The ARM Wrestle Match

on January 6, 2012
Reading Time: 4 minutes

I have an un-healthy fascination with semiconductors. I am not an engineer nor I do know much about quantum physics but I still love semiconductors. Perhaps because I started my career drawing chip diagrams at Cypress Semiconductor.

I genuinely enjoy digging into architecture differences and exploring how different semiconductor companies look to innovate and tackle our computing problems of the future.

This is probably why I am so deeply interested in the coming processor architecture war between X86 and ARM. For the time being, however, there is a current battle within several ARM vendors that I find interesting.

Qualcomm and Nvidia, at this point in time, have two of the leading solutions for most of the cutting edge smart phones and tablets inside non-Apple products.

Both companies are keeping a healthy pace of innovation looking to bring next generation computing processors to the mass market.

What is interesting to me is how both these companies are looking to bring maximum performance to their designs without sacrificing low-power efficiency with two completely different approaches.

One problem in particular I want to explore is how each chipset tackles tasks that require both computationally complex functions (like playing a game or transcoding a video) and ones that require less complex functions (like using Twitter or Facebook). Performing computationally complex functions generally require a great deal of processing power and result in draining battery life quickly.

Not all computing tasks are computationally complex however. Therefore the chipset that will win is one that has a great deal of performance but also can utilize that performance with very low power draw. Both Nvidia and Qualcomm license the ARM architecture which for the time being is the high performance-low power leader.

Nvidia’s Tegra 3
With their next chipset, Tegra 3, Nvidia is going to be the first to market with a quad-core chipset. Tegra 3 actually has five cores but the primary four cores will be used for computationally complex functions while the fifth core will be used to handle tasks that do not require a tremendous amount of processing power.

The terminology for this solution is called Variable SMP (symmetric multiprocessing). What makes this solution interesting is that it provides a strategic and task based approach to utilizing all four cores. For example when playing a multi-media rich game or other multi-media apps all four cores can be utilized as needed. Yet when doing a task like loading a media rich web page two cores may be sufficient rather than all four. Tegra 3 can manage the cores usage, based on the task and amount of computer power needed, to deliver the appropriate amount of performance for the task at hand.

Tegra 3’s four cores are throttled at 1.4Ghz in “single core mode” and 1.3Ghz when more than one core is active. The fifth core’s frequency is .5Ghz and is used for things like background tasks , active standby, and playing video or music, all things that do not require much performance. This fifth core because it is only running at .5Ghz requires very little power to function and will cover many of the “normal” usage tasks of many consumers.

The strategic managing of cores is what makes Tegra 3 interesting. This is important because the cores that run at 1.4 Ghz can all turn off completely when not needed. Therefore Tegra 3 will deliver performance when you need it but save the four cores only for computationally complex tasks which will in essence save battery life. Nvidia’s approach is clever and basically gives you both a low power single-core, and quad-core performance computer at the same time.

Qualcomm’s S40 Chipset
Qualcomm, with their SnapDragon chipset, takes a different approach with how they tackle the high performance yet low power goal. There are two parts of Qualcomm’s S40 Snapdragon chipsets that interest me.

The first is that the S40 chipset from Qualcomm will be the first out the door on the latest ARM process the Cortex A15. There are many advantages to this new architecture, namely that it takes place on the new 28nm process technology that provides inherent advantages in frequency scaling, power consumption and chipset size reduction.

The second is that Qualcomm uses a proprietary process in their chipsets called asynchronous symmetric multiprocessing or aSMP. The advantage to aSMP is that the frequency of the core can support a range of performance rather than be static at just one frequency. In the case of the S40 each core has a range of 1.5Ghz to 2.5Ghz and can scale up and down the frequency latter based on the task at hand.

Qualcomm’s intelligent approach to frequency scaling that is built into each core allows the core to operate at different frequencies giving a wide range of performance and power efficiency. For tasks that do not require much performance like opening a document or playing a simple video, the core will run at the minimum performance level thus being power efficient. While when running a task like playing a game, the core can run at a higher frequency delivering maximum performance.

This approach of intelligently managing each core and scaling core frequency depending on tasks and independent of other processes is an innovative approach to simultaneously delivering performance while consuming less power.

I choose to highlight Nvidia and Qualcomm in this analysis not to suggest that other silicon vendors are not doing interesting things as well. Quite the contrary actually as TI, Apple, Marvel, Broadcom, Samsung and others certainly are innovating as well. I choose Qualcomm and Nvidia simply because I am hearing that they are getting the majority of vendor design wins.

The Role of Software in Battery Management
Although the processor play’s a key role in managing overall power and performance of a piece of hardware, the software also plays a critical role.

Software, like the processor, needs to be tuned and optimized for maximum efficiency. If software is not optimized as well it can lead to significant power drains and result in less than stellar battery life.

This is the opportunity and the challenge staring everyone who makes mobile devices in the face. Making key decisions on using the right silicon along with effectively optimizing the software both in terms of the OS and the apps is central going forward.

I am hoping that when it comes to software both Google and Microsoft are diligently working on making their next generation operating systems intelligent enough to take advantage of the ARM multi-core innovations from companies like Qualcomm and Nvidia.

These new ARM chipset designs combined with software that can intelligently take advantage of them is a key element to solving our problem with battery life. For too long we consumers have had an un-healthy addiction to power chords. I hope this changes in the years to come.