News of the day (February 16, 2011)
Following the announcement of the alliance between Nokia and Microsoft, Intel has taken the opportunity to bring us up to date on its mobile activities at Mobile World Congress . Nokia’s decision to go with the Windows Phone 7 platform is indeed a disappointment for Intel – they and Nokia had announced at MWC a year ago that they were working on a common software palform: MeeGo. Based on Linux, MeeGo brings the Intel Moblin platform together with Nokia’s Maemo and is aimed at netbooks, smartphones and tablets.
The MeeGo ‘tablet’ interface in portrait mode with applications installed. Not as attractive as an iPad…
In announcing that 100% of its smartphones would eventually be moving over to the WP7 platform and that only one MeeGo product would be launched before the end of the year (possibly the N9, with rumours telling us that it would be the first mobile phone on the market to use an Atom), Nokia has clearly demonstrated its lack of interest in the platform. Although it’s open source, Nokia’s contribution to the project has been huge with more than 600 engineers assigned to it. The jobs of these employees, like those working on Symbian, are now under threat.
MeeGo UX tablet in landscape
With respect to the hardware, Intel should be introducing the Medfield platform before the end of the year. Intel has already worked on two mobile phone platforms in the past (Menlow et Moorestown) and although we did see the prototypes (notably from LG for Moorestown), no model actually came onto the market. Anand Chandrasekher gave a quick presentation of a Medfield chip at MWC, saying that the SoC at 32nm would not require an additional chipset (Moorestown currently does).
Chandrasekher presented a (new) prototype based on Medfield (and running on Android!), saying that other models were under preparation. He once again highlighted the fact that Medfield will represent a better energy/performance ratio than the equivalent ARM solutions (Cortex A9 dual and quad cores like the NVIDIA Tegra 2 and 3). Intel naturally has an advantage when it comes to advanced fabrication processes, but it remains to be seen if Medfield will improve energy consumption at idle as this was quite high with Moorestown. The x86 instruction set requires decoders that are particularly power hungry in comparison to the ARM RISC architecture. Although almost negligeable on modern PC architectures, the x86 architecture overhead once again becomes a problem with very low processing power cores.
The LG GW990 phone based on Moorestown and announced by Paul Otellini at CES 2010 and suspended in the middle of last year
Note that on the graphics side, where Intel had opted for the Imagination Technologies solutions (SGX535 derived from PowerVR and identical to what you find for example in the Apple Cortex 8 ‘A4’ SoCs), Medfield will use a Silicon Hive solution. This company, a spin-off from Philips, had already announced a partnership with Intel last year for forthcoming Atom generations. Intel took the opportunity to announce the acquisition of this company at MWC .
At the Mobile World Congress currently being held in Barcelona, NVIDIA has unveiled a roadmap for its SoC Tegras and given some more detail on Kal-El which is likely to be marketed as Tegra 3.
Before going into more detail, we should remind you that the current Tegra 2 was unveiled in the course of summer 2009 though it was officially launched in January 2010. It will therefore have taken over a year for products based on it to see the light of day. This holds for a few smartphones and tablets that are about to be launched. Not long ago, NVIDIA launched a new version of this component, the Tegra 2 3D with the CPU clock boosted and 3D screen support. The Tegra 2 3D could however be supplanted by Kal-El by NVIDIA’s partners, as Kal-El availability has been announced for this year: during the summer for tablets and Christmas for smartphones!
Tegra 2 is based on two ARM Cortex-A9 cores clocked at 1 GHz (1.2 GHz for the 3D version) with 1 MB of shared L2 cache. A small ARM 7 conducts proceedings because, in addition to the two main cores, the Tegra 2 includes numerous small specialised processors: video playback, video encoding, audio, image processing and the GPU of course. The GPU is derived from GeForce 6/7 architecture and has 8 ‘cores’ in total. It’s difficult to know what NVIDIA means by core here but four are dedicated to the vertex shaders and four to the pixel shaders as it isn’t a unified architecture.
