Is there any reason that USB 3.0 couldn't run over Thunderbolt? Since it's using a PCI-Express link, all you'd need is a chipset built into a Thundebolt dock or some sort and you should be able to get at least 2 USB 3.0 ports thAt could run saturated without slowing each other down.
Plus, I also think that we'll be seeing some PCI-Express add-on cards this year as this technology is VERY attractive to the production professional who needs to deal with a lot of data at once. I would assume other connectors which allow a direct connection to a computer instead of over a network cost MUCH more than what Thunderbolt does.
Since there's PCIEx4 coming out of Thunderbolt, is it possible to have a separate discrete graphics card running outside of the laptop? That could be interesting...
On paper it's definitely possible. And this would help explain why we didn't see any external PCIe setups this year at CES. However there are some caveats.
1) Intel hasn't told us what the latency is like. A 3 meter cable isn't very long, but PCIe is a very fast connection - a single lane runs at 5GHz. Without knowing much about the link layer, I can note that at 5GHz, even light only travels 6cm/Hz. For the copper implementation of Thunderbolt this would be worse. A reasonable guess would be that you're looking at up to an additional 50 clocks, on top of the latency between the Thunderbolt controller and the CPU.
The point of this being that I'm not sure what the tolerance is for PCIe latency with today's GPUs. It may not work out of the box if the GPU expects a response within X cycles.
2) Performance will be impacted. PCIe x2 is 8GB/sec, so a single Thunderbolt channel is only a bit more. You can run a GPU on such a setup, but you will be severely bandwidth limited, and this of course gets worse the faster the GPU. As a result performance would not be comparable to desktop parts. For something like a Radeon 5770 (I throw this out as it's a reasonable upgrade for most notebook GPUs), I'd guess it would achieve 75% of the performance, but that's very much a shot in the dark.
So in short, yes, it should be possible. If not with this generation of GPUs then with the next one. But with a copper connector it's going to be bandwidth limited.
You have 2 channels though, which gives you PCIe 4x bandwidth, testing has found that in most games a 4x link only results in a few percent of penalty. There're exceptions like MS Flight Simulator X that take major hits but very few other games do.
The fact that there are 2 channels is correct. However at this point we have no idea if the channels can be teamed, hence my numbers assume they cannot.
While bandwidth might not be impaired by such a link, latency might very well be. And the result of increased latency might be much more visible than a 10% loss of framerate (or there might be no issues at all). And this might even change from an application to another.
I wonder why Toms stopped testing below 4x. While not as realistic a test case, it covered people using desktop GPUs on laptops (via a breakout box and a 1x PCIe ribbon cable) and showed worst case performance.
OTOH IIRC that test showed a much larger 4x penalty than had previously been seen with on the 280 or 8800 tests.
1 cycle @ 5GHz would be more like 2*10^-9 seconds giving you .6 meters, i.e. 6 cm.
For quick mental checks think about how large an antenna would have to be to pick up a 60 meter wave with any efficiency. An even easier thing to remember is that the wavelength of a 1 GHz wave is about a foot.
While I believe you're right about it being 6cm (as Google agrees), 6cm is .06m, not .6m. .6m is 6dm (decimeters), a measure which is hardly ever used, as meters and centimeters cover most human uses well enough without need for a unit in between.
"2) Performance will be impacted. PCIe x2 is 8GB/sec, so a single Thunderbolt channel is only a bit more. "
Arrgh. 1 Gigabyte(GB) = 8 Gigabits (Gb) You're confusing people because Thunderbolt is only 10Gb/s, not 10GB/s.
A PCI Express 2.0 x16 Link is 64Gb/s, that's the slot commonly used for video cards and is over six times faster than Thunderbolt in it's current incarnation.
When Thunderbolt is optical and supports 100Gb/s then we can start to look at external video cards.
"When Thunderbolt is optical and supports 100Gb/s then we can start to look at external video cards."
That's probably the key point. The issue at the moment is when 100Gb/sec Thunderbolt will arrive. PCIe 3.0 arrives within a year, and it will push PCIe x16 to 128Gb/sec, at which point Thunderbolt would again be slower, although not by nearly as much as before.
And yes, I screwed up the units, thanks for catching that. This is what happens when you write a comment at 1 in the morning.
Why does carrier frequency have anything to do with signal propagation velocity? Wait it doesn't. I would suspect that the protocol overhead is far greater factor than the velocity of light in a conductor, which at these length scales is going to be less than 20 ns. Hardly a delay worth mentioning.
It's still on the level of several GPU/CPU cycles - so if you need to actually interactively communicate between the two, both sides will lose massive amounts of their potential performance. Of course, if you need massive intercommunication for your computation you're lost in any case, it's only that lately the physical limitations are becoming evident where previously there were merely implementation problems (we could probably link up a huge amount of 386's today without any loss of performance - the same was not possible in their era) So if you're only streaming, signal propagation isn't a problem. It's only in latency bound scenarios where insanely high speed individual components will not be able to run at their full potential. Along this self same axis, we are probably already at )or quickly approahing) the hard RAM-latency wall, until we move that too onto the CPU...
