WiFi 6 gets 1.34 Gbps on the Raspberry Pi CM4 | Hacker News Reader

WiFi 6 gets 1.34 Gbps on the Raspberry Pi CM4

jeffgeerling.com

311 pointsgeerlingguy204 comments

> Have you ever started a project that should take a couple hours with a fifty dollar budget, and realized at the end you spent a whole month on it and spent close to a thousand bucks?

Ah man, costs aside, that's my last weekend and probably the next few ones as I've been trying to make my old Motorola Xoom tablet run a modern kernel. When/if that's completed, the return on investment will be a net loss.

But the experience may net you something... an ounce of pride, or maybe a pound of regret, depending on how it goes!

Definitely! Already to have successfully compiled a working kernel is an achievement as most of my last attempts over the years failed. But I really want to satisfy those demanding user-space binaries who complains the kernel is too old.

It's quite satisfying to have grown from "struggling to build from source" to hunting 2.36 era patches and diving in the code to port it. A lot of actual learning in the process.

It'll all be worth it when you post about your hard work on hacker forums and get a deluge of angry comments telling you how you could have gotten thrice the power in half the time for 1/5th the cost.

>net loss

Well, you could have just sat on the couch watching mindless tv instead.

I'd definitely read a writeup on this. Have you seen postMarketOS?

Nice! Didn't knew postMarketOS. This[1] is mostly what I'm trying to do and their wiki make me think I might have underestimated the task. haha.. This seems to be a good place to try to contribute!

If I can convince myself to not fall into the rabbit hole of rebuilding my blog first, I'll definitely write about my porting efforts :)

[1] https://wiki.postmarketos.org/wiki/Mainlining

A trick I've used is just start writing in plain text, then fix it up into Markdown later. If you've got a kernel building you're already way further in than most. Give me a yell if you need someone to bounce ideas off.

Can you expand on the process? Why did porting the kernel cost you $1,000?

I meant except for the monetary part. Let's say, I've put about 20h in it and I expect it takes me 3 times that. Just counting the time put in that kind of project, the outcome itself isn't worth it.

Considering this is a 10 years old android tablet that cost at the time ~300$. I probably could have bought directly a more modern one which is known to be flashable with linux.

The sentence I cited resonate for me as I often dive in a project and follow in the rabbit holes even though I know it's a bit irrational to do it.

About how to do it: - you first have to find every git repo that may have relevant patches for this or similar device. There's the one open-sourced from the manufacturer, Google has one dedicated to the Tegra chipset, with more recent stuff but without any specifics about that device. - determine what parts are relevant: definitely the board definition, the drivers. But then, that file weirdly named which appears to handle interrupts? Well better take it. - adapt to changes in kernel interfaces: I've only come across name changes for now, but you can imagine it gets more complex - finally: compile, flash, test! And probably keep doing that for a few cycles.

That's for the reply. I was thinking you meant labor costs but I wanted to double check.

I have a number of similar projects and can relate to how easy it is to follow rabbit holes.

A bit of rant, but the observations in the article just confirm what I have felt about WiFi in the past decade.

> When I tested the Compute Module 4's onboard WiFi 802.11ac chip, I got up to around 80 Mbps, even when using an external antenna.

> But when I tested the speeds, I was disappointed to find the upgraded card could only pump through 300-400 Mbps on the XPS 13, due to the two-by-two antenna performance and configuration on the Dell. My MacBook Pro, which doesn't have WiFi 6, can get speeds up to 900 Mbps over 802.11ac, so the design of the entire system is extremely important.

The landscape of WiFi technology is a complete joke. The consortium and the manufacturers market this crap about 802.11ax being faster, but that's only under ideal, lab conditions. In real-world environments, it's worse than black magic that sometimes works and sometimes doesn't. 5Ghz has such terrible penetration power that it's almost useless. You can barely use a single 5Ghz router to serve a 70m^2 home at acceptable performance. Unlike many houses in North America, homes in Hong Kong are typically built in concrete with rebar, and 5Ghz signals just get absorbed to oblivion within 3 to 5 meters. If you want top WiFi performance in your whole home, you need multiple access points and wire through the wall. But wait, what was the promise of WiFi again? Less wires! The irony!

I completely agree! In a not-overly-large flat in Hong Kong, we have three APs to provide complete coverage, and only two of those are connected via Ethernet. The third is a repeater!

~10 other networks are also visible; the density hurts performance as well.

A good "secret" is to use the DFS frequencies most routers won't choose by default. High end consumer routers will allow you to select these frequencies, and usually there's no nearby interference.

I have ~70 nearby networks visible but none on my DFS channel. Your signal can drop if the AP detects radar but unless you're near a weather radar station or airport it works fine IMO

I dunno, a lot of commercial APs for hotels+such just spam the spectrum, creating a BSSID for every FCC-authorized channel. Like one box with 30 MAC addresses, one for every possible channel. It's crazy.

I guess that makes sense. I still haven't seen many using the DFS frequencies though. Presumably because they don't want to deal with testing related to avoiding radar.

In my area every channel is saturated except for DFS.

If you're desperate there is always that Japan-only WiFi channel (assuming you aren't in Japan) that no other country has approved use of... almost all WiFi gear supports it under Linux.

You didn't hear that from me.

How it works depends on your router, when it detects radar, my router just switches to one of the congested channels and never switches back to the DFS channel, which makes it rather impractical to use.

Having multiple access points is perfectly reasonable. I don't think it's fair to blame the Wi-Fi spec for this.

If you want anyone to blame, maybe blame the ISPs or router manufacturers that market the idea that a single router can cover multiple rooms because that's just not possible with nowadays' interference levels and bandwidth requirements.

Having multiple access points should be considered normal and expected. Maybe we should understand and embrace the laws of physics and their limitations instead of trying to pretend they don't exist?

Sounds about right for a technology that operates on unregulated frequencies - as Shannon's law dictates that there is only so much bandwidth in the air. The primary reason for bad wifi in any densely populated city is not construction material but the two dozens of routers trying to yell on top of each other.

I’ll take an AP in every room gladly, for me it’s about less wires to the individual devices not less wires throughout the house as a whole. But it does suck if you live somewhere that doesn’t have wired Ethernet ports on every room.

However I’ve found that mesh works surprisingly well these days. With 3 access points I’m able to cover ~2200 sqft with no problems, and it works well enough that I haven’t bothered setting up wired backhaul even though I have Ethernet ports in every room.

Short range of wifi is intentional, unfortunately. They picked frequencies absorbed by water vapor in the atmosphere. On top of that, they picked high frequencies that cell carriers don't want because they have little penetration.

The short range is intentional to allow many networks function in dense areas. Multiple AP's is really the only option

> On top of that, they picked high frequencies that cell carriers don't want because they have little penetration.

This is true somewhat, but frequencies in the 2-6 GHz range are pretty common for LTE, and cell carriers have spent many billions on the licenses.

They picked frequencies that were unlicensed. And at time 2.4GHz was likely good balance between bandwidth(frequency) and available technology.

And the 2.4GHz was free because microwave-ovens used this frequency. As those licensing bands specially in 40s/50s would not have wanted to deal with potential issues. Thus it was designated for microwave-ovens. And later freed up for any use. Thus being available and free.

Too late to edit my comment but you're right. Apparently 2.4 being chosen for microwaves because it excites water is a common misconception. It was just a trade off between penetration and difficulty producing high frequencies back then

> 5Ghz has such terrible penetration power that it's almost useless.

To a degree, that is a good thing. High density living spaces (apartments) end up with so much interference on 2.4GHz that it becomes unusable.

For reference: a few years ago I lived in an apartment where I counted at least 35 different 2.4 GHz networks. This wasn't even a high-rise but in a street with normal 3 story apartment buildings. Wifi only worked on channel 13 reliably because many access points don't include channel 13 in their "random channel" configuration because it's only available in Europe.

try wiring your home with enterprise grade APs ... the stability and performance when multiple devices are connected is vastly superior to anything consumer. And that's not only because the APs are better designed but because most firmware on consumer equipment is total crap based on old outdated sdk. On the other hand enterprise APs spend a lot of time optimising the AP even after launching it.

Do you consider Mikrotik enterprise grade? Because that’s what I use at home.

Specifically, I use this: https://mikrotik.com/product/rb4011igs_5hacq2hnd_in

It’s not AX, but still does N and AC over 5Ghz nonetheless.

I kind of don't get people discerning things into "enterprise," and not grades.

