Ask HN: Will consumers ever have a use for 10Gbit internet

24 points by Gigachad ↗ HN
I had this discussion with a friend. While 10Gbit Ethernet seems to be slowly dripping in to consumer hardware, it seems much less clear if users will ever actually have a use for 10Gbit networking.

If we look at the biggest network uses for consumers right now, its mostly video streaming and games. A 1 Gbit connection can already stream video at well above the maximum desired quality we will likely ever want. And better codecs are coming out which reduce the networking requirements even more. For video games it seems that large games would go with an asset streaming model like microsoft flight sim which can easily be done with 1 Gbit.

Are there any future developments that might make 10Gbit desirable in the consumer space?

45 comments

[ 3.5 ms ] story [ 57.1 ms ] thread
I can potentially see VR create a demand for such high bandwidth. With high resolution fully immersive videos perhaps
You won't need anything more than Stadia for that

> An internet connection speed of 10 megabits per second (Mbps) or greater is the minimum recommended to use Stadia. A slower network speed can cause issues while you play games.

> To play at up to 4K resolution, you'll need an active Stadia Pro subscription and a network speed of 35 Mbps or greater. Check below for hardware requirements.

That's the current requirement for the compression they use. I'm sure that there's a lot of heavy optimizations under the hood to support 4K at 35Mbps, which is most likely lossy as well.

Bluray video requires about 144Mbps at the highest quality right now. A bulk of this can also be audio formats with spatial information, like Dolby Atmos, or DTS:X.

VR might need even twice that amount for both eyes. Combine that with more dense sensors, probably improvements in processors and memory allowing better lossless decoders, and you can cross 1Gbps easily. I can also see multiple users at home needing this at the same time.

100mbps h264 is fine. In fact, for existing content we got mostly indistinguishable results at 50mbps. H265 gives you another 30%.

Now there might be people who can notice the difference but in our limited human studies testing on in house VR employees they couldn’t (try it out yourself if you have an Oculus link - the Oculus PC tool lets you adjust the bandwidth live if I recall correctly). You do want to start initially with a high bandwidth so that the initial P frame gets the most detail and then you can decrease the bandwidth which applies to the subsequent I frames.

Obviously VR resolution keeps increasing every year or two so this will grow but you’ve got other techniques that combat this like ML supersampling.

I think an overlooked piece of 10GBe is lower latency and more consistent performance as packet scheduling is the most challenging aspect of streaming VR. There could also be other use cases where higher bandwidth is more critical but video streaming VR likely isn’t it.

VR Streaming is impossible. The latency is nausea (and vomit) inducing. I don't mean that as an hyperbole, it really is.
That sounds like you're still streaming in traditional 2D with a static viewpoint being rendered somewhere in the cloud.

Full pre-rendered VR streaming would need to ship video of the entire environment for both eyes (or a reasonably large subset), and the client would be responsible for rendering your point of view. There'd be no latency that way, but the bandwidth needs would be significant given 4K per eye and a 90-degree FOV. That's on the order of 16K per eye, or 32K @ 120Hz video for static location. Add on some extra bandwidth for volumetric video to support position tracking, and you're looking at some even sillier numbers. Granted, a headset might spontaneously combust with that kind of data rate.

You can only accelerate your head so fast. Whatever resolution you pick for your headset, expanding the area by 50% is probably more than enough to let the client compensate for movement.
Cool to hear that what I do almost every day is impossible :)

AirLink actually works really well for me, but it does need a clean 5Ghz channel with nothing else on it. Any other device that is connected (even if it's not doing anything) will cause stutters. I have a separate Unifi 6 Lite AP hanging right above my head for it with a separate SSID.

Are you streaming across the room or actually from a datacenter? The impossible part seems to be that there is some latency that is pretty much unavoidable when you stream from 500km away.
Across the room normally but I signed up with plutosphere as well which offers VR streaming from the cloud. I have yet to try it because they have some capacity problems, but initial feedback from users that got in were good. I want to try it mainly to evaluate and for travelling as I already have a great game PC.

But latency does not seem bad enough there to worry about. Not everyone is as sensitive to that anyway and once you have your VR legs it becomes even easier.

I prefer local computing over streaming anyway. But like I said for travelling it might be a great option.

Do cars ever get driven at their maximum rated speed, no, but its nice to know its possible and to have the choice to do so.
? What are you comparing it to.