No, I think it was originally supposed to support everything, but they cut it back probably for cost reasons. I'm sure interpreting more protocols increases the complexity of the controller, ect. Also, the change to copper may have had an influence on that decision. Or maybe it's an intel/apple conspiracy to eliminate USB.
It sounds like something along the lines of a PCIe USB controller + Thunderbolt controller would work. Where this will really come in handy is having things like a USB3 controller built into a computer monitor. That would eliminate at least one cable from the monitor the PC.
Thunderbolt technology is like Firewire on Steroids with the same problems plaguing it. Too complicated and need to connect to display adapter. Not to mention royalty fees. USB is a cheaper, simpler technology and it works.
"Thunderbolt technology is like Firewire on Steroids with the same problems plaguing it. ... Not to mention royalty fees. USB is a cheaper, simpler technology and it works"
Wrong.
Directly from the article: "Similarly, there’s no per-port licensing fee or royalty for peripheral manufacturers to use the port or the Thunderbolt controller."
usb 3 is dead..at least for Intel. Since Intel is the ONLY validator for USB 3.0 Intel isn't going to bother with usb 3 anymore as USB 3 doesn't reap the billions thunderbolt is going to reap. Every one of those little ic's that's needed on each end of the cable..Intel gets paid for those. It's not an open tech..so be expecting others to have issues(linux...other oses).
"The connection supports a daisy chain topology, and Thunderbolt also supports power over the cable, though it isn’t clear how much. I’d expect no less than the 5 or so watts that USB 3.0 supports."
"Thunderbolt also provides 10 watts of power to peripherals, so you can tackle workstation-class projects on the go."
At least as implemented by Apple, Thunderbolt supplies 10W of power, which I believe is less than Firewire but more than USB 3.0.
And is the Thunderbolt controller really that big? I guess we won't be seeing Thunderbolt flash thumb drives anytime soon. And I can't think Apple could be happy with this since putting that controller in the iPad PCB will already be difficult much less the iPhone or iPod Touch. Thunderbolt would seem to be ideal to sync 64GB or probably upcoming 128GB iDevices.
"And is the Thunderbolt controller really that big?" From pictures on the Intel site it looks like the optical one is the size of a phone jack but much flatter. Sexy!
Yeah, that Thunderbolt controller was quite the porcine wonder!
planned obselescence...same idea that its not profitable for any company to make 50 year lightbulbs....unless you charge buttloads of money for them...then who would buy them? Intel wants you to buy their next boards too. hehe
Or did Intel pay Apple for the use of the part of Light Peak that depends on Apple's display port technology by getting Sandy Bridge processors to Apple so quickly.
DisplayPort is not "Apple's...technology". It's a fully spec'd out standard on its own, and is actually a breath of fresh air after using TMDS-based connectors for a LONG time. As for the port, Mini-DisplayPort has been part of the DisplayPort standard since 2009. Intel didn't need to pay Apple to use it at all.
in fact, that's why Apple uses it. Apple doesn't have to pay anybody anything to use a DisplayPort adapter, unlike an HDMI adapter. There's no licensing costs on DisplayPort, unlike HDMI.
My Dell monitor and my AMD 6970's all use DisplayPort and Mini DisplayPort. Many things use it. It's not an Apple tech in way, they have nothing to do with it.
Do I understand the article correctly that we now avoid spending big money on GbE adapters and just link two computers with Thunderbolt and run it as an ethernet connection?
If so, I'm a big fan, particularly if we could do this through a hub. Any ideas on the maximum cable length?
You can run networking over Firewire. Judging from the popularity of that feature, I don't think that networking over Thunderbolt would be a very popular use case.
Networking over firewire is quite popular for diagnostic and data transfer scenarios. For example, if you firewire two macs together, you can boot either of them off of the other.
For casually tossing large files around Thunderbolt sounds more attractive than many other ad-hock network solutions I've used.
I think Gigabit ethernet has come standard with most computer and laptop motherboards made in the last several years. What do you mean, "spend big money on GbE adapters"? Or did you mean 10GbE?
I also don't see why you would use this over a **much cheaper** 10GbE ethernet cable. For the distances its capable of you could go with much cheaper SATA3 for 6Gb/s, or probably SATA4 by the time this becomes available on non-Apple stuff.
My guess for the wire based lengths would be that you could get cat7 performance which in theory means max 40GbE at 50 meters. Much less in practice of course. Lets assume only 2x10GbE at 10 meters or 30' based on currently shipping speeds. Depending on cost and the latency constraints mentioned earlier the lengths may be specced even shorter. Chaining extends the total length.
Optical of course should add a nice multiple to the lengths.
I hope we do not waste too many sputnik-broadband $ on thunderbolt. I want the optical goodness!
While the current implementation is copper based having copper wires for both data and power, the future implementations will have fiber for data and copper for power. The power lines will not be eliminated from the fiber implementation. The cables and connectors will be a combination of fiber and copper.
