. This lavishness was acceptable for PCs, where binary compatibility was paramount and cost and battery life were less important, but performance was delivered more by brute force than by elegance.
However, these excessive designs are not a good match to the smartphones and tablets of the PostPC era. RISC dominates these “Personal Mobile Devices,” because
- It’s a new software stack and software distribution is via the “App Store model” or the browser, which lessens the conventional obsession with binary compatibility.
- RISC designs are more energy efficient.
- RISC designs are smaller and thus cheaper.
The table below from Microprocessor Report supports these last two claims[2]:
Comparing performance per megahertz, x86 is 4% - 8% faster than ARM or MIPS. More significantly, this table suggests ARM and MIPS have 40% - 50% better energy per MHz and their size is a factor of 3X to 4X smaller than x86.
Independent of these architectural battles, Personal Mobile Devices rely on “Systems on a Chip” to reduce size, improve energy, and to lower costs. If processors are available as IP blocks, any company can create a single SOC rather than use many separate chips on a printed circuit board, as is the case with PCs. Thus far, there is no serious x86 IP competitor to the many fine RISC IP options, so SOCs based on x86 can only come from AMD or Intel.
RISC vs. CISC in the Client and in the Server of the PostPC Era
If Personal Mobile Devices are the clients of the PostPC Era, then Cloud Computing is the server. Virtually all PostPC apps will have one foot in the client and one in the cloud. While RISC has a substantial lead in PMDs, CISC leads in the commodity server market that is the building block of Cloud Computing.
Interestingly, binary compatibility again plays a small role in Cloud Computing, and cost and energy efficiency again play a much larger role than in PCs. Moreover, when you acquire 100,000 servers at a time to build a Warehouse Scale Computer, custom microprocessors could make sense. RISC competitors would need 64-bit addresses, ECC-protected memory, and good virtual machine support to compete in the Cloud, but the door is not slammed shut as it was in the PC Era.
Conclusion: RISC Reascendancy for Round 3
Note that the volume is on the side of PMDs in the PostPC Era: there will surely be 100 chips built for PMDs for every chip made for Cloud Computing. For 2010, even if you include the whole PC market—which you would expect to fade eventually in the PostPC Era—the RISC chips still outnumber CISC chips by 10:1 to 15:1.
Depending on your perspective, a happy result of the latest round of the RISC-CISC Wars is RISC reascendancy.
_________________________________
[1] “Dawn of a New Day,” Ray Ozzie, http://ozzie.net/doc...n-of-a-new-day/, October 28, 2010.
[2] “Broadcom Shows Off New CPU,” Linley Gwennap, Microprocessor Report, November 22, 2010.
David Patterson has been Professor of Computer Science at UC Berkeley since 1977. He is one of the pioneers of Reduced Instruction Set Computers, Redundant Arrays of Inexpensive Disks, and Network of Workstations in addition to being co-author of two widely-used textbooks on Computer Architecture, now in their 4th editions. He is a member of the US National Academy of Engineering, the US National Academy of Sciences, and the Silicon Valley Engineering Hall of Fame. Recently, while exploring the perceived value of Personal Mobile Devices in the PostPC Era, he made the shocking discovery that iPads make excellent Christmas presents for your adult children!
Share This Entry:
All company and product names appearing in the ARM Blogs are trademarks and/or registered trademarks of ARM Limited per ARM’s official trademark list. All other product or service names mentioned herein are the trademarks of their respective owners.
11 Comments On This Entry
Please log in above to add a comment or register for an account
Sebastian Garcia 
02 February 2011 - 09:18 AM
Professor Patterson, it is a nice surprise to see your blog post on ARM's website. I just want to ask what are from your point of view the research hot topics for computer architecture in the years to come. By the way, it would be great to know your thoughts about specific ARM processors architecture, so I hope that you can write about it in more technical posts. Best regards.-
David Patterson
03 February 2011 - 03:20 PM
Sebastian,
My view of Hot Topics in computer architecture:
CPUs vs. GPUs
MultiBigCore vs. ManySmallCore
Heterogeneous vs. Homogenous Multicore
-Dave
My view of Hot Topics in computer architecture:
CPUs vs. GPUs
MultiBigCore vs. ManySmallCore
Heterogeneous vs. Homogenous Multicore
-Dave
sols
03 May 2011 - 05:42 AM
There are several counter-arguments:
1) In a smart phone all compute-heavy tasks (i.e. video/audio codecs, graphics, baseband DSP) are performed by hardware accelerators, general purpose processor performs relatively compute-light (but software-complex) tasks. People are trying to come up with hardware acceleration even for HTML parsing and rendering. So, the performance and efficiency of general purpose processor might not matter as much.
2) For both smart phone and server what really matters is system level performance and energy consumption, including processor, LLC cache, memory, IO. It is likely that efficient processor such as Intel Atom dissipates only 10-30% of the total system power, so reducing this by a small factor is not as compelling as when one compares only core power dissipation.
3) CISC might be more energy efficient in certain cases, e.g. "Understanding Sources of Inefficiency in General-Purpose Chips", ISCA 2010. Of course, x86 is probably complex in a wrong way but well-designed complex ISA might be more energy efficient than RISC at least for some classes of applications.
