Hummingbird vs. Snapdragon vs. OMAP vs. Tegra 2: ARM Chips Explained

By Ryan Whitwam

What's the same, and what's different in the world of SoCs?

Nvidia seems to finally be getting some traction in the mobile CPU game. The Tegra 2 chip will be in several phones and tablets in the next few months, and is the first dual-core mobile processor available. In response Texas Instruments, maker of the popular OMAP mobile chips, is talking up their own upcoming OMAP4 cores. With all this talk of clock speeds, some important details of mobile processing get lost in the shuffle. When we talk about OMAP, Snapdragon, Tegra, or Hummingbird, we're really talking about a system-on-a-chip (SoC). There's more to it than just the clock speed. 


Samsung Hummingbird

first debuted in the Galaxy S Android phone. The Hummingbird uses 45nm ARM Cortex A8 architecture at its heart with the ARMv7 instruction set. This part of the SoC is sometimes called the application processor. The 45nm manufacturing tech means that more transistors can fit on the die than in previous generations. But, the ARM architecture is just the starting point for chip design. Samsung (along with partner Intrinsity) had to build out some other features to differentiate the part. Intrinsity changed the logic design of the standard A8 allowing certain binary operations to be run with fewer total instructions. That means the same processes are completed, but in less time. When all is said and done, the Hummingbird is estimated to gain 5-10% over ARM's original tech. 

Samsung emphasizes power management while maintaining a high operating clock. Most Hummingbird cores are clocked at 1GHz, but the upcoming Infuse 4G will be clocked at 1.2GHz. Hummingbird also packs 32KB of both data and instruction cache, a variable-size L2 memory cache, and the ARM NEON multimedia extension. This last feature is a real advantage for Samsung. With NEON, the Hummingbird is capable of better hardware video encoding and decoding, high quality graphics, and better sound processing. 

The A4 comes from Intrinsity, and is very similar to the Hummingbird.
a very powerful graphics unit that easily outclassed the Snapdragons when it was released, though the new generation Snapdragons have caught up somewhat.  

Qualcomm Snapdragon

first generation Snapdragons were found in phones like the HD2, Nexus One, and Evo 4G. Now a new generation of Snapdragon chips have stepped things up. Unlike the Hummingbird, the Snapdragon does not use an ARM-designed application processor. Rather Qualcomm designed the core (called Scorpion) to have many features similar to the ARM Cortex-A8 architecture, but it does still use the ARMv7 instruction set.  
Going this route means an slight increase in instructions per clock cycle over the standard Coretex-A8.
The last generation Snapdragon chips from the Nexus One era were using the 65nm process tech. Now phones like the Desire Z, Thunderbolt, and Desire HD are using new Snapdragons on 45nm technology. Just as with the Hummingbird, this means more transistors, and better performance. Even the G2 clocked at 800MHz can best a last gen Snapdragon at 1GHz.  

Qualcomm has chosen to go with Adreno GPU SoC elements (acquired from AMD). On phones like the Nexus One, the Snapdragon used the Adreno 200, which was a poor choice. It just couldn't stand up to the competition. The relative multimedia robustness of Scorpion was able to carry this SoC over the finish line, but just barely. The current generation Snapdragon uses Adreno 205 GPUs. This GPU is just as fast as the Hummingbird, if not faster. We applaud Qualcomm for realizing there was an issue, and fixing it.  

Qualcomm also adds both GPS and cellular antennas to their SoC. This is not necessarily the standard practice in the industry. This allows phones to be designed and built more easily, and with slimmer designs since additional components are in one package. It explains why we see so many manufacturers working with Qualcomm. Although, the differences mean a lot of model numbers to keep track of. If you want to look up a phone's specs, make sure to use the whole code. 

Texas Instruments OMAP

there are a number of models out for the OMAP SoCs. That can make it hard to know just what's inside. 

The OMAP34x series are made using the older 65nm process, the OMAP36x use the newer 45nm technology. TI has added a few interesting elements to the package to go with these cores, some more useful than others. There is an IVA 2 imaging accelerator that supports hardware encoding of camera sensor data. This is paired with an integrated signal processor (ISP) that handles all the data crunching for image and video capture. The result is better battery performance and (hopefully) faster image capture.   

All OMAP SoCs in the 3-series use the PowerVR SGX530 GPU. This is capable part, but is starting to look a little slow compared to the GPU in the Hummingbird or new Snapdragon. To put things in perspective, this is very similar to the GPU in the iPhone 3GS (announced in June 2009). OMAP also has the advantage of using the NEON ARM instruction set to boost multimedia performance.  

TI has already been working hard to get the OMAP4 ready to go, and they expect the chips to begin showing up in devices later this year. One device of note expected to have the chip is the BlackBerry Playbook. These chips will have the newer PowerVR SGX540 GPU, a faster IVA 3 image accelerator, and oh yeah - dual Cortex-A9 1GHz cores. This chip will be a competitor to Nvidia Tegra 2, but it's still in production.  

Nvidia Tegra 2

Tegra 2 platform is about to explode all over Android. This SoC uses dual ARM Cortex-A9 cores clocked at 1GHz. It uses the ARMv7 instruction set as well. That alone makes it notable. It will be found in devices like the Motorola Atrix phone and Xoom tablet.  

The Cortex-A9 architecture uses a 40nm manufacturing process, so the transistor density is higher on this chip than the other A8-based chips. Power to the cores can be very closely monitored and controlled by the system, but both cores must be at the same power level (i.e. no individual control). Each core has 64KB of total cache, the same as individual A8 cores. Tegra 2 also uses dedicated ISPs that support up to 12MP camera sensors.  

The Tegra 2's other big trick is in the GPU department, just like you would expect from Nvidia. The GeForce Ultra Low Voltage (ULV) chip is called an 8 core GPU by Nvidia, but the truth is that these are not the standard shader cores from the desktop space. Half of the cores are for shaders, and half are for pixels. Still, the graphics performance is expected to be faster than the PowerVR SGX540. Tegra may pull ahead on battery use as well thanks to the multiple caches that can keep data readily accessible to the GPU. The GeForce GPU also allows for HDMI output support, and we expect more manufacturers take advantage of it like Motorola has with the Atrix. 

Qualcomm, Samsung, and TI are coming, but Tegra 2 looks like a tough act to  follow.