The only reasons for the fiber implementation are increased distance and increased speed for the data. Intel has claimed that under the fiber implementation they expect to have 100 Gbps per channel by 2020.
The power supplied is DC power. Within reason, this is independent of the data and the distance of the cables. Therefore no significant change to the power lines will be required when it goes to fiber for the data side of the link.
"The power supplied is DC power. Within reason, this is independent of the data and the distance of the cables. Therefore no significant change to the power lines will be required when it goes to fiber for the data side of the link."
The problem of delivering low-voltage power over 3m and over 100m are completely different. PoE has to do the same thing (deliver DC power over 100m of cabling). In order to do so, it requires a relatively high voltage. I believe that PoE puts 57v into the cable, and expects to get at least 37v at the other end after resistance loss is accounted for. If they go for the same 5v as USB, they're going to have major issues at 100m.
"Tunnel a PCIe lane over the link, and you can dump it out on a peripheral and use a local SATA, FireWire, USB, or Gigabit ethernet controller to do the heavy lifting. Essentially any PCI Express controller can be combined with the Thunderbolt controller to act like an adapter."
So can thunderbolt transport USB and SATA or not? Would it be possible to design an external hard drive with a Thunderbolt interface, and have it implement SATA? Or USB 3.0? Or how is this supposed to work exactly?
I want 12 Thunderbolt ports and no USB/eSATA/HDMI/DVI/Displayport.
you can transport USB 3 or whatever else you want over thunderbolt if you "tunnel a PCIe link" over the link.
meaning you have to use a PCIe adapter to send the data as PCIe protocol, so thunderbolt can transport it. then, on the other side, the signal gets dropped into another PCIe adapter and either converted into something else, like USB 3, or maybe SATA, or just sent as PCIe to wherever.
it was originally going to carry ALL of the various protocols that are around today, and then drop data off at whatever adapter was waiting for it with no middle-men, but intel changed that for some reason. i think it must be cost-related, but i'm jaded and distrustful.
... at 2 metres. DP 1.2 is 17.28Gbps at the same distance. I love the idea but i think they need more bandwidth to support multi-monitor, high res/refresh setups since if you are sending data and video over the same wire it may become bottlenecked faster than you think. I also don't get why there is a 10Gbps interface when the fastest internal drives are capped at 6Gbps and therefore an eSATA drive on a SATA 3 controller will send data (obviously not video) at the same rate as a 'Thunderbolt' port.
As many people are asking though, it will support any other interface you can put the controller on, so you can have Thunderbolt to USB adapters or eSATA or Graphics or RAID cards or Network or whatever you can plug into a normal PCIe slot.
Seems like today 4 more of my PCIe lanes are wasted on a not needed (at this time) technology. As for the future when it moves to fibre and we ditch the crappy SATA interface for internal storage, I think I'll be sold then.
It may be cool that I can plug any usb device into any version of the port and it will work, but backward compatibility slows down forward advancement. Everyone should realise this due to the sandy bridge woes that were caused by a transistor used for some type of legacy operation. New port = more win, sooner.
My biggest question is about the licensing of this proprietary link? Will Intel license this technology out to other integrated circuit makers like NEC, AMD, TI, NVIDIA etc? And at what cost will they license this out for? If Intel will be the only company producing Thunderbolt chips then I am not interested at all because you will not see widespread adoption of this interface like we see with USB.
I have no doubt Intel will license out the controller but I doubt many will use it to integrate it into anything for now like usb 3.0 but worse. That chip is massive and is likely a 45 or 32nm part so even at 28nm it would be kinda large to put on die with a chipset or soc. It would be interesting to put it on a video card though since they could use it to output all the video signals and it kinda would make sense with PCIe 3.0 just around the corner you'd have alot of unused bandiwth on a 16x port to abuse
Not supporting other protocols is a mistake. Instead of supporting motherboard's quality controller in drivers there is need to support myriad of controllers with different quality and drivers. We will definitely see Thunderbold -> USB3 -> SATA combination for HDD's and thats bad.
I just read over on pc perspectives that this may not be true. It's possible that this particular implementation of lightpeak (thunderbolt) is an apple only thing, and that there could be others which use different adaptors and support other protocols.
This info is in the list of updates they made at the end of the article.
It's all very fancy, but I am 99.99% certain this will see the same fate as FireWire and e-SATA: Technically superior but relegated primarily to large-scale external storage for enthusiast/professional use. Though really, even that is in doubt since USB3 will never bottleneck mechanical storage devices. USB3 is "fast enough," has years of head start, has no associated licensing fees, and most importantly is fully backwards-compatible with the peripherals and cables everyone already owns. Thunderbolt is commercially licensed, currently only available in a handful of very overpriced computers, and is only backwards-compatible with a port that no one but select enthusaists with Eyefinity6 setups and die-hard Mac fans uses anyways.
Inertia is an extermely powerful force in consumer electronics. Even Apple's legendary buzzword marketing couldn't get any consumer traction for DisplayPort, Mini-DVI, or Firewire. I really don't see how this will turn out differently.
