by Anton Shilov
12/25/2013 | 11:54 PM
There are numerous trends that have been ongoing in the recent years on the market of central processing units (CPUs): increasing share of x86 CPUs with integrated graphics engines and heterogeneous multi-core chips in general, growing importance of low-power and mobile x86 processors, emergence of high-performance ARM system-on-chips for mobile and server applications, and some others. In 2014 all those trends will have a tremendous effect on the market.
The vast majority of modern microprocessors for client personal computers – both from Advanced Micro Devices and Intel Corp. – feature integrated graphics processing units (GPUs). Those heterogeneous multi-core accelerated processing units (APUs) fully support OpenCL (as well as DirectCompute) and can accelerate applications that can take advantage of this API [application programming interface]. Next year APUs will play even bigger role than in 2013.
Early next year AMD will unveil its all-new code-named Kabini APU with heterogeneous system architecture that will bring acceleration of certain programs to a whole new level thanks to hUMA [heterogeneous unified memory architecture] and heterogeneous queuing technologies. Given the fact that next year AMD is not going to introduce new FX-series central processing units with significantly boosted performance compared the existing products (and therefore demand towards FXes will decrease), the share of APUs in AMD’s shipments will get higher.
By the end of the year Intel will also speed up its client chips with code-named Broadwell microprocessors that will brint high-end integrated graphics engines to desktop performance-mainstream offerings, which will basically mean that the majority of new Intel CPUs will carry advanced OpenCL accelerators. As a result, more software developers will be inclined to optimize their highly-parallel apps for chips with OpenCL-compatible heterogeneous multi-core design.
With Kabini and HSA, AMD seems to have a technological lead over Intel when it comes to heterogeneous computing in general. The problem is that there are loads of apps that rely on general-purpose x86 cores and therefore run faster on Intel’s chips. AMD’s main task for the next year will be the same as for the previous three: to communicate its GPGPU-related performance advantages over Intel to end-users and PC makers.
Thanks to the fact that AMD has started to offer APUs for server applications, it can be expected that such chips may start to gain traction there. However, it is impossible to expect these microprocessors to gain any significant market share since those who need performance in highly-parallel apps buy special-purpose accelerators or coprocessors, whereas those who do not need maximum speed do not have programs that take advantage of GPU compute units.
There is also a backside of the rise of heterogeneous computing. As we see with AMD, the company is boosting its GPU-compute capabilities in its chips, but performance of its x86 cores has been growing very slowly over the years. Unfortunately, it does not look like for AMD this is a priority right now. Therefore, do not expect breakthroughs in terms of x86 execution when it comes to Steamroller-based APUs. Perhaps, things will get somewhat better with code-named Excavator cores in 2015.
At present, there are no many reasons to expect dramatic x86 performance gains from Broadwell family of microprocessors. Still, with code-named Skylake chips due in 2015 Intel will bring a number of surprises, such as AVX 3.2 (with 512-bit instructions) and other advantages.
The demand towards ultra-thin notebooks, 2-in-1 devices and high-performance tablets is growing rapidly. Moreover, PC makers along with microprocessor developers are making those products more appealing in a bid to better compete against media tablets. As a result, next year leading CPU manufacturers will pay even more attention to ultra-low voltage chips.
At present, Intel Corp. offers a range of so-called Y-series microprocessors with 7W or even 4.5W scenario design power (SDP) based on high-performance x86 micro-architectures (e.g., Ivy Bridge or Haswell). In addition, the Santa Clara, California-based chip developer offers different processors with low power consumption based on low-power/low-cost micro-architectures (e.g., Silvermont) aimed at various portable devices. Next year Intel is projected to become more aggressive with its mobile offerings: the company will further cut power consumption of high-performance Broadwell chips and will enhance system-on-chips based on new-generation low-power Airmont and Goldmont micro-architectures.
AMD also offers low-power chips for all kinds of devices from notebooks to 2-in-1s to tablets. However, the company has not yet introduced such products based on high-performance micro-architecture (e.g., Piledriver, Steamroller). By contrast, it provides ultra-energy-efficient offerings featuring its low-power micro-architectures (e.g., Jaguar and Puma) only. At present AMD is interested to finally roll-out a competitive system-on-chip for media tablets, hence, this will be the company’s focus for 1H 2014. At the same time the firm cannot ignore the trend towards high-performance tablets and 2-in-1s. While AMD’s Jaguar and Puma seem to be good, they cannot compete with Haswell in terms of horsepower. So expect the Sunnyvale, California-based chip designer to expand its Kabini line with various ULP models featuring ~15W TDP and 11W SDP.
When it comes to ultra-low-power x86 microprocessors for clients (up to 10W SDP), Intel clearly has a lead over AMD since it can offer very energy-efficient [sub-10W or lower] chips based on both high-performance and power-efficient micro-architectures. By contrast, AMD can offer only Puma as a solution for low-power x86 products in 2014. In fact, this should be enough for certain media tablets, 2-in-1s and inexpensive notebooks, hence, AMD will sell plenty of its code-named Beema and Mullins APUs. The higher-end of the market will be occupied by Intel Core Y-series microprocessors with 6W – 7W SDP.
