2014 is going to be an interesting year in mobile technology. This year, we’ll see the introduction of the first programmable, DX11-class GPUs, the debut of LPDDR4, a new 20nm modem that should offer significant power savings, and the debut of 64-bit mobile ARM chips for the Android ecosystem.
Each of these new features is backed by a different major company – and that’s before we toss in the shifting dynamics at the bottom of the market, where companies like Rockchip and MediaTek are winning designs and building their own SoCs to challenge the established players.
This could be the year that Android gaming finally takes off thanks to mobile Kepler, or another 12 months dominated by Qualcomm’s high-end design wins. Here’s what the major players are planning, in alphabetical order…
Nvidia launches first programmable GPU, debuts Tegra 4i in non-US markets
Nvidia faces an uphill battle in multiple markets in 2014. The firm will debut its first integrated software modem (the Icera i500) with devices shipping in Q2 of this year. Interestingly, Nvidia doesn’t plan to bring the chip to the US market. During Nvidia’s last conference call, CEO Jen-Hsun Huang said that the chip’s launch “will likely be global, but not US.” This is likely a calculated move – the Tegra 4i’s Icera i500 modem doesn’t support CDMA, which Huang has said is a necessity to be successful in the US.
Nvidia’s launch schedule for the next-generation Logan platform (presumably called Tegra 5) is rather unclear. During the last conference call, Huang stated that the ramp up between Tegra 4 and Tegra 5 would be much shorter than the delay between Tegra 3 and Tegra 4 – but also implied we’d see more Tegra 4 hardware shipping in the next few months. It’s thought that Nvidia will first use Tegra 5 in a Shield follow-up, as that product could give an excellent demonstration of the chip’s new GPU.
It’s also unclear if Nvidia is ramping a 64-bit ARM core, or if it’ll stay with a tweaked Cortex-A15 design for the CPU side of its own hardware. With Project Denver (Tegra 6) in the pipeline and slated for 2015, Nvidia may have decided to stay with 32-bit CPUs for this cycle and pour their efforts into building a new GPU. Mobile Kepler will make or break Nvidia’s aspirations in this space. If the chip’s GPU performance is as impressive as Nvidia (NV) demoed last June, it could easily be the breakout graphics winner for the entire segment. The comparisons to the GTX 8800 and PS3 in the image below are overblown, but mobile Kepler should excel.
Whether that will make Tegra 5 the breakout solution for the mobile market, however, is open to debate. NV chose to push ahead on 28nm rather than jumping for 20nm in 2014 as both Samsung and Qualcomm intend to do. In theory, this will cut costs and speed time-to-market, giving Nvidia a several-month lead over any of the other major providers. The only problem is, NV could find itself beaten on footwork later in the year.
Then again, this strategy worked quite well for Tegra 3, which debuted on 40nm but still found wide acceptance in multiple tablet SKUs for 2012. There’s a case to be made on both sides of this point, and for now we’re going to give NV the benefit of the doubt on staying on 28nm. It’s more important for Tegra 5 (Logan) to move the bar on both CPU and GPU, with better overall power characteristics for the former and a breakaway solution for the latter.
Without an integrated modem ready for the US market, we expect NV will focus on high-end smartphones, tablets, and custom devices like Shield to drive US sales, with the Tegra 4i saved for growth markets where low cost is more important than the latest and greatest feature sets. Here, the phone will face stiff competition from the likes of Rockchip and MediaTek. Both of these companies have grown rapidly and they’re keen to capitalise on a flood of new customers in emerging markets. Nvidia will need to capture strong wins in high-end products to fund the jump for 20nm.
Qualcomm: New DX11 GPU, 20nm modems, 64-bit CPUs
Qualcomm is focusing its own next-generation efforts on model technology first. The company has already made a slew of announcements for 2014, including its Gobi 9x35 platform and the WTR3925 RF transceiver chips. Both products are cutting-edge – the 9x35 will be the first 20nm cellular modem, while the WTR3925 is the first announced RF transceiver chip built on 28nm.
There’s a reason why Qualcomm is pushing its modem technology first – its near-lockdown on 28nm LTE modems was a huge part of the reason why the company won so much market share in late 2012 and 2013. Competitive solutions, like Intel’s XMM 7160, have only recently begun shipping. Now, by jumping for 20nm first, Qualcomm hopes to lock down a significant share of the 2014/2015 market as well.