Personally, I'm looking most forward to docking stations that one day work with nearly every notebook. Just one cable from my notebook to the docking station and there you have e.g. USB ports plus eSATA, Gigabit Ethernet and DisplayPort etc.
Wow, that controller is huge. No way Apple is putting something this big in an iDevice this year.
And I would add one thing to your Thunderbolt article: That it can support multiple topologies, not just daisy chains (according to the Intel PDF). "A symmetric architecture that supports flexible topologies (star, tree, daisy chaining, etc.) and enables peer-to-peer communication (via software) between devices."
Yes, this is something most people miss -- the variable topologies allowed.
I'd bet, in the final analysis, it is like Firewire in the possible topologies allowed. The only topology not allowed in Firewire is a ring of any kind (e.g., you can't create a tree/star combination that can result in a ring).
Btw, copying the frame-buffer over to the Intel iGFX or using it's connection means there is no DisplayPort 1.2 support right? I.e. no daisy-chaining. Or does it use the discrete gpus DP for that?
Since HM67 is PCI-e 2.0 compatible shouldn't it be x2 PCI-e connected to the chip? The Promise Thunderbolt device uses that anyways. It's the cpus x16 lanes PCIe 2.0 plus the chipsets x8 lanes PCIe 2.0. I kinda see it as a successor or competitor to ExpressCard 2.0, twice as fast and not as annoyingly space consuming. A more universal and consumer as well as prosumer friendly connection. Macs will get Native USB 3.0 with Ivy Bridge and the 7-series chipets, until then you could create a Thunderbolt > PCI-e 2.0 x1 NEC USB 3.0 controller converter. I still think it's useful for slightly different uses though, you won't see self powered Thunderbolt harddrives by any means. You already have self powered USB 3.0 devices. I think it would be quite hard with a drive and SATA-controller powered by the mDP cable. Definitively won't happen to memory sticks at least :) Although maybe 10 W is enough. It just won't fit in memory sticks though :) I guess they will move to a pci-express interface though.
So after integrating the gpu onto the cpu and in theory reducing the chip count by one, they then turn around and add another IC to the motherboard. lol
Does the fact that the cabling is copper-based mean that five years down the line I will find myself with a draw full of TB wires yet somehow never with the one I need, be it copper-to-copper, optical-to-optical, or copper-to-optical? If so, I would have rather that Intel held off another six months for a pure optical solution.
From the description here, it seems like the DisplayPort pins and the Thunderbolt pins are physically separate; they simply added pins to the DisplayPort connector/cable for Thunderbolt. This would explain why the connector is backwards-compatible with DisplayPort. It seems as though Apple and Intel are concealing this fact with marketing lingo, but it's obvious to me, at least, that the reports of Thunderbolt combining DisplayPort in its protocol are false.
At any rate, I think most readers here are missing the key benefit -- it will be very easy to dock your MacBook. Because this is all based on DisplayPort and PCIe, there are many existing controller chips which could be put in a dock or inside a monitor -- SATA, second/faster graphics board, Gig Ethernet, and more USB ports. And this can probably be done with existing chips, or, at most, with only slight tweaks. I don't see this as a new protocol or standard; it's really PCIe in a cable, but that's a good thing.
USB is here to stay. Unless the plan is to give people Thunderbolt to USB converters (bad plan). USB devices are simply too widespread to be ignored. USB 3 is simply an evolutionary step we'd be silly not to take. There's even plenty of USB 3 storage out there already.
OTOH, I think Thunderbolt offers some distinct advantages over USB and therefore could find its use for some high end applications. The only problem I have with it is that it's too slow. Think of how awesome it would be to have powerful external GPUs etc.
So I think USB 3 and Thunderbolt will complement each other. Hopefully we'll now get rid of eSata, Firewire and perhaps even various video (and network) connections.
There are no USB 3 drives. They are SATA drives with a SATA-USB 3 bridge. With this technology, you could attach a USB 3 or SATA drive, by having the right host controller on the other side of the cable. The cheapest and fastest way, I think, would be to put a SATA RAID chip on the other side of the cable.
This looks very promising and I tend to trust it considering it was developed by Intel. I imagine at some point Intel will need to break away from Apple on development so that some things may be standardized.
I understand that this is just a PCIe and DisplayPort link. However, for all of the various uses and protocols, it sounds like chipsets and controllers are going to need to be embedded in devices or cables for all of that backwards compatibility with Firewire, USB, etc that they claim.
It also seems like this will be quite expensive for some time.
I also think the technology needs to be made available on PCs for the industry to catch on. Or, even ExpressCard to Thunderbolt adapters for PC users may be adequate.
My hesitation right now is that 10Gbps is a burst theoretical claim. Unless you are connecting to a very fast RAID array that is truly 10Gbps internally, you're not going to see those kinds of speeds. Hard drives or SSDs are no where near that fast yet... so don't go out and buy that new Western Digital external drive that only has one or two drives in it. It's not going to be that fast.
