Superspeed USB 3.1 Announced, Could Bring the End of Thunderbolt | Hacker News Reader

Superspeed USB 3.1 Announced, Could Bring the End of Thunderbolt

maclife.com

43 pointsbgar47 comments

Mmmmm the warm smell of link bait in the evening. This reply will probably be longer than it needs to be but I find this stuff silly and want to vent.

TL/DR: The media loves writing "Apple v.s. the World" pieces and TB isn't really meant to compete with USB.

I, as well as many others based on comments I've seen elsewhere, think there are a lot of parallels between USB and Firewire and USB and Thunderbolt (TB). USB is a flexible interface you can use to connect to a camera, keyboard, printer, video adapter, hard drive, etc. It does a lot of this in software which means there's some overhead and various versions of the spec have had their own funky issues (like USB 1 & 2 being half duplex[1]). In the end it's "good enough" and no one except device implementors will have to deal with those problems. Firewire and TB are "better" technologies, designed to do some fairly interesting things. Firewire being essentially hot pluggable daisy chain-able SCSI and TB being an external PCI Express interface that can run over fiber optic cabling (whenever that shows up). As a device maker, unless your customers need any of those cool capabilities there's no reason to to use it. Same deal as with Firewire.

My take on articles like this is they're another case of some relatively boring technical enhancement being turned into the usual "My side must win" sort of religious war by the media (and parts of the general tech public) loves to turn these things into. Clicks are clicks after all. Partly this is because TB has been labeled an "Apple technology" because they helped with some trademark paperwork[2] and embraced it at launch. Unlike Firewire, Apple didn't do anything with the spec except say "Hey lets use this type of cable/connector." But honestly, that doesn't really matter because if it's related to Apple, they MUST be in some sort of battle against some force. In this case, the evils of USB...

Honestly, Thunderbolt seems great for Apple as their design aesthetic mandates (annoyingly IMO) a minimal complement of ports and TB doubles as a display connector and a data port. They won't see a cent by it becoming popular[2] aside from the sales of Apple branded accessories, but regardless TB makes it a lot easier for their customers to plug in an Ethernet adapter, keyboard/mouse, and a couple monitors without the need for a docking station (which is what I do with my MBPr) or a USB HUB/assorted mess (which is what I did with my MacBook Air).

Intel, as the creator of Thunderbolt, stands to make the most by it becoming popular (they make all the controller chips). That said, Intel is part of the USB consortium as well and it's hard to see how they can "loose" either way this goes.

[1]http://en.wikipedia.org/wiki/USB [2]http://www.anandtech.com/show/5425/why-thunderbolt-wont-come...

Exactly. As far as I can tell, you cannot connect an external monitor to a USB port and have a GPU drive it. Thunderbolt serves a different purpose than USB: it is an interface between a small number of external devices and a small number of hardware components. The fact is that it can only at this point be useful for three things: displays, drives, and Ethernet. USB is what you use when you do not have another option. It is nice since it can support almost any kind of device because the protocol is not burnt into hardware, but of course the downside is that it has to implement the host side of things in software. Anyone trying to use a "USB video card" to play an HD video will notice how much of a toll it takes on the CPU. And when the CPU is fast enough not to be bothered by a 1080p video, we will instead be talking about 3D 8k video.

Basically, it is nice that USB just got faster, but IMHO direct access to the specific hardware (disk controller, video card, etc.) will always win.

Agreed. I used a USB 2.0 video adapter on my Air and while it was fine for a terminal, it probably managed 25FPS at best (Youtube is choppy). It also has to be by itself on its own USB controller (well KB & Mouse are fine) due to the Half duplex nature of USB 2.0. Using it on the same bus as a HD makes the display lag badly. Still cool that it can be done.

The new USB 3.0 ones probably fix all/most these problems but I just use monitors 1 & 2 for that ;).

I don't care one way or the other what apple does, but I find it ludicrous to be expected to buy a very-high-tech active cable to run a distance of five feet.

Also there is the fact that Intel can and does block devices from using thunderbolt because they don't like the category. That is not something to support. See the quote in http://forum.techinferno.com/diy-e-gpu-projects/4271-2013-11...

I'm not knowledgeable in this area, but isn't Thunderbolt a more direct interface (since it's basically external PCI-E) with less overhead than USB, making it a better choice for devices requiring low latency (audio interfaces, graphics cards, SSDs, etc.)?

Regardless, I believe Intel's Thunderbolt is here to stay, along with USB 3.x...

Having never tried to program a thunderbolt bus driver, but knowing that writing USB packet control code as being the biggest pain in the ass because the entire standard is a hodge podge mess, I wonder if Thunderbolt is easy to implement connectors for.

Really though, the problem is like most licensed Intel tech they charge out the ass for licenses. They heavily jack up Intel CPUs and Thunderbolt to subsidize their huge R&D budget.

Thunderbolt is just a long distance version of PCI Express, so existing PCIe device drivers should work just fine.

This sounds very interesting. I guess the only reason there has been no adoption of thunderbolt is just the silly license rent seeking Intel tries to pull on the interface, a lot like firewire.