All mass market electronics today is made of common list of chipsets for a device.

There is no way to change fundamental limits of hardware being used. So, the best thing to look at is the chipset, and its specs.

The best example of this is the scam called "enterprise servers" whose only distinction from no-name OEM ones is that "enterprise server" name, and 3-4 times bigger price. They are all made of sometimes 1-to-1 identical motherboards, chassis, and other components.

> The best example of this is the scam called "enterprise servers" whose only distinction from no-name OEM ones is that "enterprise server" name, and 3-4 times bigger price. They are all made of sometimes 1-to-1 identical motherboards, chassis, and other components.

It's more than just the parts used:

* Support contracts

* On-site, next-day or even same-business-day technicians

* Guaranteed parts availability

* Often from a company your org will have a procurement contract with already (Dell, HP, etc).

Mikrotik consumer routers are quite decent. A hAP² handles my apartment just fine.

I am yet to try Mikrotik but I have heard good things about it.

There's a lot of magic around MIMO, beamforming and other stuff where Mikrotik is kinda behind other manufacturers. Enterprise grade here would mean something like Cisco WAP or HPE Aruba.

Surprisingly, the "prosumer" things like more expensive Asus routers work better than you would expect.

If you want a deep dive, I suggest this page https://www.duckware.com/tech/wifi-in-the-us.html

I'll admit I'm a Ubiquiti fanboy now. I got fed-up with consumer-grade wifi gear when my old TP-Link AP+Router combi crapped out after ~20 wifi clients connected. Initially I kept the router but disabled its wi-fi AP and started using Ubiquiti APs - just one to begin with, but it handled 50+ simultaneous wi-fi clients like a champ - 6 years later I now have a UDM, PoE switch, LTE failover, three APs, and I'm considering moving-away from Google Nest to Ubiquiti for home-automation now too, because Google/Nest's video service makes no sense considering they don't let you access the video locally or save the video locally (what's the point of a video surveillance system that can be defeated simply by snipping the Internet coax connection outside?!?).

I’m a fan too! The dream machine, aps, and a poe switch were cheaper than what I’ve paid for just a single “enterprise” device before.

Ubiquiti's gear falls into a weird market segment: it's got 85% of the capability of true Enterprise-scale equipment (i.e. the kind you absolutely have to employee someone - or two, plus pager-duty, full-time for) for a fraction of the cost, but costs 2x or 3x more than the consumer-crap it's replacing - but has a _hint_ of vendor lock-in that at least with the full-time-employees you could task them with building and implementing an exit migration strategy.

My current APs are old and I got fiber recently so I am looking for cheap AX APs if they are worth the price. I know AX is still early but because I am in the market and both my daily devices are AX I might as well try (Unifi 6 Lite or LR though both don't support AX in 2.4GHz so something with 160MHz in 5GHz and AX in 2.4GHz). If you buy consumer AX now forget updates in a year or two. On the other hand the first AC Lite still receives updates and that board has gone through more than 30 revisions (I recently bought one for someone and that had board revision 36 or something). Sadly I haven't had the same luck with their wired routers. The EdgeRouters are running some old Debian and I have had trouble with their PPPoE client. So I flashed OpenWRT recently on it but it didn't run great (for my fiber) either. I am looking for either an R4S NanoPi or one of those protectli devices for just routing. I think in the router part these devices are not cutting it for me (I can do things like dnsmasq blocklisting and stubby with openwrt and cake sqm works great). Anyway I need to plan my WiFi 6 Fiber upgrade well otherwise its easy to get carried away and blow up the budget.

In high density housing the low range/penetration power is a positive. Having lived in an apartment building where all the neighbors (in three dimensions) also had 2.4ghz wifi I can tell you that having to find a frequency band that isn't congested in that situation is a hassle. And I suspect that the knowledge that congestion and selecting a frequency band is even a thing is not very common.

I’m not sure about Hong Kong, but in Japan the allowable transmit power is much lower than the US. Anecdotally, I find 2.4Ghz is usually not usable due to so many networks coexisting but 5Ghz works well. Interior walls are mainly wooden though.

My complaint is about truth in advertising. Manufacturers should not be able to add the max speed of multiple networks and claim “5800Mbps combined speed”

Hard drive manufacturers advertise unrealistic speeds that can only be achieved in synthetic benchmarks. CPU manufacturers advertise CPU speeds that cannot be sustained under large load without dousing entire thing in liquid Nitrogen.

That's why you go and read benchmarks on the Internet, to see how the devices fare in more realistic conditions.

Hold on, buddy.

> You can barely use a single 5Ghz router to serve a 70m^2 home at acceptable performance.

Of course you can't. Compared to 2.4GHz, 5GHz barely passes through walls, and it is actually a good thing. My rule of thumb for 5GHz is to have access point every other room and to lower transmit power so that they don't in the way of each other.

In effect, there are 3 APs in my 160m^2, two storey flat. I get easily minimum 400-500MBit/s on AC with multiple users in any spot in the flat but I don't use typical consumer hardware. To do 5GHz properly in a large home you need something that will be able to properly roam users while they are moving about.

Where I live I see about 250-260 2.4GHz access points at any point in time and 2.4GHz is almost useless. I only see 3 5GHz APs. The fact that 5GHz is so attenuated by the walls makes it usable in high density environment.

I have also found that one shiny laptop that was more pricey than others had trouble with lower speeds and constant disconnets. LOWERING transmit power actually helped, because it caused it to be kicked earlier from AP and switch to another, closer one.

So, to underline, you need to know technology you are using. WiFi requires a little bit of knowledge because you want to realize a miracle of passing huge amount of data at high speeds through walls. I kinda wonder why everybody complains they can pass signal only through single wall.

Are you sure the frequency is the problem? I’m living in an environment with lots of radio polution in a four story building built from concrete and rebar. I have 802.11ac wifi and I get at least 300Mbps in the worst place in my apartment.

Other argument might be FPV video. There are quite a few youtube videos comparing analog video streams over 1.3 to 5.8 GHz. The difference is there but much less than I expected.

Absorption by walls is a net positive, means you can have isolation from adjacent networks in dense r̶u̶r̶a̶l̶ urban environment.

Urban, I believe you meant to say...

Ignorant questions, if I may:

1. How come today, for home use, WiFi seems to exceed the price/performance of hard-wired? Is the Ethernet standard lagging, or lack of interest/demand, or some more physical barrier?

2. Related, for somebody who wants reliability, is Ethernet/wired still a sane choice, or does that just make them an old geezer? In urban setting with overlapping wifi cards all blasting full strength at their neighbours, is real-life wifi performance actually near as good as wired performance?

Variance and latency is much higher with wifi than with wired. Sensitivity to interference is also a problem. Dropped packet count is usually much higher with wifi, which can wreak havoc on many applications.

Perhaps things have changed since I last upgraded my home network, but iirc, advertised wifi speeds are half duplex, while wired speeds are full duplex — i.e. wifi speeds are for the combined upload/download while wired can go full speed in both directions.

It's even worse, because wireless is basically a hub. Or phrased differently: there is only one wire available for all devices on a wireless network, and your (as well as others on same frequencies) devices have to figure out which packet is for which device by themselves.

1. Taking a shot in the dark - Maybe check your ethernet cables? Not all ethernet cables are the same. If it's a cheap / old ethernet cable then it might not be rated for the same kinds of speed as your wifi. There is a chart on this page labeled "Ethernet Cable Performance Summary" with some stats for different ethernet cables.

https://www.electronics-notes.com/articles/connectivity/ethe...

2. Yes, although in my experience this is very minor. On my network - good wifi has a 3ms latency and good ethernet has a .3ms latency. While that is a 10x improvement, in the scheme of things I don't notice 3ms of latency. I live in a high rise condo with lots of interference in a major city and can't say it's caused me any issues.

Mostly just no one is bothering to make the consumer products, not that the tech isn't there. Most consumer products are the ubiquitous 1gbps Ethernet ports with 10gbps on certain prosumer/gamer chipsets. Those 10gbps boards are meant to be the upsell so they're expensive for a lot of reasons.

At a consumer level what does that buy you? The average internet connection is far far less bandwidth than that. Most consumers don't have a need to stream large amounts of data. You can just use sneaker-net for bulk transfer with a thumb drive.

> How come today, for home use, WiFi seems to exceed the price/performance of hard-wired?

While you can get impressive figures, a 1gbps wired connection is just so much better than the equivalent speed wifi connection. It’s double or triple the speed in my experience.