I don't see a whole lot of benefit from having DisplayPort in the connector. Perhaps if you could put GPU's inside of displays, you might be onto something. The display's GPU could then send accelerated graphics back to your laptop LCD. Other than that, it seems like you would be paying for it to be there even if you didn't need that aspect of it. Maybe we'll see PCIe only Thunderbolt cables. I don't see a lot of benefit from daisy chaining devices either. It was there in Firewire and I never used that feature.
I would like to see an interface that pulls double duty and connects the same way externally as it does internally, providing both power and data.
In 8ns, light only travels 2.4m. Through copper it will go even less. So either the cable needs to be shorter than that (maybe 3 feet instead of 3m?) or the 8ns figure is wrong. Actually the 8ns is crazy anyway -- why would your external peripherals need lower latency than your RAM?
Excellent point. And for some applications, the latency would be limited by a round-trip along the cable. Group velocity is likely only around 1x10^8 m/s. That would give a maximum cable length of 40cm (about 16 inches) for 8ns round-trip latency.
1. didn't see in the article if it used one of the pcie lanes to do its deal. does it?
2. it seems the apple implementation is severely neutered. why bother introducing it at all at this point in time? what do they expect it to do? link to (small) external storage? why bother? nothing really needs it currently. big external storage? you probably have it setup through the network on 10GbE. link to display? uh isn't that what the vga (for projectors or monitors) and hdmi (for tvs) currently does just fine?
Yes of course, and it's not always you need it for gaming.
But Thunderbolt isn't a replacement for real external PCIe interfaces for graphics. Having external PCIe 2.0 x2 devices for storage is great though. In servers there might be some merit to x8 cards but here 2 PCI-e 2.0 lanes should be more then fine. By a lot. It's enough for a high-end video work-flow any way. Definitively a more professional setup then to go with NEC or whatever USB 3.0 chip. Which there is still no OS X software stack for any way. From Apple. Lacie has release a third party driver though. For those that can extend their system using ExpressCard or future Thunderbolt > USB 3 converters. (Or by PCIe cards).
If this was coming on motherboards a month or two from now, I could understand the hoopla. But this won't be in laptops you can BUY for a year, quite likely a bit more than a year. So why the hoopla?
It's a delaying tactic. USB 3 is here. It works. And yes, it could transmit video if you so chose, and you wouldn't notice the diffrence between 5Gb/s and 10Gb/s. Quite likely the speed of Intel's solution is sensitive to cable quality and length, so I'm not even sure calling it a 10Gb/s connection is really reasonable in most consumer level solutions and situations. But that's fine.
Intel, come see us in a year or so when you have a product to SELL. And it'll be a year (2 years from now mind you) before you'll see this interface widespread in actual devices.
So nice technology demonstration. The product will be interesting when it gets here, but it is largely irrelevant to this year let alone next.
In the article you state... "Thunderbolt is dual-channel, with each channel supporting 10 Gbps of bidirectional bandwidth. That’s a potential 20 Gbps of upstream and 20 Gbps of downstream bandwidth"
I don't believe that's correct. I think it means you can have 10Gbps streams going up and down simultaneously.
There's no mention of 20 Gbps on the intel or apple pages about the tech - if it was capable of pushing that many bits one-way, you think they would be marketing it as such.
This is what intel says about the throughput: "Dual-channel 10 Gbps per port, Bi-directional" and Apple states: "Thunderbolt I/O technology gives you two channels on the same connector with 10 Gbps of throughput in both directions."
To me this means there are 2 channels of 10 Gbps each, one up and one down.
That really isn't the issue. If you look at Intel's datasheet you see that devices use their native PCIe drivers. That means each device would essentially be a PCIe device. Thunderbolt essentially wraps the PCIe signal making it transparent to the end device. The device simply thinks its directly connected to the PCIe bus, just like an internal PCIe add-on card is connected today. Thus, this would be similar to the existing PCIe SSD units already being sold. And would likely require drivers for every manufacturer's controller unless already bundled with the OS. This also inflates the cost of each component. The internal SATA controller on your chipset provides multiple port defraying the cost of implementing a device. Having an internal DVD on thunderbolt serves no purpose than to inflate its cost and implementation. That said, future IO devices after current optical drives have been phased out, may use a more evolved version of thunderbolt to internally connect.
Thunderbolt is most useful in external situations because computers are shrinking for the large workstation towers of old to new slimmer form factors. Thunderbolt makes it practical for machines with less internal space to host a variety of high performance peripherals and add-ons.
Direct connection is likely out of the question, but it could be proxied through thunderbolt to thunderbolt transfer cable; similar to the usb to usb transfer cables already available. However, its cost would be similar to the existing usb solution and thus make it costly to link multiple computers together; costly, in comparison to existing GbE.
Still I agree it would be cheaper than current 10GbE. However, if for no other reason that simple demand-- if consumers desire greater bandwidth than current GbE it would drive down the cost of 10GbE. Currently the majority of 10GbE cost lies in its switches; "dumb" switches would need to be developed to fill this need.