Driving cars at maximum speed is literally multiple sport branches ( F1, rally, ... ) and obviously it's testosterone inducing :p

Max. Bandwidth ... Not so much

Ever? Yes, of course. Right now ? No.

10Gbit/s Internet connectivity sounds like a lot, but it's only 10 times the speed of what we have now.

Think about RAM, CPU frequency, non-volatile storage, and graphical rendering capability these have increased orders of magnitude, as have Internet connection speed.

I don't know what it will be used for, but I'm certain that new use-cases will reveal themselves as that speed becomes broadly available.

It will certainly allow new business models, for things like storage of raw video and video editing, but broadly, I think it's unexplored territory.

Its about 800 times faster than what we get in Australia and call "Broadband" (on a really good day)
There's plenty of Americans that are getting 1/1000th of 10Gbps, and plenty more that would be thankful if they could even get that much, too.
>but it's only 10 times the speed of what we have now.

The average fast internet speed most people have right now is 100mbps and there are vanishingly few use cases which require more than this. Pretty much all video streaming including multiple users is easily covered by 100mbps. It comes down almost only to video game downloads where you want it to download in 5 minutes rather than 30. But it seems unlikely that games would gain all that much by expanding their file sizes by 100x without moving to a streaming model so users don't have to fill their storage with content they will never see.

I am more interested in a low latency connection, which then can change everything.
Unfortunately there isn't anything that will improve there, at least not significantly. We are limited by the speed of light, can't really go around that
The last couple dozen miles tend to have many milliseconds of latency, none of which has anything to do with the speed of the signals.

At least you can improve some of that by significantly overspeccing the last few hops.

With DSL you can even get fun 1st hop latencies like 30ms...
What about 3-5 people doing their own streaming on a single line? 1Gbit, especially a busy one, could be not enough.

Imagine one of them downloading huge steam games every other day while others live stream.

I know it's pure luxury, but when I had a 10gbit I enjoyed having it.

You could have 3-5 people streaming Blurays on a 1Gbit line

Netflix ? A lot (>20) of people fit on a 1Gbit line

Once they finish convincing sheeple that they need 8k or 16k or 128k TVs to perpetuate the endless cycle of obsolescence, perhaps.
Streaming networks are perpetually trying to spend less on bandwidth. I already prefer a full-HD blu-ray to 4k Netflix which has more distracting compression artifacts. Just increasing the resolution to 8k and then compressing it in a 10Mbps stream will not increase bandwidth requirements (or quality for that matter).
As a media producer / video person, 10gbe has become pretty essential and commonplace. The idea of having 10gbe internet at home would change the entire work from home / media industry.
Uploading raw videos from a DSLR, or data sets for photogrammetry.
I already do. I wouldn't use the full 10Gb bandwidth yet but definitely more than 1.. And I don't want to invest in stopgap tech like 2.5Gb.

The problem is the switches are still too expensive :(

I've been playing with streaming app/game delivery techniques, and can see where a 10x pipe might help a lot.

There is a tradeoff that has to be made when you use an efficient interframe compression algorithm (i.e. x264 or mpeg). Latency becomes a factor, and you lose the ability to instantly seek/generate a given frame without keeping track of the ones that occurred prior. You also need specialized hardware or high performance CPUs on both sides. This also has issues tolerating high frequency visual changes between frames.

When using an intraframe technique (i.e. jpeg), none of the caveats apply, with the exception of 1 new one - The amount of bandwidth required is increased substantially if you intend to serve a sequence of these approximating a video (a la mjpeg).

I've had a lot of people give me grief for this idea, but I assure you a sequence of jpegs at 60fps looks like some kind of magic butter compared to the same encoded with more efficient techniques. Especially, when you put a human in the loop with an input device. You literally don't have to worry about the type of content at all. It just looks good always. The range of devices that can quickly encode (<10ms) jpeg at 30+ fps is surprisingly large (basically anything with a cpu made in the last decade).

I've almost memorized the most important bits of libjpegturbo by this point. Can't say I've even begun to scratch the surface on the likes of AV1, etc. Don't think I'd really like to just seeing the shadow of the damn thing. This kind of simplicity has benefits that can be hard to quantify.

Any time I can make some tradeoff like this, I like to think into the future. Is the speed of the average connection going to increase or decrease from this point? Do we need 100% of the planet to have 10GbE before we can begin exploring these ideas?

