Last month, Nvidia surprised roughly no-one by announcing the successor to Tegra 3, the processor at the heart of the Asus Transformer line of tablets, the Google Nexus 7, Microsoft’s Surface, HTC’s One X and a whole line-up of Android smartphones and slates. At the end of this month, we expect to see the Tegra 4 appearing in new smartphones and tablets at Mobile World Conference 2013, potentially including HTC’s rumoured flagship 5in phone, the M7, and tablets from Asus, Toshiba and Vizio. It’s unquestionably one of 2013’s most exciting mobile chips.
There’s also no doubt of its power. Where the Tegra 3 was built around quad-core ARM Cortex A9 architecture, Tegra 4 brings us a quad-core architecture based on ARM’s next-generation Cortex A15 design. A shift from TSMC’s 40nm LPG process to 20nm LPC helps Tegra 4 reach 1.8GHz – 200MHz faster than Tegra 3 – without blowing heat and power budgets. Like Tegra 3, Tegra 4 features a fifth companion core for background tasks, though again it’s moved from A9 to A15, but now supports dual-channel DDR3 RAM. Nvidia estimates that Tegra 4 is about 20 times faster than Tegra 2 and six times faster than Tegra 3; that’s not to be sniffed at.
Admittedly, there are some disappointments. In terms of GPU power, Tegra 4 is a big step up from Tegra 3. With six Vec4 vertex units and four 3-deep Vec4 pixel units, for a total of 72 cores, Tegra 4 has six times the number of Arithmetic Logic Units (ALUs) dedicated to graphics of Tegra 3, all running at a higher clock speed than they did on Tegra 3. In desktop PC terms, it’s close in power to the GeForce 7 series, including the RSX chip found in Sony’s PlayStation 3.
However, those expecting a more efficient unified shader architecture, as found in the GeForce 8 line and upwards, not to mention a shared architecture with Kepler, may be underwhelmed. Tegra 4’s GPU architecture is understood to be an extension of Tegra 3’s with key enhancements, rather than a wholesale reinvention. Finally, where ARM’s upcoming Mali T628 and T678 GPUs support OpenGL ES 3.0, as does Imagination Technologies’ next-generation PowerVR Series 6 Rogue architecture and the Adreno 330 GPU inside Qualcomm’s Snapdragon 800, Tegra 4 does not. For these goodies, it looks like we’ll now be waiting for the chip codenamed Logan, which will presumably be labelled Tegra 5 when it emerges later this year or 2014, depending on which rumours you believe.
In some ways, this risks making Tegra 4 look like a stopgap chip. Nvidia’s own marketing puts Tegra 4 at twenty times the speed of Tegra 2 but the potential Tegra 5 at fifty times that speed. While Tegra 4 introduces a new CPU architecture, it’s only with Tegra 5 that we’ll see a radically different GPU. And if Android games start using the new features of Open GL ES 3.0, then Tegra 4 might struggle to maintain the pace against upcoming devices featuring Rogue or Mali T678 - not to mention Qualcomm’s Snapdragon 800.
However, anyone dismissing Tegra 4 out of hand would be wrong. Maybe it’s not as advanced in some respects as we might wish, but it’s still a very powerful SoC. There are also some very good reasons why it’s a major step up from Tegra 3.
Resolutions of 1,366 x 768 or 1,280 x 800 might be OK at the low-end of the tablet market, but the 2012 iPads, Asus Transformer Infinity and Google Nexus 10 have set a new benchmark for mid-range to high-end tablet screens. Tegra 4 has been built with resolutions of 2,560 x 1,440 in mind, and will go up as high as 2,560 x 1,600. It supports Google’s VP8 compression format at 1440p and H.264 High Profile; a must for efficient Internet-based streaming of 1080p video. Tegra 3 supported resolutions of up to 2,560 x 1,600 and H.264 High Profile and did a nice job of powering the 1,920 x 1,200 Transformer Infinity. However, Tegra 4 has the performance to take things up to the next level. Expect 1440p and 1600p to become baseline standards for 10in tablet screens during this year.
At CES Nvidia was pushing what it calls its Computational Photography Architecture. We wouldn’t take the name or the hype too seriously – we’re not talking about new hardware, but about software that takes the CPU and GPU in Tegra 4 and combines it with the smartphone or tablet’s Image Signal Processor to enable instant High Dynamic Range (HDR) photography, live HDR previews on the screen, HDR video and HDR burst shooting.
Why is this a big deal? In case you don’t know, HDR photography is a technique where you enhance the dynamic range of a shot by taking multiple shots with different exposure values and merging them into one final photo. Traditionally, this requires three or more shots to be taken in quick succession then merged, either in-camera or in image-editing software later. Nvidia’s scheme dramatically cuts the speed with which these shots are taken and processed, so that HDR photos appear at a speed of 0.2s/frame rather than 2s/frame, making them pretty much instantaneous. This is particularly important in a smartphone, where tiny sensors, over-stuffed with pixels, struggle to produce any kind of decent dynamic range. With the right sensor and the right lens, Computational Photography Architecture could mean better smartphone photos, and maybe a lesser reliance on Instagram-style filters for effect.