There is one note about your example. You have two computers sharing one RAID. That doesn't work. The same issue already occurs with enterprise customers with FC RAIDs that want to share it between two workstations. You need a special file system and locks to prevent each computer from corrupting data on the device. Or you have to set up two separate "partitions" thus making two separate RAIDs. The simplest solution for consumers is to have the RAID be a special NAS with each computer having a driver to access the NAS in a network like fashion. I don't expect such a solution to come cheaply.
price = adoption level i wont be buying bridge devices that hold a very high cost for a very long time. first seeing them in apple products is bad since this adds a perceived value to it which will make other companies tie an un necessarily high price to its use hoping that they can ride apple's coat tails and make higher profits.
fortunately for me USB 3 dropped in price dramatically since inception.. I can buy a 2 port host controller for 20$ and a SATA 2 controller for 15$ (and falling) if "thunderbolt" can carry these numbers, Ill be all aboard and be waiting.
I understand the beauty of this solution to be that manufacturers can essentially take any existing PCIe card out of desktop (a USB3 PCIe card for example), stick a Thunder Bolt chip on it (PCIe to TB bridge) and TB connector on it and boom... you have an external USB 3 hub.. or external whatever hub you want. Now it might be even simplier to bypass USB3 and just go straight from SATA to TB, but my point is creating a external hub for any connection that already exists in PCIe card format should be nearly trivial, no?
I'm not entirely clear however on the evolution of LightPeak into Thunder Bolt. Was LightPeak always running over PCIe, or is that something entirely new?
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
106 Comments
Back to Article
dagamer34 - Friday, February 25, 2011 - link
Is there any reason that USB 3.0 couldn't run over Thunderbolt? Since it's using a PCI-Express link, all you'd need is a chipset built into a Thundebolt dock or some sort and you should be able to get at least 2 USB 3.0 ports thAt could run saturated without slowing each other down.Plus, I also think that we'll be seeing some PCI-Express add-on cards this year as this technology is VERY attractive to the production professional who needs to deal with a lot of data at once. I would assume other connectors which allow a direct connection to a computer instead of over a network cost MUCH more than what Thunderbolt does.
mgl888 - Friday, February 25, 2011 - link
Since there's PCIEx4 coming out of Thunderbolt, is it possible to have a separate discrete graphics card running outside of the laptop? That could be interesting...Ryan Smith - Friday, February 25, 2011 - link
On paper it's definitely possible. And this would help explain why we didn't see any external PCIe setups this year at CES. However there are some caveats.1) Intel hasn't told us what the latency is like. A 3 meter cable isn't very long, but PCIe is a very fast connection - a single lane runs at 5GHz. Without knowing much about the link layer, I can note that at 5GHz, even light only travels 6cm/Hz. For the copper implementation of Thunderbolt this would be worse. A reasonable guess would be that you're looking at up to an additional 50 clocks, on top of the latency between the Thunderbolt controller and the CPU.
The point of this being that I'm not sure what the tolerance is for PCIe latency with today's GPUs. It may not work out of the box if the GPU expects a response within X cycles.
2) Performance will be impacted. PCIe x2 is 8GB/sec, so a single Thunderbolt channel is only a bit more. You can run a GPU on such a setup, but you will be severely bandwidth limited, and this of course gets worse the faster the GPU. As a result performance would not be comparable to desktop parts. For something like a Radeon 5770 (I throw this out as it's a reasonable upgrade for most notebook GPUs), I'd guess it would achieve 75% of the performance, but that's very much a shot in the dark.
So in short, yes, it should be possible. If not with this generation of GPUs then with the next one. But with a copper connector it's going to be bandwidth limited.
DanNeely - Friday, February 25, 2011 - link
You have 2 channels though, which gives you PCIe 4x bandwidth, testing has found that in most games a 4x link only results in a few percent of penalty. There're exceptions like MS Flight Simulator X that take major hits but very few other games do.Ryan Smith - Friday, February 25, 2011 - link
The fact that there are 2 channels is correct. However at this point we have no idea if the channels can be teamed, hence my numbers assume they cannot.Shadowmaster625 - Friday, February 25, 2011 - link
tinyurl / 4p9pk4dThis article shows on average the difference between pciE x16 and x4 is only 10%, for a GTX480. Who knows what a meter of cable would add.
Calin - Friday, February 25, 2011 - link
While bandwidth might not be impaired by such a link, latency might very well be. And the result of increased latency might be much more visible than a 10% loss of framerate (or there might be no issues at all). And this might even change from an application to another.DanNeely - Friday, February 25, 2011 - link
I wonder why Toms stopped testing below 4x. While not as realistic a test case, it covered people using desktop GPUs on laptops (via a breakout box and a 1x PCIe ribbon cable) and showed worst case performance.OTOH IIRC that test showed a much larger 4x penalty than had previously been seen with on the 280 or 8800 tests.