Also the wired connection doesn't degrade or die when I turn the microwave on to heat a meal. I really like the Ethernet in my house that I'm soon going to upgrade to 10 GE.

I have limited to zero success gamestreaming on wifi vs ethernet with the particular setup at my house in a townhome.

I have not tested it on my Wifi 6 phone but I'd be tentatively skeptical that you'd want to use ethernet for this use case.

My home is wired with Cat5E and a gigabit switch, and my wired connection delivers consistent, uninterrupted, near-zero-latency 1gbps connectivity 24/7.

I have two different nice wifi APs running all the latest standards and even on new devices supporting the latest and greatest the best I ever see over wifi is about 300-350mbps, plus a couple additional milliseconds of latency. Plus the random little transient wifi hiccups that just happen from time to time.

I can see about a dozen other wifi networks in range of my house, so some radio noise out there but nothing crazy.

And now that I also have symmetric gigabit fiber, only my wired connections can come anywhere close to saturating my WAN bandwidth.

If you have the option/ability, wired is going to crush wireless every time. Wifi is great for casual use and I'm glad it has gotten as good as it has, but for real work/gaming/whatever there's no contest.

I have a similar all Unifi setup, I could get 350Mbps stable connection for any devices as I used 3 Unifi APs. Recently played with the new Eero Wi-Fi 6 AP, I could get 500Mbps as it seems to be duplex two channels. That's probably the best you could get, compared with a cheap gigabit switch can give you 970Mbps easily.

I remember doing an experiment with my roommates with Engenius APs (I can consistently get ~480mbps).

For everything but direct large downloads, everyone thought that using a 100mbps wired connection felt faster/snappier than the 480mpbs wireless

Another couple for your list–

Moving a device from Wifi to wired marginally helps all your other wifi devices. Overall it can improve performance enough to not need another access point.

Even better is PoE which can let you put an AP in places that would otherwise require an electrician to install mains power outlets. Remote power-cycle is good too, even if you already have mains power for the AP.

> How come today, for home use, WiFi seems to exceed the price/performance of hard-wired?

In the review it looks like he’s using a PCIe card for WiFi. If you used a 10Gbe card on a 10Gbe network, I imagine you’d do even better.

Isn’t that how all wifi is connected?

Lots of wifi is via USB.

1. At this point I don't think 10Gbps NICs are worth that much on the second hand market. Most of the cost will be in the wiring. In WiFi most of the cost would be in the router and APs.

2. Wired is far more reliable than wireless in urban settings. You just don't have to deal with overlapping channels, physical barriers, dropped packets, concurrency between devices etc. You might get away with free real estate on 5Ghz+ nowadays but that's going to change as the ISP-provided equipment rotates.

I get a lot of wireless myself because I live in the middle of nowhere, without a single other Wifi AP in detectable range and the inner walls of my house are pretty thin. In that context, with a well configured router, I get extremely stable wireless connections with low latency (<1ms router<->NIC).

But in a city apartment I'd rather have the old 10m Ethernet cable cutting across the living-room.

The biggest problem I run into with 10Gbps ethernet is finding a consumer grade switch that supports it. You are pretty much limited to commercial offerings.

Mikrotik makes "reasonably" priced switches if you want brand new networking equipment for the home (with warranties, support etc.). But yeah it's not stuff you'd find on your ISP-provided whitelabel 15€ all-in-one box.

It's just one of these things. People who know they need 10Gpbs LAN are already completely outside of the scope of regular consumer use.

Switch availability is a big part of it. For my purposes at home, I've got two adjacent computers wired directly together with 10g, then wired over 1g to a wifi access point and router.

I live in house with similar situation, but I still prefer wired over wireless. It's rock solid and pretty stable.

1. WiFi price/performance does not exceed wired connections. Real-world WiFi performance typically falls well below the advertised maximum. You're much more likely to get the advertised maximum with a wired connection. WiFi gets more attention because running cable is hard to do. Most people are content with weak signals at the corners of their homes. WiFi is the good enough solution for most people.

2. If you can use a wired connection, a wired connection will almost always be better. The 5ghz spectrum doesn't suffer from noisy neighbors as much as the 2.4ghz spectrum, but that's mostly because 5ghz is much more easily blocked by walls and its effective range is much shorter.

When I started working from home because of COVID, I bit the bullet and ran ethernet cable in my 100 year old house. No regrets. I had been looking at WiFi mesh networks, but all of the performance reviews I saw on YouTube showed speeds much lower than the theoretical WiFi limits.

This comment is a great summary. For expanded information Ars technica did a great write up of the marketing spiel a few years back [1]. Remains an interesting read and few of the issues raised have been fixed, especially if using common 2.4/5ghz wifi.

[1] https://arstechnica.com/information-technology/2017/03/802-e...

This is another in-depth writeup: https://news.ycombinator.com/item?id=23672564

Mesh wifi is not bad of properly done.. results depend on house shape.. consider it cut in half of wifi theoretical bandwidth after each hop

5Ghz has enough spectrum that there's no need for loss. But whether anyone implements dual 4 x 4 wifi mesh? I dunno. Since smallnetbuilder.com went quiet I haven't had a good trusted place to keep up on wifi products.

Just started wiring up my house with a Ubiquiti Unifi 4x4 mesh and it's been performing as advertised.

The key to good mesh wifi is to have wired mesh routers (at least with google/nest wifi). That still means one or two cat6 runs but better than a switch and as many cat6 runs as you need in every room.

Wired connections will also have stable latency.

That's not a mesh, that's just multiple access points broadcasting the same SSID.

Wired or wireless backhaul doesn't alter mesh capabilities? Multiple access points is merely range extending and is a much poorer solution, with no proper handoff and usually at half speed.

Handoff is handled by the client, not the AP. Mesh is useful when you can’t hardwire multiple APs. Otherwise, you’re much better off with wiring up multiple APs on the same SSID/auth with non-overlapping channels.

Is 802.11r still the latest and greatest to do wired back haul with multiple AP?

> Handoff is handled by the client, not the AP.

For virtually all consumer grade gear, that’s definitely true. That isn’t exactly true if you are using enterprise grade hardware.

> multiple APs on the same SSID/auth with non-overlapping channels

Would you mind saying more? I have a 2 wired APs and 2 mesh APs in my home network. I would prefer to have them all on channel 11 2.4Ghz Wifi (since I have a zwave system as well, which overlaps with the lower part of the wifi spectrum), but I've been wondering if I should have them all broadcast on a single channel, or if they should be staggered - or spread out. Do you have any advice? Or how can I learn more about this? I think my latency is higher than it should be.

Thanks regardless!

That's where you're wrong, and this is the difference between wired meshes and multiple access points.

Wired meshes will handle the handoff (and do it better than devices, with no downtime). Multiple APs will not.

I want wired mesh but never want wireless mesh to avoid inconsistent performance. Sadly there's very few wired mesh APs meanwhile many expensive wireless mesh APs available.

I personally use the multiple AP setup myself. No downtime with handoffs on any of my devices, and I most certainly do not have any sort of “mesh” system. If you can provide some references to back up your claim, I’d love to be educated further. Otherwise my personal experience suggests you are mistaken.

So mesh is still a DS (distribution system) - to use wifi networking jargon

Also remember that in most ways, wireless is more like a hub, than using a switch. Fundamentally all wireless peers share the bandwidth (they can share it differently, but they share) - with a typical ethernet switch (i haven't seen a hub in years, let alone coax...) - you get 1gbps duplex between any two peers simultaneously (these day probably without much problems at 1gbps, depending on hardware at higher speeds and multiple ports - but much closer than with wireless, anyway).

I think the "leap" to 10gbps was a bit too far for consumer grade switches - and "too fast" - faster than your SSD - so it didn't make much sense. Hight cost, low real world benefit.

In a lot of situations, you'll likely only use 10gbps in order for your NAS to talk to multiple clients, each around 1-2gbps duplex.

2.5 GbE is starting to become quite common on midrange motherboards, still uncommon in switches though.

2.5GbE adoption for consumer PC is in progress by Intel i225 controller but sadly the controller is buggy.

Overall most boards seem to sport the RTL8125 controller instead. Which doesn't exactly inspire all that much confidence either to be honest...

https://geizhals.eu/?cat=mbp4_1200&xf=17256_1&asuch=&bpmin=&... https://geizhals.eu/?cat=mbam4&xf=17256_1&asuch=&bpmin=&bpma...