4) Processor area is so small now that it doesn't matter. In fact, according to some projections (e.g. "The future of microprocessors", CACM 5/2011) in future technologies only small fraction of available transistors can be used for processing due to power budget limitations.
5) Your comparison of Intel Atom to ARM processors is not exactly fair: given that Atom is at least 25% faster than closest competitor, one can reduce its clock frequency and supply voltage to match performance and to significantly reduce energy per instruction or unit of work (mW/MHz is a proxy).
So, in the end it seems that ARM dominates cell phone market due to historical/software compatibility reasons rather than technical reasons, just like x86 dominates PC/server. I heard a story (granted, from a bitter ARM competitor) that the only reason why ARM is dominant in cell phones is because they somehow "convinced" Nokia to ask TI to make 1st GSM phone chipset based on ARM, and that was it.
1) In a smart phone all compute-heavy tasks (i.e. video/audio codecs, graphics, baseband DSP) are performed by hardware accelerators, general purpose processor performs relatively compute-light (but software-complex) tasks. People are trying to come up with hardware acceleration even for HTML parsing and rendering. So, the performance and efficiency of general purpose processor might not matter as much.
2) For both smart phone and server what really matters is system level performance and energy consumption, including processor, LLC cache, memory, IO. It is likely that efficient processor such as Intel Atom dissipates only 10-30% of the total system power, so reducing this by a small factor is not as compelling as when one compares only core power dissipation.
3) CISC might be more energy efficient in certain cases, e.g. "Understanding Sources of Inefficiency in General-Purpose Chips", ISCA 2010. Of course, x86 is probably complex in a wrong way but well-designed complex ISA might be more energy efficient than RISC at least for some classes of applications.
4) Processor area is so small now that it doesn't matter. In fact, according to some projections (e.g. "The future of microprocessors", CACM 5/2011) in future technologies only small fraction of available transistors can be used for processing due to power budget limitations.
5) Your comparison of Intel Atom to ARM processors is not exactly fair: given that Atom is at least 25% faster than closest competitor, one can reduce its clock frequency and supply voltage to match performance and to significantly reduce energy per instruction or unit of work (mW/MHz is a proxy).
So, in the end it seems that ARM dominates cell phone market due to historical/software compatibility reasons rather than technical reasons, just like x86 dominates PC/server. I heard a story (granted, from a bitter ARM competitor) that the only reason why ARM is dominant in cell phones is because they somehow "convinced" Nokia to ask TI to make 1st GSM phone chipset based on ARM, and that was it.
David Patterson
04 May 2011 - 02:32 PM
While everyone is entitled to an opinion, what I like about computer architecture is that industry unintentionally experiments with controversial issues to help shed light on contrasting opinions.
In addition to supplying the table that quantitatively compared many instruction sets that appeared in the last blog, Microprocessor Report recently published an article about the use of x86 based Atom processors in tablet computers, one of the PostPC personal devices that I talked about :
"At IDF Beijing, Intel formally launched its Oak Trail processor for netbooks and tablets. ... Oak Trail uses the second-generation Atom Z670 processor This single-core dual-thread CPU operates at 1.5GHz for Oak Trail, down from the 1.9GHz that Intel had originally specified for Moorestown, ... which gained no customers at all]. ... With dual threading, a 1.5GHz Oak Trail should deliver performance only slightly better than that of a dual-core 1.0GHz Cortex-A9 processor. These ARM processors are available today from several vendors at less than half the price, power, and board area of Oak Trail. Thus, we don't see Oak Trail as a credible tablet product unless Windows 7 compatibility is required. Windows 7 tablets, however, have sold relatively few units, mainly in industrial and vertical applications. Medfield [a successor to Oak Trail] may close the gap on power and price, but by the time it ships, dual- and quad-core ARM processors will in production at up to 1.5GHz, leaving no performance advantage for Intel."
Linley Gwennap, "In Brief: Oak Trail Leads Intel to Tablets," Microprocessor Report, May 2, 2011
Since binary compatibility can't be used as an excuse to stick to an inefficient instruction set in PostPC devices, differences in power, price, and area seems a rational reason. Factors of two in power, price, AND board area help explain why there were 6.1B ARM devices sold in PostPC devices in 2010, a twentyfold increase over x86 designs.
Computer architects I know believe that the complicated and continuously expanding x86 instruction set--enhanced by one instruction per month since its introduction in 1978--is the technical reason behind the doubling of power, price, and area, and factors of two will matter for a long, long time.
In addition to supplying the table that quantitatively compared many instruction sets that appeared in the last blog, Microprocessor Report recently published an article about the use of x86 based Atom processors in tablet computers, one of the PostPC personal devices that I talked about :
"At IDF Beijing, Intel formally launched its Oak Trail processor for netbooks and tablets. ... Oak Trail uses the second-generation Atom Z670 processor This single-core dual-thread CPU operates at 1.5GHz for Oak Trail, down from the 1.9GHz that Intel had originally specified for Moorestown, ... which gained no customers at all]. ... With dual threading, a 1.5GHz Oak Trail should deliver performance only slightly better than that of a dual-core 1.0GHz Cortex-A9 processor. These ARM processors are available today from several vendors at less than half the price, power, and board area of Oak Trail. Thus, we don't see Oak Trail as a credible tablet product unless Windows 7 compatibility is required. Windows 7 tablets, however, have sold relatively few units, mainly in industrial and vertical applications. Medfield [a successor to Oak Trail] may close the gap on power and price, but by the time it ships, dual- and quad-core ARM processors will in production at up to 1.5GHz, leaving no performance advantage for Intel."