AnnihilatorX - Friday, February 25, 2011 - link
6cm/Hz is incorrect. The unit is inconsistent, speed is meters per second, and Hz is per second so 6cm/Hz is 6cm-second which is a distance*time unit.I think you mean
In 1cycle of 5Ghz, there is 0.0000002 (2*10^-7) seconds
In such a time, light travels (2*10^-7)*(3*10^8) = 60 meters
So 60 meters / clock
V864 - Friday, February 25, 2011 - link
1 cycle @ 5GHz would be more like 2*10^-9 seconds giving you .6 meters, i.e. 6 cm.For quick mental checks think about how large an antenna would have to be to pick up a 60 meter wave with any efficiency. An even easier thing to remember is that the wavelength of a 1 GHz wave is about a foot.
thornburg - Friday, February 25, 2011 - link
While I believe you're right about it being 6cm (as Google agrees), 6cm is .06m, not .6m. .6m is 6dm (decimeters), a measure which is hardly ever used, as meters and centimeters cover most human uses well enough without need for a unit in between.vol7ron - Friday, February 25, 2011 - link
or... .6m = 60cm, a measure which is commonly used :)PlasmaBomb - Saturday, February 26, 2011 - link
/face palm6m = 600cm
Dicer - Sunday, February 27, 2011 - link
Can't let you get away with that one....he said .6m = 60cm
/face what?
Rick83 - Friday, February 25, 2011 - link
pretty sure it's 6 cm per clock, as 5GHz = 5x10^9/s so 1/5Ghz = 0,0000000002 = 2*10^-10You used 5MHz, which is 3 orders of magnitude off.
PlasmaBomb - Saturday, February 26, 2011 - link
2*10^-9 = 1 order of magnitude off 2*10^10I think this thread shows people need to check their figures before posting...
PlasmaBomb - Saturday, February 26, 2011 - link
2*10^-9 = 1 order of magnitude off 2*10^-10Damn...
This Guy - Sunday, February 27, 2011 - link
Um, y'all wrongc = 2.8 * 10^8 m/s (speed of light)
5GHz = 5 * 2^30 s^-1
.: c / 5GHz = 0.052 m (use brackets)
.: WAVE LENGTH (i.e. frequency inverted) is ~5 centermeters.
Read mckirkus to see why this is irrelevant.
ssj4Gogeta - Sunday, February 27, 2011 - link
But how is the wavelength relevant to latency?Narcofis - Monday, March 7, 2011 - link
Also in copper you can't forget that the speed of the electromagnetic wave is about .66 the speed of light.http://en.wikipedia.org/wiki/Speed_of_electricity
bupkus - Friday, February 25, 2011 - link
GEEK FIGHT!!!mckirkus - Friday, February 25, 2011 - link
"2) Performance will be impacted. PCIe x2 is 8GB/sec, so a single Thunderbolt channel is only a bit more. "Arrgh. 1 Gigabyte(GB) = 8 Gigabits (Gb)
You're confusing people because Thunderbolt is only 10Gb/s, not 10GB/s.
A PCI Express 2.0 x16 Link is 64Gb/s, that's the slot commonly used for video cards and is over six times faster than Thunderbolt in it's current incarnation.
When Thunderbolt is optical and supports 100Gb/s then we can start to look at external video cards.
DanNeely - Friday, February 25, 2011 - link
That's just a capitalization issue, he didn't use the wrong units in his calculation. Sheesh.An 8Gb/sec data link can fit comfortably in a 10Gb/sec channel.
Ryan Smith - Friday, February 25, 2011 - link
"When Thunderbolt is optical and supports 100Gb/s then we can start to look at external video cards."That's probably the key point. The issue at the moment is when 100Gb/sec Thunderbolt will arrive. PCIe 3.0 arrives within a year, and it will push PCIe x16 to 128Gb/sec, at which point Thunderbolt would again be slower, although not by nearly as much as before.
And yes, I screwed up the units, thanks for catching that. This is what happens when you write a comment at 1 in the morning.
Bluestealth - Saturday, February 26, 2011 - link
Why does carrier frequency have anything to do with signal propagation velocity? Wait it doesn't. I would suspect that the protocol overhead is far greater factor than the velocity of light in a conductor, which at these length scales is going to be less than 20 ns. Hardly a delay worth mentioning.Rick83 - Sunday, February 27, 2011 - link
It's still on the level of several GPU/CPU cycles - so if you need to actually interactively communicate between the two, both sides will lose massive amounts of their potential performance.Of course, if you need massive intercommunication for your computation you're lost in any case, it's only that lately the physical limitations are becoming evident where previously there were merely implementation problems (we could probably link up a huge amount of 386's today without any loss of performance - the same was not possible in their era)
So if you're only streaming, signal propagation isn't a problem. It's only in latency bound scenarios where insanely high speed individual components will not be able to run at their full potential. Along this self same axis, we are probably already at )or quickly approahing) the hard RAM-latency wall, until we move that too onto the CPU...
softdrinkviking - Friday, February 25, 2011 - link
No, I think it was originally supposed to support everything, but they cut it back probably for cost reasons. I'm sure interpreting more protocols increases the complexity of the controller, ect.Also, the change to copper may have had an influence on that decision.
