Intel Executive Posts Thunderbolt 5 Photo Then Deletes It: 80 Gbps and Pam-3 | Hacker News Reader

Intel Executive Posts Thunderbolt 5 Photo Then Deletes It: 80 Gbps and Pam-3

anandtech.com

58 pointsvanburen60 comments

My experience with thunderbolt hubs has been frustrating. These ports and hubs usually advertise Gbps which I then convert to display usage + leftovers for usb data. Thing is, these hubs often have hidden restrictions. For example my thunderbolt 3 hub has support for two displays, but each one can only do 1440p/60hz despite the hdmi 2.0 spec and advertised port throughout being totally capable of pushing 120hz.

Also I had to get a usb extension for my wireless mouse dongle due to interference, not sure how they’d fix something like that though.

> my wireless mouse dongle

Are there laptops with Thunderbolt but not Bluetooth?

Bluetooth mice are often laggy and have a low polling rate

From what I heard somewhere you can achieve less latency using a USB dongle than Bluetooth. Bluetooth also can interfere with WiFi. And that mouse could just not support Bluetooth.

> Bluetooth also can interfere with WiFi.

Most wireless mouse/keyboards dongles use ESB (enhanced shockburst), which operates on the same frequency as WiFi and Bluetooth (2.4GhZ). So it may also interfere with wifi.

> Are there laptops with Thunderbolt but not Bluetooth?

There probably are desktops with Thunderbolt but not Bluetooth and a wireless keyboard+mouse set is even more relevant to be used with desktop PC than it is with a laptop.

In fact I always felt like probably buying a BlueTooth keyboard+mouse for my laptop so I could spare the USB ports for other usage. Never actually bought any because insufficient motivation - wired stuff actually feels enough, and I just bought a good USB hub for more ports.

For already about a couple of years I'm using BlueTooth earplugs. This is my first actual experience with BlueTooth. And it is far from perfect - the connection gets dropped every now and then and some times there are serious sound distortions. I've tried more than one headphones brand (never tried really expensive ones though) and many host devices.

This makes me very unsure if I actually want anything else BlueTooth.

Bluetooth keyboard / mouse does not work at boot time. They work just when the hdd /ssd is open and and the os has booted.

There's nothing special about Bluetooth - boot loaders can implement the protocol like any other software can.

Which boot loaders allow straight bluetooth keyboards to enter the bios configuration?

How's Thunderbolt going with AMD anyway? Got a 2nd hand Apple Thunderbolt Display in mint condition and really, really cheap, and used it as an excuse to upgrade my 'old' PC. Wanted a sff PC but couldn't find in the area any AMD that offered Thunderbolt.

I think it's still 1-2 years away. Rembrandt and X670 may have USB4.

You can buy the ASUS ProArt which has native thunderbolt 4 support. It's a B550 board, but expect some x570's

There are a few AMD motherboards with Thunderbolt, have been since 500 series launch. AnandTech does X570 / B550 overviews that list them.

I wonder if one day Thunderbolt+DisplayPort+USB hybrid (what some Type-C ports of today already are) will become the only connection interface we will use. It can replace HDMI (with a small and cheap adapter for HDMI-only displays), PCIex slots, OCuLink (not really an expert on these but why not?) and obviously SATA which already falls out of relevancy quickly. So why do we even need anything but full-featured Thunderbolt ports?

We already have one serial signalling method being "pan-standard." It's single mode optics transmission.

Optic transmitters already don't care what data exactly, and how is sent through them. The electrical interface in between them, and the main chip is a dumb SERDES, and it does exactly nothing except sending the data as-is.

Once on chip, the signal is free to be routed without any special electrical modulation as one wants.

Plain SFP+ transmitted are used from Ethernet, to FibreChannel, to Infiniband, to SONET, to 4G CPRI, to HDBaseT. Both fibre, and optics actually got really, really cheap.

Some ISPs even transmit DVB-T2 over the same optical cables thy use use to provide the actual Internet connection.

The only drawback is that Thunderbolt is fairly expensive and the cables are very expensive.

Thunderbolt cables are just USB-C cables. Some cheap cables may not have all the connections needed but ff you are getting a cable rated for USB4, it will work for Thunderbolt.

Not really. USB-C 3.0 cables are $10-20 while Thunderbolt/USB4 cables are $30.

A SATA cable is $2-3 so replacing SATA with Thunderbolt would increase cable costs by literally 10x.

I just wonder what the actual production price is. Perhaps Thunderbolt cables only cost this much because of being a luxury segment - nobody but rich people buying expensive peripherals for their expensive computers really need them today so why not charge them another $30. Maybe the price will drop substantially once most of the consumers will need them.

> Some cheap cables may not have all the connections needed

Well yeah, that's pretty important. A 2m cable that supports 40gbps from a reputable vendor is $59 (https://www.amazon.com/dp/B084Z65YJQ/), and that's one of the better prices I've seen. There's a bunch of no-name ones for around $50, but I'm not sure I'd trust them.