I think it was less "too fast" and more that 10gbase-t is crazy power hungry and significantly reduced run length. End result is basically 0 adoption in the enterprise so there was little to no trickle down to the consumer market.

The power draw also made it extremely difficult to make a 10gbt dongle for laptops and extremely expensive to put on a desktop motherboard.

10 gigabit is slowly becoming more practical, led largely by Mikrotik making a 4 port SFP+ 10gigE switch for $140. I just upgraded my home network to have 10gig on my NAS and my primary desktop workstation, the possibilities it opens up are really amazing and the whole project was about $400 including switch (my existing main switch already had a 10gig uplink but I needed another), 2 NICs and cables. Still not within reach of everybody, but when I priced it out 10gigE ended up being cheaper than 2.5gigE.

What kind of cable do you need to use with that?

Not op, but presumably sfp+ drop cables if the devices have sfp+ interfaces, like:

https://www.amazon.com/10Gtek-Mikrotik-10GBASE-CU-Passive-1-...

https://www.amazon.com/10GBASE-SFP-10G-AOC-Ubiquiti-Supermic...

Would prob. make sense if you got cheap 10g network cards (second hand, maybe) with sfp+ interfaces?

Other option might be a converter/adapter - but then you might be better off with a switch that has rj10 ports built in (and use cat6/7):

https://www.servethehome.com/mikrotik-srj10-review-convert-s...

I used a 30 meter Amazon branded fiber cable for the main run with 10gtek SFP+ modules and 10gtek direct attach copper cables for the short connections. Both are fairly affordable, and the 10 gig fiber SFPs were much cheaper than I expected. Cabling including SFPs was about $100 of the $400.

To be fair, beamforming is a thing now on some consumer wifi 6 routers.

Yes, but it still doesn't magically give you 1gbps full duplex across 10 devices. On the other hand, you probably don't need that.

See my other comment for more pointers on this.

I'm guessing that as we transition to wifi 6 aps and clients - we're likely to get "good" enough networking from wifi for many (most?) applications.

But I still expect eg: backing up 10 desktops in an office to a NAS will be a much better fit for wired network, than wireless - especially if done in parallel with other day-to-day traffic.

You mean OFDMA? Beam forming is already done on cheaper 11ac APs.

I think this changed in wifi6 due to ofdm https://www.networkworld.com/article/3332018/wi-fi-6-with-of...

It now works more like a DOCSIS modem with multiple simultaneous subchannels.

I sort of handwaved this away - even with ofdm there is sharing of a finite spectrum subject to noise/neighbours etc - but it will likely get much better. And radios (in the ap) is still a limitation:

https://meraki.cisco.com/lib/pdf/meraki_whitepaper_wifi6.pdf

> MU-MIMO is technology that allows an AP to service multiple clients simultaneously across a supported number of wireless streams or channels. While this capability existed in 802.11ac, multi-user MIMO will now add communication in the upstream direction. With 8x8 support, which was added during the 802.11n amendment, new APs can now support four simultaneous 2x2 MU-MIMO clients in both the upstream and downstream directions. MU-MIMO will work together with OFDMA to allow multiple clients to communicate simultaneously across multiple frequency ranges as well as multiple spatial streams.

This along with ofdm helping with the simplex problem, will likely mean a wireless ap can come close to a 4 or maybe 8 port switch - but will still be far away from a 24 port switch in terms of full duplex throughput.

This is likely to make a difference if streaming film from a NAS and on to a display, using miracast over infrastructure[1] or similar bw heavy applications - consider for example two displays and two devices doing this via one ap.

For (much) more see: https://www.duckware.com/tech/wifi-in-the-us.html

And maybe: https://www.cisco.com/c/en/us/td/docs/wireless/controller/te...

[1] https://docs.microsoft.com/en-us/surface-hub/miracast-over-i...

Ps I just discovered this was a thing :)

10GbE is just 1.25GB/s, I believe most consumer NVMe SSD (even on OEM) works faster for sequential r/w. I think the most blocker is there's few use case for faster LAN (for normal people).

Well, I’m not going to rewire my house, but I put all of the servers, including the Raspberry Pi, in the same room as the router, so they can all be connected online with cables.

I am certainly glad about 18 months ago I totally renewed the home network to GB powerline and upgraded to Vodaphone business when demon was finally switched off.

And 5G is good but you still need a wired DS and several AP's potentialy one per room.

> Is the Ethernet standard lagging

In many ways, yes. In a lot of ways the ethernet standards of today are basically the same thing that we've had since like 2000. There's no reason it can't use a lot of the same techniques that are used in wifi, other than it increases complexity. 1 Gbps is enough for anyone... right :D

Surprisingly, you'll see a lot of the wifi standards have analogs in cable broadcasting. DOCSIS is very analogous to how WiFi handles data transmission. It is pulling a lot of the same tricks.

> Related, for somebody who wants reliability, is Ethernet/wired still a sane choice, or does that just make them an old geezer?

Yeah, ethernet is still way more reliable than Wifi. You run into far fewer problems with interference or noise.

Some of the higher frequency WiFi standards (60GHz) suffer less from noise (because they can't penetrate walls very well).

> In many ways, yes. In a lot of ways the ethernet standards of today are basically the same thing that we've had since like 2000.

10Gbe is readily and widely available these days, even in consumer space, and the top end is pushing 400Gbps ethernet. For reference, 10Gbe over baset was standardized in 2006, so I'd say we have come pretty far from the turn of millennium.

10GbE in the consumer space is widely available? I can go down to Best Buy and buy an 8-port 10GbE switch? Do many consumer routers ship with 10GbE ports on them?

I guess the latest Mac Pros ship with 10GbE ports on them. Definitely a normal consumer product. I guess its probably pretty low end, as the high end of the consumer space is pushing 400GbE according to you.

Blame intel. they could've landed 10gige in their consumer choosers any time in the last decade.

You can find it's not possible to see 10GBASE-T Hub's massive heat spreader or fan.

What?

To add. Wifi is still shared medium, so as devices increase the share of bandwidth lowers. With proper ethernet switches this really shouldn't be case. As any pair of devices should be able to communicate with each other at full speed.

Commercial applications could use and afford the higher speed 10Gbit over 'fiber' and other expensive interconnects. While the consumer side really did not demand >1Gbit speeds until very recently but even now the demand is still very low.

10Gbit network adapters consumer a lot more power and are a lot more expensive than 1Gbit chipsets. Consumers's primary demands are things like better battery life, price, and smaller form factors so it makes sense to go with 1Gbit. Additionally, a lot of the most popular computers today don't even include a real Ethernet port; it's a folding mechanism which saves space but would make a 10Gbit spec Ethernet port (shielding, etc.) even more of a challenge.

Modern chipsets are starting to include 2.5/5Gbit chipsets which benefit from higher speeds but are still much cheaper than 10Gbit. I think a lot of players in this space were expecting and waiting for 10Gbit to become much cheaper and efficient but that did not happen.

Many Ultrabooks drop the Ethernet port altogether forcing the use of USB-C adapters. USB-C/thunderbolt has a theoretical bandwidth of 20/40Gbit/s which means you, in theory, could buy a 10Gbit ethernet to thunderbolt adapter.

I'm sure there's a lot more on the technical side that has kept the price/energy of 10Gbit high but I don't know the details.

> 10Gbit network adapters [...] are a lot more expensive than 1Gbit chipsets.

You're not kidding! I'm working on updating a lot of my network to 10 GbE, and it's kind of insane how you can get by with super-cheap hardware, cabling, connectors, etc. for 1 Gbps and everything works fine... but when you go to 2.5 or 5 Gbps, prices increase a bit (and the need for better cables with more shielding arises).

When you hit 10 Gbps, there's a lot more expense on every layer, especially the chips/cards that you connect to your computers.

The difference is something like $5-10 for 1 Gbps devices, vs $150+ for 10 Gbps devices (in many cases... there are exceptions).

grab connectx3s from eBay, they're 20-30 quid a pop.

Picked up a handful of old 10g mellanox cards and just direct connected my computers since even the used 10g switches are still expensive... i can reach double my 1g ethernet speeds but not near the 10g max. I'm thinking the pciv1 slots str holding them back...

Nah, 4x gen1 is sufficient, and your cards are surely 8x or 16x. use iperf for a reliable measure.

Stop complaining. My first 100 mbit switch was $4500 for all of the five ports that it had.