Whether Nvidia has better technology than other mobile GPU makers is open to question. We’re expecting big things from Imagination Technologies with Rogue, and from Qualcomm and ARM with the Adreno 330 and Mali T678 architectures. However, Nvidia does have a better grip on the games industry. It has years of experience in helping developers get to grips with new chips through the “The Way its Meant to be Played” programme, and good links with developers from the high-end PC space (Valve, Epic, Crytek, Ubisoft, DICE) and the mobile space through the TegraZone programme and app. It might not always play fair, sometimes encouraging developers to engage with proprietary technologies rather than take a more standards-based approach, but it knows how to get the games to push its hardware. By championing platforms as well as its own hardware, Nvidia helps drive gaming forward.
Vector Unit is the developer behind Riptide GP, Shine Runner and Beach Buggy Blitz on iOS and Android, and Hydro Thunder Hurricane on Windows 8 and Windows RT. “I will say I appreciate how gamer-focused Nvidia is” says Matt Small, the company’s Creative Director. “They really think a lot about game development and game performance when they design their hardware, and that’s a big part of why we like to work with them.”
While Small won’t go so far as to describe Tegra as a reference platform for Android games development, he will say that “actually we kind of use them that way, as our primary devices during development.” It doesn’t give them a reference platform they can rely on – “we still have to test and optimise for lots of different devices. Sometimes even the exact same device will have different issues depending on what carrier it’s on” says Small – but by developing for Tegra, developers like Vector Unit can see how its game should run across a pretty good range of devices.
What’s more, Tegra 4’s increased horsepower will allow for some fairly major boosts in visual quality, as you can see from CES demos of Vector Unit’s Riptide GP2 and Mafinger’s Dead Trigger 2 in action. “Games optimised for Tegra 4 will bring advanced graphics features such as specular lighting, bloom effects, high resolution textures, and MSAA” Nvidia’s Technical Marketing Director, Nick Stam, told us. “The performance gap between consoles and mobile devices may not be bridged soon, but one can certainly expect to see Tegra 4-powered devices delivering console-class gaming experiences with advanced graphics effects and support for game controllers.”
Matt Small agrees. “If you’re talking about raw performance, then even Tegra 4 is not on par with Xbox 360 or PlayStation 3” he says, but adds that “it’s getting close enough that if you put them side-by-side, if you compare Riptide GP2 to Hydro Thunder Hurricane [on Xbox 360] or Dead Trigger 2 to Left 4 Dead for instance, I think a lot of players would have a hard time telling you which one is a mobile game.”
What Smith will say is that “the graphics processor on Tegra 4 is a really big jump in performance – it’s a bigger leap than Tegra 3 was over Tegra 2. Our new games like Riptide GP 2 are going to take advantage of this in a number of ways. One of those is improved lighting, with real-time rendered shadows throughout the environment. That means bridges cast shadows on the riders and the water, riders cast shadows on jet skis, etc. The other big change is improved materials, which includes texture effects like normal mapping, per-pixel specular, detail mapping, and other effects to make the surfaces of things like rocks and metal appear more realistic and detailed.”
The net effect? “I think Tegra 4 can deliver a game experience every bit as polished, immersive and spectacular as a current gen console game” Smith says, “It’s really a question of whether developers can afford the time and budget it takes to make a game that deep and polished… and whether they can make enough money back selling it for $1.99.”
The other plus for Tegra 4 is that it’s a chip designed to power new form factors. Tegra and its rival SoC processors were originally built to power smartphones, then enhanced to power tablets. Now they can be found in anything from TVs to laptops to games machines. “Tegra 4 was designed to flawlessly power a range of different devices, such as smartphones and tablets, gaming devices, auto infotainment and navigation systems, and PCs” says Nvidia’s Nick Stam. Just as Samsung’s Exynos 5 Dual has shown up in its latest Chromebook, while Tegra 3 and Snapdragon S4 processors are turning up in Android and Windows RT convertibles, so we can expect to see Tegra 4 appearing across a wider gamut of products.
The most obvious example is Nvidia’s own Project Shield gaming handheld. It’s clearly a showcase for Tegra 4, and a sign of the kind of multiple-device convergence that the company sees coming in the future. Nvidia isn’t positioning it as a competitor to the Nintendo 3DS or PlayStation Vita. Instead, Nick Stam claims that it’s “aiming for a different audience”.
“We built Shield because we believe that gaming should be based on open platforms” he told us. “Consumers can check out free-to-play Android games from anywhere they want, while also being able to stream games from their PCs, untethered, from anywhere in the house. We saw an opportunity to bring about a better Android gaming experience, where we wanted to give Android gamers the ergonomics and controls that they're familiar with from the more traditional console controllers. PC streaming with Project Shield brings about a brand-new way to experience PC games, and with the recent explosion in both the number and the complexity of mobile games on the market, we feel a device such as Project Shield just makes sense.”
Whatever form it comes in, Tegra 4 won’t have the market to itself. In fact, with forthcoming ARM A15/Mali T678, Snapdragon 800 and A15/Rogue SoCs on their way, the competition has never been tougher. Yet you can expect Tegra 4 to have a major presence across a range of form factors, and show the way forward for next-gen SoCs. Bring on the devices, and the software to show what they can do.