The company’s next-generation high-end CPU platform, the Snapdragon 805, will pair an iterative CPU at higher clock speeds with a brand-new, DX11-capable GPU dubbed the Adreno 420. Qualcomm is staying mum with the details, but available information suggests clock speeds as high as 2.5GHz (up from 2.3GHz with Snapdragon 800), a GPU clock speed of 500MHz, and 40 per cent more shader throughput than the Adreno 300 could offer. This will also be the first Qualcomm smartphone SoC to offer hardware support for the nascent H.265 standard, ushering in dramatic improvements to power consumption while decoding video streams. That tech won’t be widely used for several years, until H.265 becomes standard, but being first out of the box with a hardware decoder is a significant achievement.
Qualcomm also has a midrange 64-bit solution coming next year: The Snapdragon 410. We’ve covered that chip extensively in other areas – it could hit the sweet spot for power consumption and multi-core performance, while beating out the Cortex-A7 by a significant margin. Qualcomm isn’t beating the 64-bit drum as hard as Samsung, but we do expect to see more 64-bit chips from the company than a simple low-end core. 64-bit chips could appear at the high-end by the close of 2014, depending on whether Qualcomm opts to launch a simple ARM Cortex-A57 or extend its own custom architecture to address the market.
Qualcomm’s toughest competition in graphics will likely be Nvidia, which has its own plans for the segment, but Krait has done extremely well against the Cortex-A15 in terms of power consumption per watt. Nvidia will need to deliver an improved variant of the original Tegra 4 CPU to fully match Krait’s power consumption.
Samsung races for LPDDR4, ultra-wide memory, image processing, and 64-bit CPUs
Samsung’s roadmap indicates that the company is focused on a mixture of near-term, incremental goals with a few technological leaps tossed in for good measure. In the incremental category we’ve got upcoming 20nm chips based on the Cortex-A57, Samsung’s steadily improving AMOLED technology, including flexible AMOLEDs debuting in 2014, and better big.LITTLE implementations coming in future Exynos phones.
Despite earlier rumours that the upcoming Galaxy 5 would be based on a 14nm chip, this appears to be incorrect; at its Analyst Day in November, Samsung’s Doctor Namsung Woo, President of Samsung LSI, stated that: “Next year  we expect to use 20-nanometre, and the year after, 14-nanometre FinFET technology.”
Samsung’s first 64-bit core, presumably built on 20nm, will be a Cortex-A57. To date, Samsung has always used ARM’s actual CPU designs, but that’s going to change – the company is working on its own implementation of an ARM core. At present, it’s expected that this chip will debut on a 14nm FinFET process, likely towards the end of 2015.
One of Samsung’s biggest pushes in 2014 will be the adoption of LPDDR4. The company actually spent a great deal of time at its Analyst Day talking about how the market for mobile memory has changed compared to its desktop/laptop counterpart. In mobile, being first to a new standard allows the company to capture far more of the total profit. The implications here are that only the first mover ever makes money on investing in a new standard, and Samsung is determined to be the first out of the gate to LPDDR4. That means we’ll see 4GB phones in 2014, at least at the high-end. Longer term, Samsung has an entirely different card up its sleeve: Widcon.
Widcon, formerly known as Wide I/O, is Samsung’s next-generation memory technology that utilises TSVs (through silicon vias) to connect a CPU to its main memory. The idea is simple: Instead of using a relatively narrow, high-speed bus to access a pool of main memory, connect that RAM directly to the IMC (integrated memory controller) in a much wider configuration by running hundreds of wires directly between the chip and the RAM.
Samsung believes it can achieve significant performance improvements compared to both LPDDR3 and LPDDR4 (Widcon 1 being the DDR3 analogue, with Widcon 2 mapping to DDR4.) Unfortunately, the company has no strict timeline for its introduction. We know that Widcon is in the prototype stage, with test chips successfully manufactured in-house, but don’t expect to see a shipping solution until 2015.
Meanwhile, companies like MediaTek and Rockchip are going to be working hard to grab the lower end of the market across the world, even if they don’t gain much traction in the US. Significant pressure will be brought to bear on all three of the primary US vendors, particularly Nvidia, whose Tegra 4i is supposed to go head-to-head across the globe without a US anchor to brace against. The low-end ARM vendors are throwing mud at the wall as well, experimenting with everything from octal-core Cortex-A7s to new quad-cores built at GlobalFoundries. MediaTek has done particularly well in 2013, posting a 34 per cent sales gain from 2012 to 2013.
Of the three players covered here, Nvidia is in the most precarious position, but could reap tremendous rewards if Kepler delivers the mobile gaming experience the company believes it can. Samsung and Qualcomm are playing to respective strengths, with Qualcomm probably the better positioned of the two thanks to a dominant modem technology and an in-house custom CPU. For Samsung, success means deploying a better version of big.LITTLE in future cores, launching new mobile memory standards, and delivering improvements in foundry technology to match its first custom CPU core.