I don't think of it as complaining—I remember 'back in the day' when a 20 MB hard drive was thousands too... that's tech.

I was just commenting on the fact that there is often more-than-a-magnitude price difference between 1-10 Gbps.

The switches and cards might be the "easy" part, those are solvable by throwing money at them.. The tough part is the Cat6a or higher cabling. If I understand correctly, properly terminating Cat6a to a spec-conforming level is hard, even professional cable puller don't always get it right. Besides all the tools and testers (which cost more money), you also need skill to ensure your cables are actually performing up to spec. And skill isn't something that you can necessarily buy with money.

WiFi is very much worse than ethernet in dense apartment housing. With gigabit internet service, getting 90% is easy wired, getting 30% wireless is lucky. In a <1000 sq ft apartment my wireless performance depends on the room. Basically on the physical layer wireless is very busy with collisions and retransmitting. Better performance can come with some tuning, but ultimately it is quite limited.

I plug in all the devices I can conveniently.

I wired my PS4 to the CAT6 cable to the router and it has zero issue. Comparing to that, when it was connected to wifi just 10 feet away, I ran into all sort of issues with latency and drop packets when playing FPS games online.

1. 10Gbe copper requires fast switching and more shielding - there was a discussion here a few days ago that it's still very expensive - there will be probably 2.5Gbe ethernet soon that is a compromise on price/complexity and speed.

2. As someone developing part-time with others on a community mesh network if you somehow can use a wire use the wire. no discussion. 802.11ax (wifi 6) does improve quite a lot, but in the end there a quite a few requirements to reach 1.3gbps. noise, distance, mimo-antennas, driver-issues, fresnel-zone should be free, enough free channels to blast 80mhz of wifi data through the air at 1024QAM - all devices must speak wifi 6 - then there is latency due to aggretation and due to the high bitrates the distance is usally very low.

On the one hand you have reliable 110mbyte/s on wire vs. maybe sometimes a little bit more via wifi.

>there will be probably 2.5Gbe ethernet soon

NBase-T is the magic, will train 2.5, 5 or 10g over cat5e, depending on the cabling

> 10Gbe copper requires fast switching and more shielding

That depends greatly on your cable length. A 50ft run (about the distance commonly found from switch to device in an average home) will _easily_ operate trouble free and full speed over Cat5e. It’s only necessary to use Cat6/Cat6A/Cat7 when you are going over long cable lengths (most reputable sources say ~125ft is the max for reliable 10GbE over Cat5e, adding an additional ~50ft for Cat6, and up to ~330ft for 6a/7)

> How come today, for home use, WiFi seems to exceed the price/performance of hard-wired? Is the Ethernet standard lagging, or lack of interest/demand, or some more physical barrier?

"Per-port", 2.5gbps is cheaper than fast wireless. But one fast wireless interface on a router can work in a star configuration with many devices. You can get a 2.5gbase-T card for $30. You can get an 8 port 2.5gbase-T switch with 2 10gbit SFP+ uplinks for $450. This will run circles around any wireless network in practice.

Decent 802.11ax adds more than $30 to your device in most cases. And a router with good 802.11ax is still a couple hundred bucks.

It'd be even cheaper, but 1gbps is good enough for 'most everyone, so we have to deal with repurposing low-end business stuff instead of consumer gear.

You can get 10 gigabit with Ethernet. 10gb switches and especially 10gb NICs from retired servers are getting cheaper.

After you upgrade to a 10gb you might need to use something other than SATA drives to reach the full potential. At least for a NAS. I’d benefit from faster restores and transferring 4K video.

> Related, for somebody who wants reliability, is Ethernet/wired still a sane choice, or does that just make them an old geezer?

Absolutely not. Wired gigabit ethernet is far better than wifi; on a small wired network the packet loss is negligible and contention is non-existent, so the network is smooth and fast. You get better than 80% of theoretical wire speed in file transfers. Remote desktops have minimum latency. Every couple months another wifi router zero-day pops up and some huge fraction of all such routers become vulnerable to remote attacks. It's nice not being involved in that mess.

My fellow remote colleagues with wifi sound terrible on their contended networks; dropping packets and getting cut off calls. The people with wired connections like myself are always glitch free.

Wifi is for toys and casual use. Ethernet is for work.

Just pull fiber and you or the next owner won't have to worry about it in the next 50 years. You can buy pre-terminated or use quick connectors. Most don't realize that fiber is cheaper to produce than copper; the expense comes from the fancy SFP switches and routers.

Wireless is horrible. In urban settings, the RF pollution will lead to all kinds of misery. If you are doing anything important with the connection, it must be wired. It's also bad security because I can drive by in a van and knock out your whole block's internet connections because so many people equate internet with WiFi.

Yeah also fiber are smaller and can help sometimes.. and have no length issues like copper

> How come today, for home use, WiFi seems to exceed the price/performance of hard-wired?

Only if you read the advertised speeds instead of measuring.

Also wired networking has the benefit of not using a shared medium, so all wired clients get full capacity each, vs wireless where all clients share the bandwidth.

Wired is so much better for almost any task I throw at it, by a magnitude at least, so the only units in my house left to wireless are phones, tablets and non-work laptops.

Although slowly, 2.5gbps ethernet is starting to land. I have a few devices in my house now with 2.5gbps interfaces, my hangup has been that I currently have a commercial GbE switch that was moderately expensive and I'm not feeling the need to replace it with a 2.5gbps switch that will be quite expensive (and if I really wanted to I could get full speed on some ports with SFP).

But not feeling the need, I think, kind of makes the point... WiFi has to continuously push to higher and higher speeds in large part because for various fundamental reasons WiFi is not able to deliver speed with consistency. Packet loss, jitter, contention, and periodic serious disruptions in connectivity are the norm for WiFi, and very high speeds offer more overhead to tolerate these problems. Unlike Ethernet, WiFi is also fundamentally half-duplex and contentious between multiple devices, so very high speeds and MIMO techniques (as well as beam steering and other more cutting-edge technology) are used to mitigate this. All of these enhancements are aimed towards allowing WiFi to consistently and reliably achieve 100mbps and more in practice - something which even the latest WiFi standards still struggle with in much of the "real world."

By contrast, GbE can consistently deliver gigabit performance, full duplex, non-collision with only slight overhead, and a very high degree of consistency.

Gigabit ethernet already allows saturation of most storage devices in consumer use, so from a practical perspective, there aren't many reasons anyone would want more--going to 2.5, 5, or 10GbE in a home network is not likely to allow you to actually do anything faster because at that volume of data you're probably reading from or writing to storage, and consumer storage devices cannot reach even gigabit speeds. Likewise there are few or no consumer devices capable of generating instantaneous/real-time data volumes over GbE, the ones that are are moving uncompressed video, e.g. HDMI at 10-20gbps, which is both beyond what "faster" ethernet specs currently readily available can do, and it's just something that few people seem to want - there aren't many situations where you're moving video over a distance longer than an HDMI cable and you aren't willing to compress it.

I actually have an unusually demanding situation as I have a project that involves moving multiple high-rate IQ streams from large-bandwidth SDRs over the network. Even here, CPU and I/O issues on the sending devices are consistently a bigger problem for me than the actual network. There are SDRs with highly optimized network support that allow them to achieve real-time rates in the Gbps range, but they cost so much that owners of such devices aren't going to have a problem investing in commercial network equipment.

It feels a bit lame to say that "GbE is fast enough," but for consumer cases I think this is largely true. The only thing I can see changing it is moving full-rate video around, but video streaming has already been captured for consumers by vendor-specific systems that use compression, and "works fine" within an ecosystem even over WiFi.

I mostly agree. On the other hand, the existence of hdmi/thunderbolt switches/chaining kinda proves there's life for networking at 10gbps and beyond - 4k@100hz stereoscopic video etc.

But for better or worse, we didn't get ethernet for that.

> there aren't many reasons anyone would want more--going to 2.5, 5, or 10GbE in a home network is not likely to allow you to actually do anything faster because at that volume of data you're probably reading from or writing to storage, and consumer storage devices cannot reach even gigabit speeds.

a sata ssd is a bit under 6Gbit. new fast pcie v4 ssd's are a bit shy of 6Gbyte (48Gbit), & your budget models are still frequently 16Gbit & up.

it really makes me sad how widely accepted it is that consumers don't need good connectivity. meanwhile usb4 is arriving & allows direct computer to computer links over regular usb-c cables at 40Gbit. hopefully there's growing discontent at this old old old 1Gbit, & also meh 2.5Gbit ethernet, as ways to shuffle data around.

