Saturday, July 20, 2013

by Bobby Situkangpoles

Pop quiz: What’s the similarity between HP Slatebook x2, HP slate 21, Nanotech Nuvola NP-1, Toshiba Excite Pro and Toshiba Write Pro? Never heard of them? Good answer. The similarity between these never-heard-of-‘em devices does not stop there. They also happen to be devices reported to be shipping with NVIDIA’s next gen Tegra 4 on board, as reported here by Android Authority. 

Earlier this year, NVIDIA stole the headlines with the launch of the Tegra 4 line of chips. Back then, Anandtech reported that the new Tegra chips will begin shipping in Q2 2013. Well now Q2 is behind us and apparently, apart from NVIDIA’s own SHIELD, the next highest profile devices in the list are Toshiba tablets. 

Last year, NVIDIA did rather well. Apart from having a number of big name manufacturers opting to use their Tegra 3 chip inside some of their high profile devices, NVIDIA was also the SoC supplier of choice to power both Google and Microsoft’s hero device, the Nexus 7 and the Surface RT. 

Of course, one can argue that NVIDIA is doing fine since their PC products are still the incumbent ruler of the PC world. However, recent reports have shown that the PC market had just been through 5 consecutive quarters of dwindling sales while mobile is accelerating at a rapid pace. Mobile is so important, that Intel has been putting a lot of effort to break into the mobile game. Their latest success is Samsung’s Galaxy Tab 3 10.1 which is powered by an Intel Atom chip. 

We will talk about this in another piece, but for now let’s get back to NVIDIA. 

Where did NVIDIA go wrong? In the past few years, NVIDIA has made their name popular in the ultra-mobile space by becoming the first to launch new generation of chips. Tegra 2 was the first commercially available dual core ARM Cortex A9 mobile SoC, while Tegra 3 can be said to be the first to break the four-core barrier.

While being first is good to guarantee headlines, it also comes with a number of early adopter downsides. Chief among which is the fact that, much like the superheroes that inspire NVIDIA’s product codenames, NVIDIA’s chips also have their own particular weaknesses, just like Superman’s kryptonite. 

Tegra 2 came without support for NEON SIMD extension, a feature that can accelerate multimedia and signal processing algorithms. This feature was already supported on a number of single core SOC’s at the time Tegra 2 was launched. Lack of NEON support was not Tegra 2’s only Achilles’ heel however, as the SoC also lacked dual channel memory bus. This limited the SoC’s memory bandwidth, thus performance, compared to their contemporaries. 

While Samsung’s first generation Exynos 4 dual and TI’s OMAP 4 - both can be considered the Tegra 2’s generational peers - can support up to 6.4 GB/ memory bandwidth, Tegra 2 had to make do with less than half of that, at 2.66 GB/s. 

The launch of the quad core Tegra 3, a year later, solved the NEON problem. However, that does not mean that the chip was free of any kryptonite when compared to other chips from the same generation. 

The first was the weakness carried over from Tegra 2. NVIDIA’s then new quad core chip still did not have support for dual channel memory. Although NVIDIA tried to improve its bandwidth by increasing clock, bringing bandwidth up to around 4.2 to 5.3 GB/s, those speeds are still slower than other SoCs from the same generation. The truth is, it was still slower than the two dual core chips from previous generation mentioned earlier. 

The second weakness of Tegra 3 was a more obvious example of the downside of being first. NVIDIA’s quad core SOC came with a 40 nm process while other quad core alternatives of the same vintage were already on 32 nm even 28 nm process. This deficiency hurt NVIDIA’s offering in both performance and power efficiency compared to other quad core chips at the time.

Tegra 4 was again the first ARM Cortex A15 based SoC to enter the market. This time around, it comes with a 28 nm process (thus making it on par with its soon to be contemporaries), support for out of order NEON instruction, and finally dual channel memory bus. But it also has its own kryptonite. 

See, not only that Tegra 4 requires an external DRAM (memory module), does limiting its implementation to larger devices (as reported by Anandtech), it also does not come with an integrated modem. One of the main advantages of the last few generations of Qualcomm chips is the fact that they have their LTE capable baseband module built into the same chip as the rest of the SoC, making implementation more efficient both technically and economically. 

Qualcomm's Snapdragon 600 

Tegra 4 requires a separate ICERA i500 block to provide the baseband for it to work on cellular networks. It’s not like NVIDIA did not realize this shortcoming though. They came up with the Tegra 4i which comes with both an integrated DRAM and a modem using the same IP as the ICERA i500. Making it a true integrated SoC solution for mobile devices just like those that Qualcomm offers. However, as it seems to be the case with all NVIDIA mobile SoC’s it wouldn’t be complete without its own kryptonite.

NVIDIA’s Tegra 4i has four Cortex A9 cores instead of the more current, Cortex A15. This makes it challenging for us to compare it to both Samsung’s Exynos 5 (Cortex A15) and Qualcomms Snapdragon 600 or 800 which use their own custom Krait cores. 

Let's do a recap of how things are with NVIDIA's SoC's at the moment.

Tegra 4 does not have integrated LTE capable modem and PoP DRAM, both the modem and memory have to be provided on separate modules. This not only increased the level of technical complexity and cost for OEMs but also limit the application of the SoC into smaller devices. 

To remedy the problem and provide a practical solution for smaller devices like smartphones, NVIDIA offers the Tegra 4i which has everything integrated inside the SoC. The only problem is, it comes with last generation ARM cores. Thus, it is safe to assume that OEMs looking to launch any high end "hero" device in the near future are not likely to be interested in the Tegra 4i. 

The problem is, most current and upcoming high profile mid range devices, like the Samsung Galaxy Mega 6.3 and Motorola's Moto X, are already on Snapdragons. 

Is there any reason not to be worried about NVIDIA's current efforts in the ultra mobile space?