It is rumoured that by the end of the year AMD will introduce a consumer-class SoC based on ARM Cortex-A57 general-purpose cores. Theoretically, this will likely give AMD an edge over Intel’s SoCs when it comes to power consumption and graphics performance. However, with ARM inside, AMD will compete against several very powerful application processor designers who use ARM’s technology. The result of this rivalry is completely unknown.
The race for lower-power consumption by two makers of x86 chips has its side-effect as well. In a bid to make their chips consume less energy, AMD and Intel ceased to focus on pure performance, but started to make huge compromises. As a result, the horsepower of client chips is growing relatively slowly these days.
Three recently announced mobile application processors – Apple A7, Qualcomm Snapdragon 805 and Snapdragon 410 – predict the capabilities of next-year’s mobile system-on-chips better than any prophet. The general trends are pretty clear: ARMv8 64-bit architecture/instruction-set, new-generation graphics engines (e.g., ARM Mali T700-series, ImgTec’s PowerVR “Rogue” 6-series, Nvidia Kepler, Qualcomm Adreno 400-series, Vivante Vega and so on), support for ultra-HD resolutions and HEVC codec in addition to support for new types of cameras as well as 4G/LTE advanced connectivity.
Apple has clearly managed to leave everyone else behind with the world’s first application processors with 64-bit ARMv8-compatible cores. While at present there are not a lot of advantages that the 64-bit may bring to end-users, in about two years’ time there will be many programs that will benefit from the new technology. Given the fact that end-users will be looking forward smartphones and tablets with 64-bit app processors, developers of SoCs will make everything to speed-up their plans. At present it is hard to predict who among the leading designers of mobile SoCs will be the second after Apple to release a chip with 64-bit processing capability.
Potentially, Nvidia Corp.’s Tegra 5 can feature Cortex-A57 cores since the company is not yet ready with its Denver cores (which are due in late 2014 – early 2015). While performance provided by Cortex-A15 should be enough for a lot of tasks, Nvidia just cannot lose marketing war against other SoC developers, hence, it will likely adopt ARMv8 cores from ARM for its 2014 Tegra line.
In addition, it is also logical for Samsung Electronics to install a 64-bit chip into its upcoming Galaxy S5 smartphone so to be on par with Apple. The problem is that the firm historically launched app processors for flagship handsets several months ahead of the device; assuming that the S5 is scheduled for a March launch, the processor is either days from its formal announcement or the S5 will be based on another [non-Samsung] chip, perhaps, not a 64-bit one.
Qualcomm has already announced Snapdragon 410 app processors with 64-bit processing tech aimed at mainstream smartphones. Actual products based on the new chips are due in the second half of the year. Keeping in mind that at the Consumer Electronics Show 2013 Qualcomm introduced the whole line of this year’s SoCs, it can be expected that at the CES 2014 the company will also reveal its 2014 roadmap, which will likely include different 64-bit SoCs aimed at various market segments.
Mediatek, Rockchip, Via Technologies and other smaller app processor designers will be unable to significantly accelerate their roadmaps due to limited resources. Moreover, even if they can, their customers – who produce inexpensive smartphones and tablets – will prefer more affordable rather than feature-rich SoCs. Still, by the end of the year even smaller makers may adopt low-power/low-cost ARM Cortex-A53 cores.
New-generation ultra-mobile graphics processing units – from ARM Holdings, Imagination Technologies, Nvidia, Qualcomm, Vivante and some others – will not only deliver new levels of performance for mobile devices as well as support for new resolutions, but will also enable GPU-compute technologies on smartphones, tablets and other consumer electronics.
The latest graphics cores from various designers fully support OpenCL application programming interface and therefore can perform general-purpose processing on GPUs, thus significantly boosting compute capabilities of mobile application processors. While it is impossible to expect teraflop-class performance from a low-power chip, even 100GFLOPS can provide substantial advantages in demanding applications that can take advantage of GPU computing.
Thanks to support of the latest graphics APIs including OpenGL ES 3.0, OpenGL ES 2.0, OpenGL 3.x/4.x, OpenCL 1.x and DirectX 10, and enhanced performance, the new GPUs will also bring new levels of graphics effects to mobile games, thus further improving their positions against those designed for dedicated game consoles.
Infinity Blade III running on Apple iPhone 5S with A7 SoC inside.
Apple again was the first with adoption of new-generation graphics processing unit: the A7 features Imagination Technologies PowerVR “Rogue” 6-series GPU. At present Apple’s A7 is the world’s highest performing mobile graphics processor.
Qualcomm also formally rolled out its Snapdragon 805 with next-gen Adreno 420 graphics engine. Besides, the Snapdragon 805 is the world’s first commercial mobile 1GP/s ISP (image signal processor) and the world’s first mobile SoC with support for HEVC and 4K resolution. This is also the first chip to be made using 20nm process technology at TSMC.
Nvidia is projected to reveal Tegra 5 SoC with Kepler architecture graphics either at the CES 2014 in January or at the Mobile World Congress 2014 in February. It is unknown when actual products based on the new code-named Logan application processor are set to emerge.