> hopefully there's growing discontent at this old old old 1Gbit, & also meh 2.5Gbit ethernet

Nobody is stopping you from going all in with 40GbE. It can be done over twisted pair copper even if you are so inclined, but I prefer QSFP+ modules with fixed twinax cables personally.

it's priced at enterprise prices. my actual hope/expectation is that eventually a lower tier enterprise switches emerge with 25gbit ports, & that becomes low priced enough that producer people start thinking of adopting. there's no point to relying on 10gbit, the total switching aggregate is tiny but prices for the switches are sky high.

I’ve picked up used gear on eBay and also from office liquidators and had good success going that route. Yes, you don’t get warranty coverage then, but I typically buy a backup just in case. Yes, you will pay 5x the price per port even used, but you’re also getting nearly 5x the performance too.

I have some infiniband gear I play with. second hand markets are great.

but I think there's a corrupt systematic drive against innovation & progress that I've tried to advocate against & speak to in this thread. I think the standards have failed to rise because they have been actively suppressed & kept low.

intel embedding dual 80gbps transceivers on their mobile chips (in displayport output) is a tell that this really is just stuckness, intransigency. I can acknowledge that intel is working at much shorter ranges, but all in all, I think host-bus-adapters are way way way overpriced & maintain price simply by the vast quantity of radically inferior products they feed us in the consumer market.

> radically inferior products they feed us in the consumer market

I think you need to try to a bit less hyperbolic and stop and apply logic. Out of your own social graph, how many besides yourself would be willing to pay 10-100x the price for consumer home networking than they do now? The reality is, outside a few folks like you and me in the 0.001% of consumers, there just isn’t even a want for 10+ GbE, let alone a need. Heck, probably >50% of consumers just use whatever craptastic all-in-one device their ISP gave them for their home networking. The market is just not worth it, it’s still too niche. I also don’t feel this is a chicken or egg situation either. The applicability of 40GbE networking will likely never be useful for the average consumer unless there’s a radical new technology that hits the market.

As such, if you want the gear today, stop considering yourself an average consumer and instead buy gear targeted to the audience who commonly has needs similar to yours.

> WiFi seems to exceed the price/performance of hard-wired

WiFi absolutely does not. This is a test in lab conditions, with a single, stationary device and no stats on packet loss. Unless you're happy to use your device in these conditions (but at this point why not use a wire?), this does not represent the real-world conditions in which you'd be using the device.

If he tries again with people moving around and moving the devices around, with more people using the wireless network at the same time, he'd see much more different results especially when it comes to stability.

Stability is the main problem with Wi-Fi. It's relatively easy to get a huge speed test, it's harder to maintain a stable latency and no packet loss. It doesn't help that operating systems aren't good at reporting this information to the user, so the Wi-Fi always appears to work (with decent speeds) until it totally craps out when you launch an interactive application (like calling or multiplayer gaming) where low latency is paramount and the error correction/retransmission algorithms can't do their thing fast enough).

Single all in one modem/router/access points are popular because they are incredibly easy go set up and that is about it.

The only people who don't run Ethernet cable are those who are too lazy to bother.

The second any sort of performance becomes important you will need to find dedicated devices with Ethernet running inbetween.

My parent's house is 100% wireless however I had to run extended lengths of Cat 5e to set up multiple wireless access points. We have multiple 4k TVs on different sides of the house. There was no way we could keep decent speeds and wireless coverage without running cable.

Even for my Android phone. I will occaisonally use a wired Ethernet connection to speed up downloads.

Q: 1a. How come today, for home use, WiFi seems to exceed the price/performance of hard-wired? Is the Ethernet standard lagging, or lack of interest/demand, or some more physical barrier?

A: WiFi does not exceed the price/performance of hard-wired Ethernet. They are different enough it is difficult to compare properly - for instance, for wired Ethernet, do you include the cost of the wire installation, termination, etc? I don't think it is reasonable to include - complex wiring gets more expensive - potentially by an order of magnitude! Fiber can be much more expensive than copper.. or cheaper. Fiber transcievers get very expensive for very long distances (miles) etc.

So to do a simple comparison - 10GbE vs. Wifi. Let assume you are adding these to a computer which doesn't have them. The Wifi adapter is $26.90 A 2.5GbE adapter is $30

What about the network switch? (or wifi access point) Typically with wired ethernet, you do this as a "cost per port" - currently about $25/port on the low end (rated at 10Gbps, by the way, but able to step down to 2.5Gbps).

This (top performing "high end") Wifi router runs $250. It may officially support 200+ clients but that just means it can theoretically talk to them at all. Good luck pushing actual data to/from them! In practice, to match up in cost/performance with a 10 port ethernet switch, it would need to drive the full 2.5Gbps for 10 clients simultaneously! I guarantee you it can't.

And the price equation only gets better if you go full 10Gbps - which is probably the "sweet spot" right now for price/performance, buying in SOHO-sized quantities.

Q1b: The ethernet standard is pretty excellent, and currently goes to 400Gbps, and if you have the cash to buy 400Gbps equipment - you can basically expect it to work to spec - if the switch fabric says it will support X Tbps concurrently - it will. Good luck with that on Wifi.

There is a TON of demand - you just don't see it in your house. Or your neighbor's house. Or your friend's house.

You should though! Ethernet is AWESOME - and back when we had wired landlines, people ran those wires all over their house. It isn't that hard, or that expensive, and you'll get MUCH better, more predictable performance from doing it, and moving as many devices as you can to wired ethernet. Even your WiFi access point benefits from an ethernet cable running to a location where it's signal works best, rather than some closet off in the corner of your house.

Q2: Related, for somebody who wants reliability, is Ethernet/wired still a sane choice, or does that just make them an old geezer? In urban setting with overlapping wifi cards all blasting full strength at their neighbours, is real-life wifi performance actually near as good as wired performance?

A: No, real life wifi is nowhere near as good as ethernet. Switch to ethernet as much as you can! Of course, it doesn't make much sense to run an ethernet cable to the tablet you read sitting on your couch. But every desk, printer(?), your WAP(s), etc.

Latency and reliability on WiFi, even the best, pales in comparison to a simple UTP cable. Besides that, you're always just one step away from some security issue in the WiFi stack that might leave you wide open to abuse from nearby stations.

The rated speed of your access point is essentially shared with all clients in that channel. So if you've got 16 devices, they all share that few gigabits of potential airtime assuming they have all nearly perfect connections to the AP. My 16-port gigabit switch can handle 32Gbit of switching, significantly more throughput than the latest WiFi can even hope to achieve in ideal circumstances. Adding more capacity is buying another switch. Depending on the existing density and RF noise of the area, you might not be able to add much more density even with adding more APs.

I think its more that we're seeing wireless get somewhat competitive with its wired counterparts in the home, not necessarily that its surpassing it. Its still not as reliable as wired. Unless you live in a anechoic chamber you're going to get dropped frames from time to time, requiring retransmits and increasing your jitter. If you're experiencing that on your wired networking, you either messed up your cable runs or you need to step outside of your microwave oven.

That said, 10GbE+ outside of the datacenter has been a long time coming. Its there, but its not very widespread. Most devices tend to use a good bit of power and are rather expensive. On top of that, unless you're doing something datacenter-like (using a SAN or building HPC cluster) you usually don't experience much of a difference. In the few instances where you might want more throughput, you can often set up your devices in a LAGG to combine multiple gigabit ports to increase aggregate throughput.

> How come today, for home use, WiFi seems to exceed the price/performance of hard-wired? Is the Ethernet standard lagging, or lack of interest/demand, or some more physical barrier?

10-gige is already cheap and 40-gige is coming down in price. There is just no need for it at home. I have my house wired with 10gige, and > gigabit fiber internet, and literally can’t notice when switching to wireless. 500 mbps should be enough for anyone.

Markets don't innovate unless pushed into competition & ethernet has had so little pressure to innovate. I think there's a lot of parallels here to another article today, on webcams being ultra-uncompetitive. Discussion also talked about monitors in the same way. There's a very low expectations for these products & few people trying for better.

There is 2.5Gbit and 5Gbit ethernet happening. These are both significantly easier to implement than 10Gbit, which everyone is withholding, reserving for "enterprise" ("who needs that much speed?") but the whole thing screams scam, when far far more complicated wifi chips are so cheap, when 40 & 100Gbit are not that expensive. 10Gbit is not easy but all these 5Gbit, 2.5Gbit options; it strikes me as more about tiering, about having a range of products to sell, rather than update to something modern. Everyone is upgrading at minimum possible speed while leaving pricier options free of competition.

