Comments Locked

23 Comments

Back to Article

  • eastcoast_pete - Wednesday, June 19, 2019 - link

    @Anton: There is 8K video available, at least enough for demo purposes. If, as you report, a static image led to the Realtek's chipset getting quite toasty ( "...how hot the SoC's heatspreader got"), I actually doubt that they simply couldn't find a short 8K demo loop to show off their new creation. A video decoding and processing chip running hot with a static image doesn't augur well for later use with actual video. Rather, it looks like the RTD 2893 is not ready for prime time (or anything else in 8K), and the static image was all this chip can handle at this time.
  • nathanddrews - Wednesday, June 19, 2019 - link

    I think you missed the part about it not being final, but rather a demonstration of a chipset they're working on for release in 1-2 years.
  • Santoval - Wednesday, June 19, 2019 - link

    That's beside the point. The "chipset not being ready" and the mentioned "scarcity of 8K content" are two entirely distinct things.
  • Azethoth - Thursday, June 20, 2019 - link

    I think you missed the part about it not being final, but rather a demonstration of a chipset they're working on for release in 1-2 years.
  • mode_13h - Friday, June 21, 2019 - link

    Um, are you the same person that just posted as @nathanddrews? If not, you certainly share his issues. See above.
  • B3an - Thursday, June 20, 2019 - link

    "Realtek ultimately brought the RTD2983 to Computex to show potential customers that it's ready, and that interested parties can start designing their devices based on the SoC"

    Seems very much ready to me.

    It's the actual consumer products that will not be ready for a year or two. So if this finished SoC is getting hot from simply displaying a static image, then that's just embarrassing. There's also a lot of 8K videos online if you look around, including on YouTube, so it seems as if they purposely showed a static image because they didn't want this thing melting...
  • levizx - Thursday, June 20, 2019 - link

    There's no way they can do 8K (or even 4K) AV1 efficiently right now, it has to be using some sort of FPGA instead of the actual final IC.
  • mode_13h - Friday, June 21, 2019 - link

    That was my thought. I was surprised to see something already doing AV1 decode @ 8k.
  • mode_13h - Friday, June 21, 2019 - link

    Are you schillin', dude?

    1. The article clearly gave the reason as being a lack of content.
    2. The article clearly stated:

    "Realtek ultimately brought the RTD2983 to Computex to show potential customers that it's ready"

    As for 1-2 years, that was the projected time frame in which end devices were expected to ship. Nowhere in the article did it say anything about this SoC being "not final".

    This was not a long article. You've either got serious reading comprehension issues or some other kind of agenda. Either way, you lose.
  • Kevin G - Wednesday, June 19, 2019 - link

    It depends on how it handled a static image. Many of the real time video-over-IP codecs leverage M-JPEG/M-JPEG2000. The result is a constant ~900 Mbs data stream regardless if the actual content is a still image or full motion video. The encoding algorithm is 'dumb' in this sense that it can't detect and highly compress still images while maintaining the necessary latency desired.

    Realtek could be leveraging a similar scenario as I described above but there is not enough detail to determine either way. This *could* be perfectly indicative of how the platform will behave with video.
  • Azethoth - Thursday, June 20, 2019 - link

    Also it literally said that it is able to decode video. So I think the onus is on disbelievers to prove that, bizarrely, they are lying about being able to do video and that somehow they will fool 3rd parties into buying a SOC that can only do still pictures. Some truly thoughtless comments so far.
  • mode_13h - Friday, June 21, 2019 - link

    > the onus is on disbelievers to prove that, bizarrely, they are lying about being able to do video

    Hey, if you want to buy early silicon based on specs alone, be my guest. What we know is their demo very much falls short of their claims, and the stated reason doesn't add up. I'm not saying they can't ultimately deliver a SoC that meets their specs, but I think the onus is very much on Realtek, at this point.

    > Some truly thoughtless comments so far.

    Yeah, like the idea that anyone is going to make an on-the-spot purchasing decision, as you imply.
  • mode_13h - Friday, June 21, 2019 - link

    Yeah, it's a bit of a fallacy that still images are inherently cheaper to decode. If it truly was an AV1-encoded keyframe, then it's potentially more expensive to decode than a typical P frame.

    However, the thing you can't tell from a still image decode is just what framerate it's managing. That could be the real reason they didn't show anything involving motion.
  • Santoval - Wednesday, June 19, 2019 - link

    "or anything else in 8K"
    As you can read in my comment above there are plenty of cameras able to shoot perfectly 8K video. TV sets able to output 8K content have also been demonstrated. However encoding and decoding 8K video with AV1 is an entirely different thing. H.265 can also be used (but *not* H.264, it doesn't support 8K), but 8K probably pushes it to its limit in terms of efficiency, while it's notorious license issues are a deterrent.