Samsung will adopt a new-gen mobile graphics engine for its system-on-chips that will power its next smartphones and tablets. What is unclear is whether Mediatek, Rockchip, Via Technologies and other developers of inexpensive app processors will also include all-new graphics processors into their new chips.
With the rapid development of mobile devices in the recent years, system-on-chips that power smartphones and tablets got significantly more advanced than they were just three years ago. In many ways, they are closing the gap between themselves and processors aimed at fully-fledged personal computers. Some day we will see similar capabilities and technologies on microprocessors designed for completely different applications.
In the recent months Sony, Samsung Electronics and Qualcomm introduced their smartwatches, OmniVision unleashed reference design of smart glasses with LCOS display, marking the beginning of the consumer wearable computing era. Next year there will be more of such devices as well as specially-designed system-on-chips that will power them.
Qualcomm Toq smartwatch
Simplistic SoCs for ultra-low-power devices have existed for years, but with such well-known products as the forthcoming Apple iWatch, processors inside them will get publicity that they have never had before. Moreover Intel will also attract attention to its Quark platform, thus attracting a lot of public interest.
Omnivision smart-goggles prototype
In a bid to fulfill demand for wearable computing platforms, companies like Broadcom, Qualcomm and others will also offer appropriate solutions specifically tailored for consumer electronics-class devices. In a bid to better compete against Quark, they will most likely introduce special brand-names for such system-on-chips.
Intel Quark chip
The success of wearable computing products will be determined by proper software, not hardware, at least initially. Therefore, do not expect the market of smart-watches or smart-goggles to explode in 2014. However, it will emerge and therefore there will be competition between hardware makers.
While a number of noticeable events will happen on the market of servers, micro-servers and embedded applications, do not expect any particular breakthroughs in 2014. ARM Holdings will finally become a small player on the market of servers and micro-servers. AMD will introduce a range of its ARM-based solutions and Intel will continue to offer both extreme-performance as well as low-power server system-on-chips.
A slide from an Intel roadmap document. Image first published by VR-Zone web-site.
One of the most noticeable events due in 2014 is the emergence of Intel Xeon E5-2600 v3 “Haswell-EP” microprocessor with up to 14 cores as well as dual-socket Grantley platform based on C610-series “Wellsburg” chipset. The latter will bring support for quad-channel DDR4 memory as well as a number of all-new server-class capabilities. In addition, Intel will roll-out all-new Atom-class Xeon system-on-chips (based on Airmont architecture) designed for high-density/micro-servers.
AMD’s server plans for 2014 are pretty much well known. The company will refresh its Opteron CPU for dual-socket and quad-socket configuration with Warsaw design that will continue to pack 12 or 16 Piledriver cores like today’s chips for the same market segment. No significant performance improvement is expected. For uniprocessor servers AMD will introduce Opteron “Berlin” chip with four Steamroller cores, GCN graphics compute units and HSA features. Berlin will replace current quad-/octa-core Opteron chips for 1P machines. Essentially, AMD will try to persuade server software designers to optimize their applications for GPUs. Finally, the Sunnyvale, California-based company will roll-out its first ARM-based server product code-named Seattle with four or eight ARM Cortex-A57 cores.
Following its new strategy, AMD will offer a family of new chips aimed at embedded market, including Bald Eagle (2/4 Steamroller cores, AMD GCN graphics cores, most probably HSA features), Hierofalcon (4/8 ARM Cortex-A57, 10Gb Ethernet) and Steppe Eagle (2/4 Puma cores, AMD Radeon HD 8000 graphics). Those products should help AMD to address various embedded products, particularly those that need high graphics performance.
In 2014 the microprocessors will continue to do what they have been doing very well throughout their history: gain functionality as a result of gaining transistors. However, the main focus for many client x86 chips will be low power consumption, but not performance breakthroughs. When it comes to ARM-based client chips, then there is a different focus: high performance and limited power consumption.
The actual market of central processing units will mirror the market of computing devices in general.
Desktops will continue to stagnate, which is why performance of mainstream desktop parts will grow very slowly. The only good news for PC enthusiasts for next year will be emergence of next-gen Intel high-end desktop platform code-named Haswell-E with DDR4 memory support.
Traditional notebooks will continue to lose the pace of growth, leaving the market for 2-in-1s, ultra-thin notebooks and high-performance tablets. Therefore, expect the rise of ultra-low-voltage processors based on various micro-architectures: modern consumers clearly perform portability and battery life to performance.
Ultra-mobile devices like media tablets and smartphones will continue to close the gap with PCs and thus will get chips that support 64-bit computing capabilities, up-to-date graphics processing units, support for higher resolution displays, improved connectivity and other things.
Servers will continue to evolve in different directions. There will be feature-packed multi-core chips for high-performance machines and there will be low-power system-on-chips aimed at micro-servers and high-density servers. While ARM-based CPUs for servers will be launched, do not expect them to gain a serious market share in 2014: interested parties will first test-drive them before building large-scale deployments.