Meanwhile, intel just shipped two 40Gbps USB4 ports on their new cpu chip. Which can directly connect two computers with just a regular cable. Unlike ethernet this is very short range though, measured in meters. Although rumblings about optical keep showing up, which would throw everything into chaos. Wifi though, now, is multi-gigabit. It's making ethernet look bad.

>WiFi seems to exceed the price/performance of hard-wired?

Those are best case performance. i.e Standing next to the Router. In most cases you wont be getting 1Gbps from WiFi.

But having said that, WiFi is improving at the rate simply because there is a much larger market to drive its volume, and amortised R&D. WiFi is included in every laptop, tablet, Smartphones and many other applications. That is nearly 2 Billion devices a year. Compared to Ethernet, you are looking at Desktop PC, Servers and other much smaller niche like NAS. Not to mention faster WiFI ( aka WiFi 6 ) is an easier sell to consumers than faster Ethernet which is really only aiming at prosumers.

First draft of WiFi 7 will arrive in a few months time.

>is Ethernet/wired still a sane choice

Always use wired if you have a choice.

Hey Jeff -- Just wanted to say I really appreciate you taking the time to write stuff like this up. I know that may be 'the point' of what you're doing, but I also know from personal experience that writing up a comprehensive review/guide/benchmark (whatever you want to call this style of writing) can easily double to quadruple the time investment for a project.

Thank you for sharing the troubles and pitfalls for the rest of us to learn from, or even just reference.

I'd say doing the post + video at least 10x's the time involved in one of these projects.

A lot of people can get to the 'I did this thing and I got it to work' stage. Getting to 'I did this thing and I can get it to work from scratch a 2nd time' is better. Getting to 'I did this thing and here's my documentation, enough to help you do it too' is my goal, for pretty much anything I work on.

I wasn't even talking about the video, just the write-up.

I haven't done much, but video production seems like a black hole of space and time where things are never 'done' and eventually you just release it because it's the only way to stay sane.

> Getting to 'I did this thing and here's my documentation, enough to help you do it too' is my goal, for pretty much anything I work on.

As they say in the biz, 'You're doing The Lords Work'. Again, thanks, and much appreciated.

Eh, clickbait title. Suggests speeds of built-in raspberry WiFi which is not the case.

The implication is more on the reader. I don't expect the title to be as verbose as "WiFi 6 gets 1.34 Gbps on the Raspberry Pi CM4 with pcie wifi adapter."

It's wordy enough as is. And anyone who is in the market for this news knows that you aren't getting that with the stock radio.

> anyone who is in the market for this news knows that you aren't getting that with the stock radio.

Well, yes, but the suggestion that it is possible from this clickbaity title is why I'm here to begin with.

If we're concerned about wordiness, we could have dropped "Rasberry Pi CM4" from the title as this detail seems irrelevant when the external wifi adapter is doing the heavy lifting.

That's the main point of this, though—until the CM4, it was not possible to get this kind of card working easily on any kind of Raspberry Pi.

There are some other SBCs (usually more expensive) with the capability, but being able to (easily) put AX WiFi into a Pi project enables some new use cases.

Why not? This looks like a much better title to me.

Of course, maybe the article wouldn't have been read as much then...

This also reminds me of the many people who complain about my YouTube thumbnails from time to time.

I do have the 'open mouth' from time to time (I hate it but do it)—and the reason is not because I think that's appealing, professional, or amazing.

It's because through a lot of A/B testing over the past year, the same quality and subject matter in a given video without a dumb thumbnail gets at least 30% less impressions on YouTube.

What use is making great content if your marketing around it can't get people to read/view it?

You have to hold your nose to do any marketing (IMO), and there's a reason most of us on HN distrust salespeople and marketing people. But they exist for a purpose. I've had to dip my toes in those murky waters to be able to sustain the open source work I'm doing.

'Be the change you want to see in the world'?

Or just 'decide if you want dumb clicks or reasoned interested'.

I hate what you describe about YouTube. To me the worst offender is a chap who used to go by 'Stephen the Robot', but I think he changed the channel name, his actual content is excellent, but I have to sit holding the remote, so that I can skip all his repulsive self-reactions and crash-zooms on his face. It's such a shame, he's working on an awesome project, it's really interesting, (open source scratch built pick and place machine) but the delivery seems to be designed for 'tweenagers' on TikTok. (For context, I'm in my 20s, hardly old and grouchy. I actually might not be older than him even, unsure.)

The key is to get the dumb clicks (to pay the bills), while not going so far as to disenfranchise the core audience.

Agreed. It took me a while to realize that this was just about compiling Linux drivers for a network card connected through PCIe.

Raspberry and Linux Drivers issues aside.

This is 1.34Gbps on a 2x2 5Ghz 160Mhz WiFi 6 with theoretical speed of 2.4Gbps. I could get very close to 800Mbps on my WiFi 6 on iPhone. So I was expecting 1.6Gbps with 160Mhz.

With Wireless speed over taking wired, we really need 2.5 /5Gbps Ethernet, but the move to these standards are so slow.

What? Cat 6a is 10Gbps so the actual speed is probably more than 5Gbps.

I think the parent post is referring to the low rate of adoption of >1gbps networking. Switching and routing is expensive and not many devices support it.

They’re presumably referring to 2.5GBASE-T and 5GBASE-T. Some motherboards and WiFi 6 WAPs are now including NICs meeting these standards but not 10GBASE-T. Still not commonplace throughout, though. Both 2.5 & 5GBASE-T are rated for 100 m on CAT 5e.

https://en.wikipedia.org/wiki/2.5GBASE-T_and_5GBASE-T

Yes the point of NBASE-T or 2.5/5BASE-T is that it should be a lot cheaper than 10GBASE-T.

10G Ethernet has gotten a lot cheaper, certainly viable for enthusiasts. I'm not sure what I would use it for except for faster access to my NAS, and that's coincidentally also the one device category where 10G ports are somewhat common. For everything else 1G per device seems plenty, and with Ethernet you actually get the advertised speeds in typical scenarios.

A network card for 10Gpbs costs at least €100 per device here, whereas 1Gbps goes for less than €10. A switch with at least 4x 10Gbps ports costs over €300 (whereas 4x 1Gpbs costs about €20).

Using 2.5Gbps ports is more cost effective, with switches available at a third of the price of a 10Gbps version and network cards going for about half the price of a 10Gbps port.

A literal 10x price increase is not exactly what I'd call "viable for enthusiasts". With 802.11ax, WiFi has surpassed ethernet for all practical purposes for end users. You can't _rely_ on ax working at full speed, but if you try it and it works, it's a whole lot cheaper and easier than setting up wired internet.

It's sad, but this is only the case because network device vendors want to keep cashing in on "enterprise" hardware with ludicrous pricing. Prices have barely dropped over the last 4-5 years on many 10Gbps switches, probably because they're mostly used in data centres anyway.

I've run into annoyances with 1Gbps ethernet a few times, usually when transferring large files between PC and laptop or from my NAS. After 10 years of universal gigabit ethernet, I somehow felt like it shouldn't be faster to copy everything from my laptop to an external SSD and then copy it back to my desktop than sending the data over a short wire connected to both computers, but here we are.

> A literal 10x price increase is not exactly what I'd call "viable for enthusiasts".

As an "enthusiast" here that's because you should be looking at used gear. You can pick up things like the Mellanox Connectx-2 SFP+ used cards for cheap, like $20-30 USD (local availability is of course a crapshoot though).

If you want/need 10Gbps over cat6, though, then yeah it gets more expensive. But if you're just looking for 10gbps between a couple of systems and can run a cable, SFP+ with some DAC cables are very much within reach.

> With 802.11ax, WiFi has surpassed ethernet for all practical purposes for end users. You can't _rely_ on ax working at full speed, but if you try it and it works, it's a whole lot cheaper and easier than setting up wired internet.