Linley Gwennap, "In Brief: Oak Trail Leads Intel to Tablets," Microprocessor Report, May 2, 2011
Since binary compatibility can't be used as an excuse to stick to an inefficient instruction set in PostPC devices, differences in power, price, and area seems a rational reason. Factors of two in power, price, AND board area help explain why there were 6.1B ARM devices sold in PostPC devices in 2010, a twentyfold increase over x86 designs.
Computer architects I know believe that the complicated and continuously expanding x86 instruction set--enhanced by one instruction per month since its introduction in 1978--is the technical reason behind the doubling of power, price, and area, and factors of two will matter for a long, long time.
sols
06 May 2011 - 05:54 AM
Yes, I agree that industry unintentionally experiments with various issues. In 1997 Intel acquired StrongARM from DEC, produced XScaleprocessors with ARM ISA and tried hard to get into cell phone chipset market without success. Now Intel is trying to do the same thing with Atom with x86 ISA, also without success so far. It is hard to come up with more controlled real life experiment than that. So, one can conclude that ISA doesn’t matter and Intel’s failure in cell phone chipset market is not because of ISA – after all they tried ARM first.
On the other hand, it is not clear why of all RISC ISAs ARM is so dominant in cell phones. Does ARM have any technical advantage over other RISC ISAs MIPS/SPARC/Tensilica? I could not find any numbers or technical papers that show such an advantage. Maybe, SPARC was not suitable for embeddedsystems in the 80s-early 90s because of register windows. MIPS seems to be perfectly suitable and, in fact, is widely used in embedded systems but not in cell phones/smartphones/tablets. Tensilica designed ISA later specifically for embedded systems, provided code density (>20% better than ARM on EEMBC) by variable size 16/24-bit instruction formats (ARM did similar thing with Thumb-2only in 2003), avoided RISC ISA pitfalls such as delay slots and tried hard to penetrate cell phone market - without much success. It is ironic that of all RISC ISAs ARM is actually the least elegant, overloaded with unnecessary CISC-like features such as condition codes, pre/post-increment addressing with shifting/scaling (even x86 addressing modes are simpler).
So, one can conclude it is not technical merits of ISA that matter but other factors such as being selected by big player, i.e. IBM selecting Intel x86 for IBM PC and Nokia selecting ARM for GSM phone for whatever reason. Of course, afterwards ARM had to execute and deliver “good enough" solutions.
Of course, in the 80s and early 90s because of limited transistor budgets it probably made sense to put RISC rather than CISC into embedded systems. However, since then it is legacy/software compatibility that kept ARM in the leading position. Software compatibility is less important for cell phones than for PC but nonetheless it is important because of ecosystem of software tools, OSs and difficulty of porting applications to different ISA. Small difference in ISA would not be sufficient for cell phone manufacturers to switch from ARM to another ISA.
On the other hand, it is not clear why of all RISC ISAs ARM is so dominant in cell phones. Does ARM have any technical advantage over other RISC ISAs MIPS/SPARC/Tensilica? I could not find any numbers or technical papers that show such an advantage. Maybe, SPARC was not suitable for embeddedsystems in the 80s-early 90s because of register windows. MIPS seems to be perfectly suitable and, in fact, is widely used in embedded systems but not in cell phones/smartphones/tablets. Tensilica designed ISA later specifically for embedded systems, provided code density (>20% better than ARM on EEMBC) by variable size 16/24-bit instruction formats (ARM did similar thing with Thumb-2only in 2003), avoided RISC ISA pitfalls such as delay slots and tried hard to penetrate cell phone market - without much success. It is ironic that of all RISC ISAs ARM is actually the least elegant, overloaded with unnecessary CISC-like features such as condition codes, pre/post-increment addressing with shifting/scaling (even x86 addressing modes are simpler).
So, one can conclude it is not technical merits of ISA that matter but other factors such as being selected by big player, i.e. IBM selecting Intel x86 for IBM PC and Nokia selecting ARM for GSM phone for whatever reason. Of course, afterwards ARM had to execute and deliver “good enough" solutions.
Of course, in the 80s and early 90s because of limited transistor budgets it probably made sense to put RISC rather than CISC into embedded systems. However, since then it is legacy/software compatibility that kept ARM in the leading position. Software compatibility is less important for cell phones than for PC but nonetheless it is important because of ecosystem of software tools, OSs and difficulty of porting applications to different ISA. Small difference in ISA would not be sufficient for cell phone manufacturers to switch from ARM to another ISA.
Quick Links
My Blog Links
»
Blog Tags
»
Search My Blog
»
