The Intel CE4100 (codenamed Sodaville) was introduced on September 24, 2009 at the Intel Developer Forum. Based on the Atom CPU, it is a full blown SoC. A basic block diagram of the CE4100 is shown below.


 

The x86 core in the SoC is based on the Atom CPU. We suspect that the CE4100 is a slightly modified single die implementation of the Lincroft and Langwell SOCs used in Moorestown, albeit in the 45nm process. While the Moorestown platform had a 32 bit DDR interface, the CE4100 increases this to 64. Other than that, the core components for the OpenGL and video decode support seem to be the same.

The Atom core with 512K of L2 cache is the host processor in the SoC. Though the internal L1 cache details are not public, it is likely to be the same as that of the Lincroft SoC which had 24K of data cache and 32K of instruction cache. Owing to a single die implementation, as well as the process geometry, the TDP of the SoC must be a bit high compared to Lincroft or Langwell taken alone. The higher DRAM bus width also contributes to an increase in the TDP.

The memory controller can support two separate channels of 32-bit DDR2-800/DDR3-1333. This is in contrast to the single channel 32 bit DRAM support in the Tegra 250 which was the earlier SoC under consideration. A NAND flash controller helps the system boot from attached Flash storage. The ability to boot from NAND flash has the potential to reduce the board costs.

Next, we shift our attention to the most interesting gray box in the diagram. This block consists of the video decoder, display processor and the graphics processor. While the block diagram removes any doubt that the graphics processor used is the same as the one used in the iPhone 4 (Imagination Technologies PowerVR SGX535), the origins of the other two components are not entirely clear. We have been led to believe that Imagination Technologies is behind these components as well. The decoder would be a member of the PowerVR VXD series, and it is indicated that two simultaneous HD streams can be decoded. There is also hardware acceleration for decoding JPEG pictures, so one may possibly look forward to snappy photo slideshows in products using this SoC. The scaling, noise reduction and deinterlacing features of the display processor need proper programming in order to be able to deliver good results, and the quality of the Intel drivers would ultimately decide the HQV scores for media streamers based on this SoC. The display controller also needs proper configuration in case a product based on this SoC is supposed to end up supporting native frame rates. It is also responsible for OSD blending, subtitles and miscellaneous video functions.

From the perspective of the Boxee Box, the Audio/Video inputs go unused. However, using this SoC in a DVR / PVR system would make the usage of these inputs necessary. For media streamers, HDMI 1.3a is the key feature. Considering that this SoC was launched almost one year back, the absence of HDMI 1.4 is excusable. For general I/O, we have GbE support, but the Boxee Box only enables 100 Mbps Ethernet. The 2 SATA ports also go unused, but both the USB 2.0 host ports are taken advantage of. Some of the other I/O ports are configured for supporting SDIO. Without a look at the board, and knowledge of the pin configuration, it is not evident which I/Os are configured for supporting the SD card.

The CE4100 also sports a dual audio DSP from Tensilica, the Tensilica HiFi2. It is capable of downmixing / decoding two lossless HD audio streams (as per Blu-Ray bitrate specifications) simultaneously. The Transport Processor would be useful for a STB product. We have been given to understand that the security processor is not disabled in the Boxee Box. This should enable the Boxee Box to access premium online content and also get a license from the Blu-Ray consortium.

Tegra 2 Out, CE 4100 In Analyzing the Boxee Box Specifications
Comments Locked

51 Comments

View All Comments

  • ganeshts - Monday, September 13, 2010 - link

    T2 is infinitely more capable than A4.

    Boxee just didn't want to ship a streamer with such crappy specs :)
  • tipoo - Monday, September 13, 2010 - link

    Exactly, lol...Cant wait for Cortex A9 (what the T2 is based on) smartphones to trickle down.
  • sprockkets - Monday, September 13, 2010 - link

    Except for the fact that you are both wrong, at least on this front:

    MPEG4 is the old "divx" lame codec. You aren't going to find anything HD encoded with this, and if anybody does, they are idiots.

    Apple TV supports H.264 at the main profile (which is just high profile without 8x8 transform) and at LV 3.1, which is defined as 1280x720 at 30FPS at 14mbps max bit rate.

    But let's face it, if 2.5ghz dual core processors from Intel take around 70% of cpu time to decode 1080p WITH the deblocking filter on, what makes you think a 1ghz ARM cpu can, WITHOUT custom decoding hardware?
  • ganeshts - Monday, September 13, 2010 - link

    Both T2 and A4 are aimed at the tablet / smartphone market. No one should expect these app processors to decode HD video on the host processor.

    While T2 uses nV's custom decode engine, Apple uses the PowerVR VXD line (which is the same as that used in the CE4100), BUT, the clocking rate is probably much lower on the A4 compared to CE4100 because it is supposed to be for mobile usage. In addition, DRAM bandwidth available on both A4 and T2 is probably not good enough for high bitrate Blu-Ray material. (They have 32b DRAM interface, while CE4100 has 64b DRAM interface).

    As for T2 vs A4, my comment was mainly on how powerful each SoC is. (On paper T2 will beat A4 hands down). As for comparing the exact video decoding abilities, it is not possible to comment for 2 reasons:

    1. Tegra 2 hasn't reached any reviewer yet.
    2. Apple will never allow any stream into the A4 device unless it goes through iTunes, in which case, it gets auto-transcoded.
  • sprockkets - Monday, September 13, 2010 - link

    We can if nvidia simply tells us what profile at what level it can do. At least Apple, of all people, post what exactly it can and cannot play.

    Even later firmwares of the old Cowon A3 claimed it could play HP h.264 video at 1280x720 with TI's OMAP from like 2 years ago.
  • ganeshts - Monday, September 13, 2010 - link

    nV has no obligation to tell that except to customers who want to use T2 in their products. Apple, on the other hand, has to, because they are selling a product directly to the consumer.

    Anyways, nV's marketing team seems to have screwed up here.

    I haven't tested Cowon A3 personally, so can't comment on that.
  • mindbomb - Monday, September 13, 2010 - link

    Tegra 2 can play 1080p30 baseline profile level 3.1
    decent for a phone, embarrassing for an htpc.
  • mindbomb - Monday, September 13, 2010 - link

    No, mpeg 4 part 10 is h264.
    mpeg 4 part 2 is what you are thinking of, but it is pretty unpopular outside of people transcoding their dvd's, so when people say mpeg 4 video, its implied that they mean h264. (especially if its apple saying it)
  • mindbomb - Monday, September 13, 2010 - link

    oh wait, they mentioned it after a section specifically detailing h264 playback. You were right, they were referring to mpeg 4 part 2.
  • sep332 - Monday, September 13, 2010 - link

    Isn't CIFS effectively the same as SMB?

Log in

Don't have an account? Sign up now