[CES 2011] It’s been exactly one year since NVIDIA announced and showed Tegra 2 to the world, and NVIDIA did not win a whole lot of contract (“design-wins”) with handset and tablet makers – until recently. If you follow this site regularly, you have probably noticed that Tegra 2 devices have been a common occurrence in the past couple of months, and you can expect this to continue going forward.
First of all, NVIDIA used to focus their software (drivers) efforts on Windows CE. It turns out that it was a mistake, although one that might be explained by the fact that an alliance with Microsoft and its early DirectX made NVIDIA successful in the past (RIVA 128 days).
NVIDIA has since turned its efforts towards Android, the Operating System (OS) that OEMs (Original Equipment Manufacturers) really want. This new direction, combined with the realization by wireless carriers that multimedia and gaming capabilities do generate money, have propelled NVIDIA from being a marginal player to being a real threat for arch-rivals like Qualcomm, Ti, Freescale, Marvell and others players in that space.
Multimedia and Gaming Do Matter
Two years ago, people were reasonably happy with their smartphones if they could browse the web decently, and even that is still a challenge for some platforms… Today, games are the type of content that brings the most money in the eco-system (app stores), so all the sudden, carriers want to have a great gaming phones. And guess who has been powering gaming graphics for a couple of decades?
Master the basics first
Being a “fast gaming machine” is not good enough. First, Tegra 2 has to prove that it is good at the basic things: battery life, web browsing. According to NVIDIA, Tegra-equipped phones browse the web a whole lot faster than established (but also older) competitors out there. The graph above shows its relative performance in an array of web benchmarks. This is basically a demonstration of the multi-core nature of Tegra 2, as web browsing is something that can be made to be multi-core friendly. Other multi-core architecture should exhibit similar improvements.Tegra 2 also uses a known principle: “perform the task quickly, and get back to sleep”. The idea is to allow the chip to enter a very “fast” mode (which consumes more power), so that it comes back to a “rest” state more rapidly. NVIDIA says that Tegra’s rest state is lower-power than many competitors. Although this might help battery life, the primary goal is to increase performance significantly – without making the battery situation worse. It might be counter intuitive, but using multiple cores at a lower frequency can lead to a significant power consumption reduction for a given task.
Then go to the next level
With basic stuff out of the way, the Tegra 2 team has identified two areas where they can make a big difference: Flash and 3D gaming. As you have seen, Flash might be available in select phones, but it doesn’t mean that it is usable at all. In fact, Flash (in the browser) is mostly unusable at this point.
Adobe Flash performance is typically slow and it is a CPU (and power) “hog”. On PC, Adobe, Intel, and many companies like AMD or NVIDIA have contributed to the Flash player code (or low-level portions) to make it what it is today. In the mobile world, this has not really happened yet – until now. If a company isn’t capable of optimizing Flash itself, it’s just not going to happen – that’s the bottom-line. With Tegra, NVIDIA has managed to accelerate the whole Flash pipeline in hardware. Years of working with Adobe Flash on the “PC” side is paying dividends here.
Most people expect NVIDIA to do well here, so it’s not really a surprise, but is it still a technological feat. OpenGL ES 2.0 found on mobile phones is equivalent to DirectX 9, so that’s hardly new for a company like NVIDIA. However, there are a few details that are particularly interesting, like the use of an “Early-Z” (a technique to quickly discard unseen pixels), programmable blending, or cache memory that reduces memory traffic (and therefore power). Some of these techniques are used to simply increase performance in discrete PC graphics, but here, there’s also a huge focus on power preservation.
3D Image quality is another thing that is often overlooked in mobile graphics. For example, NVIDIA says that it is the only company that has a shipping product with anisotropic filtering, a technique that eliminates jittery pixels at a distance, without making the image murky. It is something that we take for granted in PCs, but it’s actually new in handsets.
Finally, there’s Anti-Aliasing, a technique that eliminates “jaggies” on the edges of objects. NVIDIA thinks that its fast CSAA technique can make a difference in terms of quality/power. It probably will.
2D imaging is typically not associated with Tegra or NVIDIA, but this is hugely important for mobile devices. What NVIDIA brings to the table is raw performance: by being able to process images much more rapidly, the user can either experience a more interactive photo app, or snap more photos/sec to create content like panoramas. NVIDIA says that its throughput can be anywhere from 6X to 20X higher than established platforms like the OMAP 3430 (Droid X, Nokia N900…) and the PC110.
Content is King
It’s all nice, but we all know that hardware alone doesn’t win (or Samsung would rule the world by now). NVIDIA needs to convince developers to adjust their apps for Tegra, and it’s not easy when your installed-base is still small to non-existent. Developer want to sell games to more people, so it’s sometime hard to justify additional work for a nascent hardware platform.
It’s difficult, but far from impossible. Given that games are indeed the most lucrative type of apps, NVIDIA can bring some added-value to developers. For one, they have one of the best (and the largest) developer support team in the industry. That group can provide insightful knowledge and techniques useful for general 3D programming, but particularly good to exploit NVIDIA’s chips. Secondly, that team can also get into a game and optimize or add special effects, if the developer is short-handed.
Finally, NVIDIA can provide some marketing muscle to a developer in need of attention. This partially solves one of the most intricate “app developer” headache: discoverability. Help from a heavy-weight like NVIDIA can be like “gold” for a developer.
To be fair, once a game has been built for a generic OpenGL ES platform, it’s not that hard to make an improved version. Sure, it costs a bit more to build better assets, and tweak the graphics — but it is still an incremental effort that might be well worth it.
To make a long story short, NVIDIA will take the fight to a turf that it is well familiar with: the cut-throat high-performance gaming terrain. If it can turn a bunch games into “benchmarks” (and win them), the numbers will simply speak for themselves.
It has been a long time in the making, but Tegra is now getting real traction, and there hasn’t been a week in the past couple of months during which there was not an announcement of a new tablet or handset using Tegra 2. As the design-wins come in, NVIDIA will apply even more pressure to improve on every single metric that it has set for this line of product. With this influx of competition, you can expect a ferocious competition among chipmakers, and the end result for all of us is ever faster device — within the same power envelope.