Hacker News Reader

Comment by 0xcde4c3db | Hacker News Reader

> Supporting the 48Gbps bandwidth is the new Ultra High Speed HDMI Cable. The cable ensures high-bandwidth dependent features are delivered including uncompressed 8K video with HDR.

This is interesting, because I've seen several companies promoting "visually lossless" compression (which is apparently a euphemism for lossy compression that the vendor doesn't think you'll notice) to reduce bandwidth requirements for cabled 8K video transmission. Does this represent some kind of defeat for them, or were they targeting applications where Ultra High Speed HDMI wasn't going to be realistic in the first place (e.g. the various systems that transmit video over UTP cable)?

Also, does anyone know how USB 3.1 SuperSpeed+ cables stack up? Obviously they're specified for 10Gbps for USB devices, but that's in the context of USB's requirements for the attached transceivers, so I'm not sure what it implies about the performance of the cable itself.

View on HN

It also supports DSC[1]. Visually lossless is defined as indistinguishable difference between A/B of predetermined challenging images. You can check the images they used and run the test yourself if you so desire.

[1]https://www.vesa.org/news/vesa-updates-display-stream-compre...

Comparing static images is not really an appropriate test for a video compression algorithm, for obvious reasons. You can write lots of stuff that works on a single frame but looks like garbage on a video stream.

Can you give a link to video source files that could be used to compare the two? The link I turned up looks like it needs specialized testing hardware...

https://www.youtube.com/watch?v=dFbpcBuQg9s

It doesn't involve any inter-frame coding or prediction, so evaluating stills is fair. It compresses a single line at a time.[1]

[1]: http://www.vesa.org/wp-content/uploads/2014/04/VESA_DSC-ETP2...

It might introduce temporal artifacts, which would not be seen in stills.

At these kinds of speeds, why would we use it for just video? We could make server interconnects extremely fast.

Modern HPC interconnects are already faster than 48Gb/s

Those are usually quite expensive. If HDMI requires 48Gb/s capable transceiver chips to be built into every TV, they will get really cheap very fast. It might not be a viable solution in the end, but it could work for some applications.

40 gbps is only Infiniband QDR speeds, you can pick up an adapter for sub-$50 right now.

And plug it into what?

Your servers?

If you mean switching, you don't need a switch for simply connecting two computers. Presumably that was the use-case supported by "HDMI-as-network-layer".

If you watch around, you can get Mellanox Voltaire rack switches for $200 or less. QDR gear is "obsolete" and often surplused at very attractive prices. Everyone else has moved onto FDR or EDR speeds.

Your can get e.g. a qlogic qdr infiniband switch for a few hundred dollars

Over typical HDMI distances, you don't really need an expensive transceiver. 100GbE can be done over direct attach cables that cost less than many current "premium" HDMI cables. Much of the cost beyond that is related to the MAC and host interface being able to do something interesting with the packets.

You can get 40 Gbps infiniband cards for $200 now a days.

The switches are a bit expensive (you can probably find them for 3-4K if you look hard enough), but that's not that expensive for a data center.

Isn't the high bandwidth path for HDMI uni-directional? Not half-duplex, but actually one way only, as in the receiver doesn't have the hardware to transmit back (on those lines). I'm not terribly knowledgeable about HDMI but I think that's basically the case, someone please correct me or elaborate if they know more.

There are some other bi-directional communications channels for control and other features like DRM (HDCP) or even Fast Ethernet but they don't get nearly that much bandwidth.

Reply on news.ycombinator.com