    Beside AV1 there is also the upcoming H.266/VVC (Versatile Video Coding), which is the actual successor to H.265/HEVC. This will not be royalty free (obviously) but it will have much more simplified licensing terms (similar to H.264). It will not be finalized before late 2020 - early 2021 though, so AV1 has quite a head start. A possible codec war is brewing...
  • nandnandnand - Wednesday, June 19, 2019 - link

    Hopefully we will see solid progress on AV2 before VVC hits the scene. MPEG needs to lie with the fishes.

    https://rethinkresearch.biz/articles/ao-media-look...

    There are features that didn't make the cut in AV1 for various reasons, including too much increased complexity. With AV1 already beating HEVC, AV2 could be their chance to throw everything and the kitchen sink in.
  • GreenReaper - Wednesday, June 19, 2019 - link

    AV1 for real is big news, if it's being hardware decoded. Apparently the RTD1311 does the same for 4K:
    https://www.realtek.com/en/press-room/news-release...
    https://www.realtek.com/en/press-room/news-release...
  • nandnandnand - Wednesday, June 19, 2019 - link

    Yeah, this is the first product I've heard of that will support AV1. We need to see more of that.
  • GreenReaper - Wednesday, June 19, 2019 - link

    There have been others, apparently, but this is the first 8K:
    https://en.wikipedia.org/wiki/AV1#Hardware

    I'd be glad to see Level 5.1 decode capability on a AMD "4600G" Zen 2 7nm. Little point in 8K content I can't view and probably can't stream (4K 60FPS Main is up to 40Mbit): https://en.wikipedia.org/wiki/AV1#Profiles_and_lev...

    It'd match what they offer for other video formats - unless this changes with Navi's codecs:
    https://en.wikichip.org/wiki/amd/ryzen_5/2400g#Gra...
  • nandnandnand - Wednesday, June 19, 2019 - link

    So 3 products with AV1 decode, including another RealTek SoC announced a couple weeks ago. Not much.

    As for AMD Navi, it looks like it will not decode AV1:

    https://www.techpowerup.com/256481/amd-navi-radeon...
    https://www.techpowerup.com/img/cf96IBWB2r9zYZ25.j...

    I don't know if it will be any different for laptop or desktop APUs, but that doesn't look encouraging. Looks like we'll have to wait for Zen 3 and "7nm+" GPUs if we want AV1 hardware decode support.
  • GreenReaper - Thursday, June 20, 2019 - link

    I'd be a tad surprised if the video codec block can't be upgraded separately from the rest of the graphics hardware. It's its own dedicated subprocessor. Case in point: Raven Ridge, which added hardware decode of VP9 (a big deal for laptop battery life given YouTube's preferences):
    https://www.x.org/wiki/RadeonFeature/#index9h2

    There was other Vega graphics hardware out, but it didn't have that feature; some may still not.
  • Santoval - Wednesday, June 19, 2019 - link

    "Since Ultra HD content is scarce, Realtek demonstrated the RTD2983 platform at Computex using a static picture encoded using one of the aforementioned codecs."
    While the content itself might be scarce the equipment to shoot at 8K is not. RED already has no less than three "brains" (sensors) that can shoot at 8K with RED cameras, Panavision has the Panavision Millenium 8K DXL (with a RED sensor), Ikegami has the Ikegami SHK-810 8K camera, and even Canon is about to release the Canon Cinema EOS System S35MM 8K camera.

    While the above cameras are out of reach for the average Joe, Realtek could have easily rented one of them for the demonstration, or -why not- even buy one (except the Panavision, that's rent-only) to employ as a testbed for their 8K platform.
  • mode_13h - Friday, June 21, 2019 - link

    Yeah, considering the development costs for such an ASIC, claiming lack of test material suggests an extreme lack of resourcefulness. They could've probably even asked some of their customers or partners for 8k clips.
  • mode_13h - Friday, June 21, 2019 - link

    > Since Ultra HD content is scarce, Realtek demonstrated the RTD2983 platform at Computex using a static picture

    ...or, the chip can't currently handle decoding full motion @ full framerate.

    It's somewhat unthinkable they could bring a custom chip to this stage, without ever having run any full-motion clips with any of the target codecs. I mean, you could even render & encode some Blender scenes yourself, if you had to.

    But they could also just rent an 8k digital cinema camera and shoot some scenes with it. It's really unbelievable that "they just have no clips to display". I smell a rat.

Log in

Don't have an account? Sign up now