For all practical purposes it hasn't. Remember the airspace itself is shared. That old 802.11ac laptop or phone or whatever streaming youtube or netflix will murder your theoretical 802.11ax bandwidth. It's not unlike claiming that an 8-port gigabit switch is "8Gbps" (assuming a full-throughput switch, which is common-enough) - that's basically how wifi marketing speeds works. It's the shared available bandwidth, not the per-client bandwidth. If you actually care about bandwidth then 802.11ax is still going to be practically inferior to 1gbps ethernet runs. But for most people as long as the wifi is mostly faster than their internet connection, which is typically far less than 1gbps anyway, then who cares? More significantly latency with wifi still remains unreliable, so anyone trying to game will still also be better served with wired connections. Even if "theoretical peak bandwidth" is lower.

> I'm not sure what I would use it for except for faster access to my NAS

Uncompressed 4:4:4 4k video at 60hz is only a small bit beyond the reach of 10G ethernet. Imagine being able to remote display from anywhere to anywhere else in the building with absolutely zero loss. Use a huge server as your desktop machine without having to listen to those fans all day...

meanwhile most DP nowadays use compression in the background and you cant even turn it off

Just imagine how much you could get with a wire.

Getting a wired connection above 1 Gbps seems non-trivial -- there are products that do 2.5 Gbps, and other products that do 10 Gbps. It seems like you'd have to pick one or the other and hope the rest of the world standardizes on the same speed as you.

10GbE is a standard. It's expensive for home users, but it's very much the standard for faster-than-gigabit ethernet.

https://en.wikipedia.org/wiki/10_Gigabit_Ethernet

> there are products that do 2.5 Gbps, and other products that do 10 Gbps

And 5, 25, 40, 50, 100, 200, 400 Gbps.

> seems like you'd have to pick one or the other and hope the rest of the world standardizes on the same speed as you

Huh? A connection between A & B will just be as fast as the slowest link between them, be it 400 Gbps or 6 Mbps, regardless of the various other speeds. And a 10GBASE-T NIC will happily autonegotiate with a 10BASE-T NIC and everything in-between.

Not that I imagine it causes problems in practice, but as a point of order I think 10G copper PHYs cannot negotiate below 100BASE-TX.

Maybe, not sure. I can’t immediately find anything that would specify as such, but I suppose adapters could opt to not support some modes (and I guess technically I may have been overreaching a bit on that bit anyway since autonegotiation is only mandatory at all in 1000BASE-T and later).

some 10G-BASE-T cards don't support the 2.5G and 5G modes as those got standardized later.

Some even don't support 100 Mbit. But most do. :D

I imagine around 2.5Gbps as that seems to be the speeds the compute module is getting using a SATA RAID. If we want to be optimistic then it would be 5Gbps as that's the limit of PCIe on the Pi.

Real-world usage it seems like somewhere around 3.5-4 Gbps is the limit on the PCIe x1 lane on the Pi—I'm still doing more testing around that.

I'm actually testing a 2.5 Gbps card with the Realtek RTL8125B chip in it now: https://github.com/geerlingguy/raspberry-pi-pcie-devices/iss...

Optical maybe, but copper has more limitations. Signal propogation in air is faster, and with things like beam-forming there's big advantages to having multiple sending and receiving points spread apart.

So to push speeds in a similar way with a wire you might end up needing a mesh of wires.

Power delivery and (relative) protection against signal degredation are, for me at least, the remaining big reasons to wire things in. Speed not so much.

I love how detailed the video is, awesome!

I recently wanted to cross compile something rather obscure for my armv7 based e-book reader. When I first researched the device I found out that the manufacturer of its chip no longer exists. I ultimately gave up after trying many different things such as compiling different versions of gcc, setting env variables and using the configure script but I just couldn't get it done. Compiling on the device would have been even harder because it only has a statically linked busybox and some proprietary stuff on it. As someone who is younger than the internet, it can be really hard to try something without being able to google it because no one has even tried that yet. But after seeing this video and the fact that you ultimately got this to work, I feel motivated to try this again tomorrow!

I think the mentioned Mikrotik switch should work with 2.5Gbps port (if you run the router in AP mode)? I've run successfully the mentioned Mikrotik switch against 2.5Gbps NIC and 5Gbps NIC to reach the advertised speed cross computers.

Yep; it was a matter of an older transceiver I was using. I am using a different one and it supports 2.5 GbE now. Sadly, I can't edit video on YouTube like I can edit or update a post :)

And... is there a raspberry pi you can add storage to that has enough bandwidth to write those 1.34 Gbps?

I know they have to keep costs down and all, but they could try to put out a version with a m.2 slot. Even sata would be better than these sd cards.

USB 3.0 offers 5GBit of performance, so yes

Something that doesn't use 100% cpu? :)

And yet my RPi4 can't get more than 9.1MB/s over ssh when wired to cat5e, or 17.14MB/s over netcat. I know the connection is good as testing with my laptop I can get gigabit speeds, but the pi insists on being sluggish.

That's not the networking performance though. RP4 can hit line speed on Gb Ethernet: pi@kodi:~ $ cat /proc/cpuinfo | grep Rasp Model : Raspberry Pi 4 Model B Rev 1.1 pi@kodi:~ $ uname -a Linux kodi 5.4.72-v7l+ #1356 SMP Thu Oct 22 13:57:51 BST 2020 armv7l GNU/Linux pi@kodi:~ $ iperf -c 192.168.0.39 ------------------------------------------------------------ Client connecting to 192.168.0.39, TCP port 5001 TCP window size: 214 KByte (default) ------------------------------------------------------------ [ 3] local 192.168.0.214 port 44960 connected with 192.168.0.39 port 5001 [ ID] Interval Transfer Bandwidth [ 3] 0.0-10.0 sec 1.04 GBytes 891 Mbits/sec

Put 4 spaces in front of every line. Btw 891 mbps is not line speed

I just setup my rpi4 as a gigabit router with a USB Ethernet for the wan port. I can hit 980mbps and negligible latency. It’s got slightly more spread on small packet sizes compared to a 300$ gaming wifi router. Where it really sucks is VPN and cryptography since there is no hardware acceleration. It sits mostly idle just doing nat and ipv6 forwarding (and pihole)

I’m going to document it once i work out the kinks. Ipv6 was a real learning experience and had to workaround issues with lack of good native ipv6 support in the tooling and configuration.

> Where it really sucks is VPN and cryptography since there is no hardware acceleration.

I know that OpenVPN is super slow on the Pi, but have you tried WireGuard?

No. At this point I trust my ISP more than any VPN provider.

You can be your own VPN Provider, all you need is a VPS.

RaspOS? Had issues with it being oddly slow. Threw Ubuntu or armbian don’t recall which and it went away

A friendly warning, don't upgrade to WiFi 6 just yet, wait for WiFi 6E to future proof yourself. That adds 6Ghz channels, which obviously would contain next to no contention this early in the deployment.

> Apologies for the thumbnail—it performs a thousand times better than not-open-mouthed-Jeff in my A/B testing.

Really, 1000x better? That's insane if it's true (and explains those irritating thumbnails on YT).

The CM4 and the Pi4 hack expose one PCIe lane, correct? Is it possible to use some sort of... port multiplier / switch / hub to run more than one PCIe device on a Pi?

The RPI4 has a 1x lane of PCIe 2.0. Maybe you could split it into two 1x PCIe 1.0?

You can split it to two 1x PCIe 2.0, or perhaps even 2x, 4x, probably even 3.0, 4.0, with the right chip, but in any case you need a PCIe switch, and they will share the uplink bandwidth.

There's at least one reported case of a PCIe multiplier card working with the Pi4: http://labs.domipheus.com/blog/raspberry-pi-4-pci-express-it...

Yes; I'm testing two switches currently—see https://pipci.jeffgeerling.com/#pcie-switches-and-adapters

I've been farting around with Friendly's devices. Several of them have 2x PCIe, which I'm using/will be using for persistant storage, but I don't think you could compare anything they have to the compute module, especially with respect to PCIe.

Is there any flavor of hardware that lets me use PCIe as layer 1 for a TCP/IP stack? Specifically, with a switch in the middle instead of daisy chaining?

pcie switching is a thing, but it's wasted on IP.

Are you sure? Isn't the point of NVME that it has lower latency than speaking SATA to flash memory?

I've seen switches for putting a whole bunch of devices onto the same motherboard, but can you have a whole bunch of motherboards talk to each other directly?

http://www.dolphinics.com/products/MXS824.html for example.

My point was that you don't buy those and then layer TCP/IP on top of it.

thanks for respecting system theme on your website.

Of course! The day the dark mode feature came out on Safari I added a darkmode theme. I also set background to 000 black, so it consumes less energy on OLED displays.

Reply on news.ycombinator.com