Licensing fees have nothing to do with it. The reason is that Thunderbolt is expensive to manufacture, and mass-market computers have tiny profit margins.

Adoption is starting. Newegg lists a variety of Thunderbolt gadgets. Mostly it seems to be storage products, where 10 Gbps Ethernet is needed but makes Thunderbolt look cheap.

How do you consider the USB standard a mess? I am quite impressed with it as a low-cost peripheral port that requires minimal hardware resources. I agree, it can be a pain to write for. My biggest difficulty was crafting HID descriptors! Yikes!

Have you ever tried to push more than 6 keys on your keyboard? e.g. in local two-player games.

Modern gaming motherboards come with a PS/2 port for a reason...

Modern gaming keyboards come w/ full n-key rollover over USB. My Noppoo Choc Mini is capable of full NKRO whereas my Model M can't do the same over PS/2.

USB doesn't impose a simultaneous key press limit that low, the keyboard does. The 'Report count' field of the 'Key codes' usage page of the keyboard's USB endpoint descriptor determines the maximum number of simultaneous key presses. This field is a byte, so it could be up to 255.

The keyboard endpoint example in the USB HID spec does have 'Report count' set to 6.

Greg Kroah Hartman wrote about implementing Thunderbolt for Linux: http://www.kroah.com/log/linux/hardware.html (not a pretty story)

I suspect Intel has realised that people who need Thunderbolt have no practical alternative. May as well charge what that market will bear for a while.

AMD does have a low cost alternative - Lightening Bolt http://en.wikipedia.org/wiki/Lightning_Bolt_(interface) - and I will admit that if it takes off and beats Intel/Thunderbolt I will smile a bit. It will be AMD64 all over again.

I'm curious as to how the hell Intel can charge huge license fees on a pci express passthrough port. Though, it doesn't sound like Lightning Bolt is similar - the Displayport part, maybe, but USB is serial and has high latency compared to the pcie bridge.

> I'm curious as to how the hell Intel can charge huge license fees on a pci express passthrough port.

They aren't - they charge the fees for Thunderbolt. You can't claim something supports Thunderbolt without paying licensing fees.

> it doesn't sound like Lightning Bolt is similar

You can draw a venn diagram and find that all three have an overlap. A desire for fewer cables, to be able to connect lots of stuff, and bang for the buck. Display, storage, peripherals etc. Outside of the venn diagram, Thunderbolt does have low latency, but most don't need or want to pay for it. If Thunderbolt support was as pervasive and cheap as USB, it would be a different conversation.

I agree. I have a hard time seeing thunderbolt dying because of USB 3. People said FireWire would die when USB 2 came out, but it's still with us (never nearly as popular, but still there).

I just wish the USB people would get a brain about naming. Full Speed? High Speed? Super Speed? Super Speed Plus? The numbers work fine, and don't confuse the hell out of consumers.

I guess that's why we shouldn't use them.

Some folks I know working on the technical side refer to this iteration as Ludicrous Speed seeing as the naming has gotten so convoluted.

Correct, although I would also mention high-end video; probably the killer app for Thunderbolt. As you said, audio interfaces are a big one, particularly at the high-end, where high throughput is desirable but low latency is essential. As for graphics cards; it is really a general expandability solution, which is much simpler for less-technical users than opening up a PC and attempting to install an expansion card. The new Mac Pro design seems to follow this line of thinking.

Thunderbolt will always be a more niche market than USB, but I could see it having an even longer life than FireWire (which is still popular for high-end video and audio applications).

Speaking of the Mac Pro, I was wondering whether the 6 TB2 ports are independent and more importantly, could they be aggregated (like for external SLI or Crossfire)? That would be really interesting...

> but isn't Thunderbolt ... a better choice for ... audio interfaces

I certainly see your point and am in no way disagreeing with you, but this made me laugh.

May I know why? I've used USB and Firewire interfaces, and while both sucked at driver support (plus Firewire is pretty bad on PCs in general), the latter was much better, especially for live work...

Anyone knows how far this 'basically external PCIe' actually goes? If it's really PCIe, does this mean I can get any of my current PCIe cards, solder a TB cable onto them and use them as such provided I have proper drivers?

Yeah. There exist dead expensive adapter boxes and AFAIR they only get either 4x or 8x speed, but not all 16 PCIe lanes.

Here's a guy connecting a graphics card to his mac via Thunderbolt.

http://hexus.net/tech/news/graphics/58429-macbook-air-uses-g...

    "Normal" GPU + external Powersupply
     |
     |
   Adapter: PCIx1-Slot to Express-Card
     |
     |
   Adapter: External Expresscard-Slot to Thunderbolt
     |
     | <-- TBolt cable
     |
   MacBook

nice.

Thunderbolt having direct memory access makes me slightly uneasy. I'm more inclined to fill the ports with superglue than actually find a use for them (I won't mind you). If the price of the accessories comes down, then it'll certainly be an incredible boost for some applications.

If anybody is interested in learning about them, there's direct key stealing attacks for both TB and FireWire if you look around.