>getting a cable rated for USB4, it will work for Thunderbolt.

While USB4 does mandate the use USB-C connector, support of Thunderbolt is not mandatory and is entirely optional.

> and the cables are very expensive.

Define expensive to you? It’s cheap enough to me that I buy Thunderbolt cables exclusively as they have the most universal support for all the USB-C alt modes and PD variations.

Even if you can get them cheaply the length limits on non-“active” thunderbolt 4 cables make them a poor choice as a universal replacement for a usb c cable in many cases. Unless you have active cabling (hundreds of dollars or more typically for thunderbolt 4 and extremely rare) the Thunderbolt 4 spec maxes out at 2m for passive cables. You can have other flavours of USB C cable in longer lengths and still conform to spec.

I’m in no rush to replace my laptop charging cable with a 2m one. Similarly, I use a ~5m usb c cable for a VR headset.

> I’m in no rush to replace my laptop charging cable with a 2m one

That’s the beauty of having choice. I have zero need or desire for a longer than 2m cable for my laptop.

IIRC these things depend on the port and the cable supporting it. If so, we'll have a universal form factor for all sorts of functions that depend on the capabilities of the port and the cable both being clearly labeled which, in practice, are requirements not always satisfied (or understood by most users). For the most part in the past there was one cable that did one thing, fit one specific port, and basically did what it was expected to do without any fuss.

I expect more confusion and frustration from non-power-users.

I believe most of the people writing here (let alone those using OCuLink) can afford always buying the best full-featured cables to avoid the problem entirely.

For those who can't afford this the whole problem can be solved by better labeling of the cables.

Whatever the case, to me this doesn't seem a reason why we should keep different ports for everything on actual computer mainboards when there is thunderbolt.

If a mainboard manufacturer is worried about end-users being unable to get proper cables - just put some in the box.

im not necessarily opposed to it, but there are tradeoffs that come with a universal port. USB consortium tried to fix this with the coloroed tongues inside the female piece, but either manufacturers thought they were ugly or the number of different functions outstripped the colors your could reasonably use.

i doubt think many on HN will have trouble with this, except when their family asks why it doesnt work when the cable is plugged in and cant read the fine print on the sheath.

Dual USB-C[1] can also replace NEMA-1 power connectors.

[1] https://m.xkcd.com/2493

It's one of those ideas that sounds really good, but winds up being a bit of a nightmare. Just looking at what we currently have with USB-C things get very complicated. A USB-C port can support one of several different speeds of USB or Thunderbolt. It may also support native Displayport alternate mode or it may not. That's just the ports. The cables become even more tricky. Even if you know your USB-C/Thunderbolt port is capable of Displayport, is your cable capable of it? Most USB-C cables out there are USB 2.0, because it's way cheaper and many users only care about charging their phones or the occasional file copying. Even if you do have the right calbe it may only support lower bandwidth so now you're stuck trying to figure out why you can't do 4K60P. Now how do I explain all of that to a less technical user?

> Now how do I explain all of that to a less technical user?

"You need that expensive cable labelled as USB4 to do anything other than charging your phone".

I guess it is pretty difficult and expensive to make on both sides from a notebook or on front and back on a pc very fast ports.

When compared to internet speeds, the amount of data you can send over local transfer is amazing. I really hope this pans out.

At the same time my office Internet speed has already exceeded that of SATA.

Office internet speed, or office LAN speed?

My home internet speed exceeds SATA.

You can get more than 750 megabytes per second on your home internet connection? I would love to see a screenshot of that speed test.

Office Internet speed - it's 10G. The LAN speed actually is less - just 1G because of the on-board NIC - I just don't need more.

Thanks. I just checked the SATA spec, and hadn't realised just how slow it is.

(High-speed storage I/O has long relied on multiple datapaths / busses, for pretty much just this reason, in dependent of disk I/O standards, going back decades.)

I want the holy grail monitor:

    38” Ultrawide
    144Hz
    200+ DPI (retina)
    HDR

For retina-quality DPI like that we’re talking something like 7680x3200, with HDR it’ll be 30 bits per pixel, 144 times per second.

Without any compression that’s 106,168,320,000 bits per second, which Thunderbolt 5 can’t handle. I guess I need to wait for Thunderbolt 6.

HDMI 2.1 can already do 8K @ 120hz with only 48Gbps, using DSC, and DisplayPort 2.0 is supposed to handle two or more 4K/8K screens @ 144hz. I think we'll have more trouble getting a GPU that can drive that..

The older my eyes get, the more I wince when hearing DPI numbers like that. Scaling of UI elements is never as perfect as we pretend it is. For example, getting both GTK and Qt to simultaneously scale properly on my workstation is something I've given up on, nevermind the bios screen (and I'm only at 109 DPI). Shakes fist at cloud

If a visitor were to go to such a facility, it's quite likely they'd have to sign an NDA and/or that their phone would be confiscated for the duration of their tour of this new technology. But let an executive in, phone in hand, and he does the very thing they're trying to prevent.