For Kal-El, NVIDIA has retained the same architecture and 40nm fabrication process with TSMC. There are now four ARM Cortex-A9 cores, still with 1 MB of shared L2 cache and the GPU is up from eight to twelve cores. NVIDIA hasn’t given any detail on the distribution of these cores, but we imagine there will be four cores for the vertex shaders and eight cores for the pixel shaders, so as to have enough processing power to supply higher resolution screens. Note that each ARM core also now has an MPE vector unit, which wasn’t the case before.
NVIDIA has moreover announced that it has developed its video playback processor so it can support up to 40 Mbps in h.264 high profile, corresponding to the Blu-ray standard. This is an important development because Tegra 2 is limited to h.264 main profile with a throughput of 20 Mbps. Support for 3D Blu-rays requires continuous 60 Mbps support and therefore isn’t yet in the bag.
NVIDIA is still using a single channel LPDDR2 memory controller for the new SoC and won’t therefore be able to count on faster memory to give higher bandwidth.
While Kal-El seems promising on paper, there will be competition from Qualcomm and T1 who are also preparing new SoCs. TI recently presented the OMAP 5, the first SoC based on the Cortex-A15. Two of these cores will be included, with 2 MB of L2 cache and a dual channel LPDDR2 memory controller, as well as an SGX544 GPU from PowerVR. In contrast to NIVIDIA and T1, Qualcomm develops its own ARM cores and has just presenented Krait, its new CPU core. This will be integrated into the forthcoming Snapdragon SoCs which will exist in 1, 2 and 4 core versions and will clock up to 2.5 GHz. The quadcore version is likely to be reserved for tablets and will include a new in-house GPU, the Adreno 320, announced as as powerful as the Xbox 360 and PS3 GPUs.
Competition is looking tough and energy consumption is likely to be crucial on these components as battery life is one of the most important considerations for these peripherals, but also because it will be a determinant when it comes to fixing the final clocks, and therefore levels of performance. On paper, Qualcomm and T1 have an advantage here because they have developed their new SoCs using the 28nm fabrication process. They are however dependent on the development and yields of this fab process and should it prove inefficient, or come late, this advantage may yet turn into an open field for the forthcoming Tegra. Unless of course Intel decides to come to the ball?
One thing’s for sure, the world of smartphones, tablets and other MIDs is sure to be active over the next few years, as you can see by the NVIDIA roadmap which slates a new
super hero Tegra each year:
Those who’ve been following the updates to our APU E-350 test (in French on our sister site HardWare.fr – English translation to be published here shortly) will have noticed that we came across some problems with the acceleration of video playback with Flash 10.2 on AMD platforms. Adobe has just updated its release notes and confirms the problems that we came across:
Hardware decoding of H.264 for Video objects fails to engage on supported AMD (ATI) graphics cards. Hardware decoding of H.264 works correctly for StageVideo objects. Adobe and AMD are actively working to resolve this issue.
To recap, Flash 10.1 already had GPU acceleration. In Windows it was a basic DXVA type, with just acceleration for H.264 decoding. For Flash 10.2, Adobe has implemented the entire DXVA 2.0 pipeline under the marketing name StageVideo. Post processing is now possible (de-interlacing, but also the various filters in the control panels of your graphics card) as well as rendering of EVR type 3D surfaces (only works with Internet Explorer 9 at the moment).
StageVideo on Adobe’s official demos
To use StageVideo, in addition to Flash 10.2 on your machine, the player used by the video site needs to be updated to use the new API. YouTube’s not there yet, nor are most video other video sites. In any case, standard traditional DXVA rendering should kick in, which is no longer the case with the final version of Flash 10.2. AMD has told us that a workaround/correction is on its way.
The DecodeDevice indicates GPU support for H.264 decoding. The ProcessDevice which enables post processing is also available on StageVideo videos. Via DXVA Checker.
From launch, the GeForce GTX 460 was a unanimous hit, with NVIDIA finally offering a decent price/performance ratio. With the arrival of the Radeon HD 6850, which has a similar level of performance, the GeForce GTX 460s were drastically cut in price, making them an even better buy. Mamy models have come onto the market from NVIDIA’s various partners and we wanted to give you a report on them, concentrating on the 1 GB models that we see as offering the best value for money.
> Roundup: 14 GeForce GTX 460 1 GB cards!
Copyright © 1997-2015 BeHardware. All rights reserved..