Thunderbolt devices are currently a joke. The use case would be being able to push a laptop's GPU out to an external display card, so when you're docked you can run whatever desktop GPU you want.

But the cheapest connectors for that sort of thing can't to x16, and cost $800. Which, for the price, would let you buy an entire desktop.

The use case would be to connect to the dock with 1 cable rather than a huge port on the bottom. Or connect to an external drive bay for fast and low overhead access.

The external GPU thing has always been the most fanciful use case.

Is anyone doing anything interesting and consumer-friendly with Thunderbolt? All I've seen it used for is DisplayPort, a slimmer DVI connector.

Apogee has at least one interface that's Thunderbolt ready and I'm sure a few other audio companies are working with it as well. Edit: Universal Audio. Others have plans/announcements.

Other products: external drives, external GPU (demoed by Lucid).

While trying to refresh my memory I came across this: https://en.wikipedia.org/wiki/List_of_Thunderbolt-compatible...

The Apogee Symphony can be connected to Thunderbolt via the $1000 "ThunderBridge" (http://www.apogeedigital.com/products/symphony-64-thunderbri...), but it feels more clunky than using USB which can connect directly to the Symphony without the intermediate hardware.

The ThunderBridge gives extra benefits (the Symphony with USB alone can't go above 96kHz or 16 channels).

Yes there are a bunch of tremendously expensive external PCI card cages that cost $200 per slot and which are attached to your computer with $99 cables.

The Apple Thunderbolt Display is a nice docking station solution: video, audio, USB and Ethernet all running over one cable. And if you have two displays, you can daisy chain them and still only have one cable connected to our laptop.

Any idea what is driving these advancement? Why USB was slow in past and suddenly it is possible to get 10X more speed?

USB 1 was slow because it was designed to be a cost-effective way to connect keyboards, mice, etc. result: it did not require high-quality cabling, had the PC poll all devices for data, etc.

USB 2 stretched the USB 1 interface to its limits (for the technology of the day, and for a desired cost level). Devices could indicate that they could handle higher-frequency signalling and if so, they would be sent such signals, Backward compatibility meant that low speed devices kept using the slower timing, even if they shared the bus with higher speed devices. So, low-speed devices still use the same fraction of the total bandwidth. For example, a mouse will want to be polled a hundred times per second. It also will want be cheap, so it will use the slowest USB speed. Result: if it took 1% of USB 1 bandwidth, it still takes 1% of USB 2 bandwidth, while USB 2 is a lot faster. And because devices have to be polled, that is regardless of whether you use your mouse (yes, your mouse can take 4Mbps of your USB 2 bandwidth)

If USB 3 went along this line, things would have gotten ludicrous. Because of that, USB 3 uses different cabling. That allowed the, to also fix some issues that USB had. For instance, devices can now signal that they have data to deliver, whereas they had to be polled before.

Summary: USB 3 is a totally new protocol that uses different (higher quality) cabling and only shares some terminology with USB 2.

But are usb plugs now a noticeably different width to ethernet ports?

I'm really interested in the 100 watt power bandwidth. That borders on making usb laptop chargers, saving a port on the side. Really expensive ones to support 3.1, though.

The new Mac Pro coming out this year and its slim design is only possible by Thunderbolt. Apple expects you to use your huge Raid Arrays and other 3rd party hardware through a thunderbolt connection. USB can do this too of course, but not as fast.

If size matters at all in a workstation class device is a different question entirely, but in case you need a portable workstation (for mobile audio/video work for example), Apple is now the way to go.

Is USB 3.1 going to get faster charging for mobile devices, too? That should be one of their priorities.

I fail to see how charging would be a priority of a protocol essentially designed for data transfer. I mean, it's a nice idea and there's enough consumers that would benefit from it but imo they'd better focus on the communication part of things instead of cranking up the amount of mA it should be able to deliver.

USB is also a standard for chargers now, so why wouldn't it be?

Good. Thunderbolt is another lame Intel scam.

Thunderbolt is a back door. Generally, physical access to any machine and "all bets are off", but still, I'd rather not have an orifice on my machine that says "fuck me here please" no matter what my business. USB can be implemented in ways that are more secure, but they are often not.

Yeah, I won't have an enabled 1394 connector on my PCs; USB is highly filtered and won't give you the bus-level keys to the kingdom.

That said, USB is a huge pain in the rear to deal with, especially isochronous traffic.

Thunderbum?

In principle this could be mitigated with a suitably programmed IOMMU which can limit the operations that a DMA capable device (e.g. on the other side of the Thunderbolt connector) can do with your main memory.

Traditionally they are used when you have an operating system running in a virtual machine, but you want (for whatever reason) give it hardware access to a PCI card. Without the IOMMU, the virtual machine could program, e.g. a network card to read/write to memory that belongs to the host operating systems, hence breaking free of the virtualization jail.

But, as far as I know, it was also possible to limit DMA in firewire controllers. Was that implemented in practice? I don't know. So will it be done for Thunderbolt-Enabled computers?

Reply on news.ycombinator.com