I'll note that digital cinema, using a series of jpeg2000 frames, is capped at 250Mbps. At that quality level, done in real time, at 4k, you might need more than 1Gbps per user.
> well above the maximum desired quality we will likely ever want

we can look at computing history and see there hasn't been a limit to what people will want. However much processing, storage, or bandwidth we had at the time, there was always use for more.

More things will be offloaded to the server
Keep in mind, what you're really asking is "will consumers have a need to use over 900Mb/s pipe" , In burst... yeah, I can see that being needed in the next few years with streaming+gaming+p2p , sustained? Not in the next ~5yrs af least.

Even if the need is there, ISP+Transit cores are not up to it. But it does make a lot of sense to have a 10Gbe last mile with 500Mb/s sustained(without oversubscription) and 10Gb/s burst usage. Especially in metro-areas.

Throughhput drives demand. Customers aren't doing stuff that needs 10Gb/s because they don't even have 1Gb/s, not because they wouldn't if they did.

Personally, 802.11n and whatever crappy DSL link speed I get is good enough for me, but I don't do video, social media, or online gaming. I can rent access to machines much better positioned to take advantage of greater bandwidth if I ever need them.

> Throughhput drives demand. Customers aren't doing stuff that needs 10Gb/s because they don't even have 1Gb/s, not because they wouldn't if they did.

I think that effect diminishes as speed goes up. Network speed is just one constraint that might limit me from doing things I want to do. Others are time and money.

I think we are approaching the point where network speed will be sufficient for most people that it is no longer a constraint. At that point faster networks won't really matter to most people.

From the comments elsewhere in the thread there are plainly some edge cases where it's actively useful.

Personally though I'm still on ~60Mbps download at home FWIW and feel simply no need to upgrade. Whole family can watch - independently - Netflix/Youtube/etc. No issues WFH while they do so. I'd find the upgrade to 100Mbps or 200Mbps quite cool from a tech point of view, having started with dial-up 25 years ago, but honestly I have no real need for it.

UltraHD - 4K - needs 25Mbps or thereabouts, remember.

In the consumer space the trends I'm following are more aligned to the various 5G spectrum roll-outs. I think there's an approaching inflection point where consumers will be routing (or be QOS'ed) by their Telco ISPs to either a fiber network or wireless.

I imagine telco consumer routers becoming a MEP.

Right now, we have things like Android Studio and npm which are absolute bandwidth hogs. I'm sure a use will come for more bandwidth. I never thought I'd need anything past 1 Mbps, but it can be the bottleneck for build times now.

One thing I didn't have enough bandwidth for was doing a presentation on Google Meet to a group of people, presenting a 4K screen. I suspect it does several 1 to 1 connections, rather than broadcasting, which would explain the drop in performance when presenting to a group.

Symmetric connections (upload_speed = download_speed) have much more utility imo, and we should focus on getting that in the consumer-space before we focus on 10Gbit.
Probably. Do you use more than 640KB memory?

Having more bandwidth makes new things possible. So the question isn't what do people do now that can make use of 10Gbit, but rather if everyone had 10Gbit what could be done that we can't do now.

Things which currently blocked by not having 10Gbit internet (not saying that this is the only blocker):

- videocall with hundreds of people where everyone can see everyone else

- higher definition, extremely realistic VR

- instant transfer of gigantic files. I work as a data scientist and I'd love to just slack someone the result of a quick and dirty experiment on a 50GB dataset. It sounds ridiculous now, but it isn't.

- real-time collaborative editing of gigantic files (video, sound, tabular data).

what is the use of having a video call where you have hunders of people seen each other?? even with a 4k display you cannot display more than couple dozen people at time..

more the a couple dozen the videos get so small they are useless, at this point it is better just go audio..

usually you are only interested in the video from a couple of people and usually one at time.. and software we have today can already handle it with connections below 100mb..

i dont think you need that much for VR.. you probably can get good realistic vr with 1gb..

the last two are specific needs for specific fields, i hard say it is something that fall under consumers needs

I imagine it can be useful to be able, in VR, to see more than a dozen people.

For example, a VR concert where you can attend with a ultra-realistic avatar?

Put youtube level infrastructure in every pocket and lets see what other man-made horrors we can come up with.