Makes me wonder what a Red Team could get away with.

There's a good chance its just a marketing strategy.

So I assume this could be used in USB5 and DisplayPort 3.0 with 160Gbps? That should allow 8K Res at 120Hz without compression or something like 5K with 240Hz.

The certification of Thunderbolt still requires Intel's blessing though. So while it is royalty free and open standard there is still some barrier in getting AMD on board. Although from my view there isn't actually that much of consumer demand for Thunderbolt.

What can actually saturate that kind of a local link? PCIe 4.0 maxes out at 32 Gbps and 5.0 is increasing that to 63 for x16. x32 exists in the spec, but I'm curious what if anything actually implements that? Certainly not any consumer-grade boards.

A Thunderbolt 5 controller would probably have a PCIe 4.0 x4 (64 Gbps) connection to the host.

I would assume that display data carried over TB could come close to saturating it - 4K 120Hz requires a data rate of ~26Gbps. But I think a big main beneficiary here could be in external GPUs. A high performance GPU connected via Thunderbolt 3 currently takes a bit of a performance hit (I could be wrong but I thought in the realm of 10-20%), but if the data rate is doubled perhaps that could be minimal for the current generation of GPUs.

Thought it was giga-transfers?

Wikipedia says:

Version 3.x: 8 GT/s ×1: 985 MB/s ×16: 15.75 GB/s

You can plug a 2x200GbE Mellanox Ethernet controller on 2 PCIe3 16x connectors (linked by a bespoke cable)

Version 4.0: 16 GT/s ×1: 1.97 GB/s ×16: 31.5 GB/s

I think you can also buy eth controllers that will saturate that.

32 GB/s

PAM4 (not 3) is used a fair bit in lower cost 100GbE optical links these days, for medium distances.

We have one HN member here very adept in matters of interconnect: @deepnotderp

Saying from my IC design exposure really many years ago, those ultra fast, and ultra complex short range serial signalling methods are also ultra power hungry.

Past the PAM4, signalling over copper hits a power consumption wall where you can no longer get more speed by throwing more transistors at it, because any speed gains from extra signalling complexity start to be offset by the need to reduce the transmitter clock frequency to keep it from melting.

I believe PAM4 will be the swan song of sophisticated serial signalling before the industry switches to optics.

The industry possibly have one last trump cards in the sleeve: chord signalling, which is currently being developed by Kandabus. It's an extremely complex signalling scheme, with analog signal equalisation, and other signal conditioning techniques bordering on black magic. Its design cost, and complexity may well be higher than going straight to optics.

I believe Thunderbolt 4 has a fiber optic spec coming.

https://www.amazon.com/optical-thunderbolt-cables-Corning/s?...

> I believe Thunderbolt 4 has a fiber optic spec coming

Thunderbolt 2 had a fiber optic version.

As in fiber optic?

Apologies for being ignorant, most my experience with with fiber has been with very small runs of OM3 and OM4 as trunks between network switches. But isn't fiber optic cable less tolerant of the kind of abuse that a short USB cable would see? Being throwing into someone's purse or having it repeatedly pulled by the cable or just being folded in half over and over again, stuffed in a pocket and sat on. Is fiber really suitable to replace the kind of short range interconnects USB handles?

> But isn't fiber optic cable less tolerant of the kind of abuse that a short USB cable would see?

A USB 3, or TB cable will be dead long before a toughened single mode cable from an equivalent amount of abuse.

I had many TB cables which degraded, or died completely simply from gently rolling, and unrolling them a few times.

I had a single mode cable in the office being driver over by heavy carts, and stepped on for years without it giving in.

Multimode cables though are way more prone to signal distortion from mechanical stress because of their their thicker cores.

btw. corning builds optical thunderbolt cables.

Really? Color me surprised. Just reading a bit more into it now, I always thought that it fiber optic was made of glass which was why optical cable was always more expensive and more delicate. Never heard of polymer based optical cables until now, and combined with a more robust sheathing...

Would the total cost of cable then be somewhere in the same scale between an active circuit cable like USB 3.1 onwards vs a optical one with the laser transceivers and the like, assuming data signalling protocols?

The optical cable itself is made of glass in a plastic sheath. It's very cheap because there are millions of kilometres of it used around the globe.

And yes, the tiny few micron glass core is actually more tough than thicker copper cables.

It's only sensitive to extreme, like 180° bends, and a thicker insulation will prevent permanent damage.

The cheapest single mode SFP transmitters cost like 5-10 dollars wholesale.

If they were manufactured in volumes comparable to USB chips, they would've probably cost close to one dollar.

Reply on news.ycombinator.com