Or maybe it's an intel/apple conspiracy to eliminate USB.
nafhan - Friday, February 25, 2011 - link
It sounds like something along the lines of a PCIe USB controller + Thunderbolt controller would work. Where this will really come in handy is having things like a USB3 controller built into a computer monitor. That would eliminate at least one cable from the monitor the PC.pugster - Friday, February 25, 2011 - link
Thunderbolt technology is like Firewire on Steroids with the same problems plaguing it. Too complicated and need to connect to display adapter. Not to mention royalty fees. USB is a cheaper, simpler technology and it works.Spazweasel - Friday, February 25, 2011 - link
"Thunderbolt technology is like Firewire on Steroids with the same problems plaguing it. ... Not to mention royalty fees. USB is a cheaper, simpler technology and it works"Wrong.
Directly from the article: "Similarly, there’s no per-port licensing fee or royalty for peripheral manufacturers to use the port or the Thunderbolt controller."
hescominsoon - Friday, February 25, 2011 - link
usb 3 is dead..at least for Intel. Since Intel is the ONLY validator for USB 3.0 Intel isn't going to bother with usb 3 anymore as USB 3 doesn't reap the billions thunderbolt is going to reap. Every one of those little ic's that's needed on each end of the cable..Intel gets paid for those. It's not an open tech..so be expecting others to have issues(linux...other oses).Randomoneh - Friday, February 25, 2011 - link
Aren't HDDs and SSDs still a bottleneck with max. speeds of 700-1000 MB/s?dagamer34 - Friday, February 25, 2011 - link
Connector should be faster than the connected devices, so that daisy chaining more sense.Azethoth - Friday, February 25, 2011 - link
Single ones maybe, but arrays saturate current links eventually.ltcommanderdata - Friday, February 25, 2011 - link
"The connection supports a daisy chain topology, and Thunderbolt also supports power over the cable, though it isn’t clear how much. I’d expect no less than the 5 or so watts that USB 3.0 supports."http://www.apple.com/ca/thunderbolt/
"Thunderbolt also provides 10 watts of power to peripherals, so you can tackle workstation-class projects on the go."
At least as implemented by Apple, Thunderbolt supplies 10W of power, which I believe is less than Firewire but more than USB 3.0.
And is the Thunderbolt controller really that big? I guess we won't be seeing Thunderbolt flash thumb drives anytime soon. And I can't think Apple could be happy with this since putting that controller in the iPad PCB will already be difficult much less the iPhone or iPod Touch. Thunderbolt would seem to be ideal to sync 64GB or probably upcoming 128GB iDevices.
Brian Klug - Friday, February 25, 2011 - link
It's indeed 10 watts, Ryan also caught me on this ;)Should be fixed now.
-Brian
Azethoth - Friday, February 25, 2011 - link
"And is the Thunderbolt controller really that big?"From pictures on the Intel site it looks like the optical one is the size of a phone jack but much flatter. Sexy!
Yeah, that Thunderbolt controller was quite the porcine wonder!
Fleeb - Friday, February 25, 2011 - link
Guess this is the answer.gevorg - Friday, February 25, 2011 - link
I wonder how much Apple paid Intel for exclusivity to be the first to have Thunderbolt?Chloiber - Friday, February 25, 2011 - link
Intel says every manufacturer could release TB today if they wanted to. It's not exclusive to Apple.gevorg - Friday, February 25, 2011 - link
Then why Intel didn't include it with any of their own H67/P67 motherboards?puttersonsale - Friday, February 25, 2011 - link
planned obselescence...same idea that its not profitable for any company to make 50 year lightbulbs....unless you charge buttloads of money for them...then who would buy them? Intel wants you to buy their next boards too. heherelentlessfocus - Friday, February 25, 2011 - link
Or did Intel pay Apple for the use of the part of Light Peak that depends on Apple's display port technology by getting Sandy Bridge processors to Apple so quickly.JMS3072 - Friday, February 25, 2011 - link
DisplayPort is not "Apple's...technology". It's a fully spec'd out standard on its own, and is actually a breath of fresh air after using TMDS-based connectors for a LONG time. As for the port, Mini-DisplayPort has been part of the DisplayPort standard since 2009. Intel didn't need to pay Apple to use it at all.erple2 - Friday, February 25, 2011 - link
in fact, that's why Apple uses it. Apple doesn't have to pay anybody anything to use a DisplayPort adapter, unlike an HDMI adapter. There's no licensing costs on DisplayPort, unlike HDMI.B3an - Friday, February 25, 2011 - link
My Dell monitor and my AMD 6970's all use DisplayPort and Mini DisplayPort. Many things use it.It's not an Apple tech in way, they have nothing to do with it.
gt1911 - Friday, February 25, 2011 - link
Do I understand the article correctly that we now avoid spending big money on GbE adapters and just link two computers with Thunderbolt and run it as an ethernet connection?If so, I'm a big fan, particularly if we could do this through a hub. Any ideas on the maximum cable length?
numbertheo - Friday, February 25, 2011 - link
You can run networking over Firewire. Judging from the popularity of that feature, I don't think that networking over Thunderbolt would be a very popular use case.Telvin - Saturday, February 26, 2011 - link
Networking over firewire is quite popular for diagnostic and data transfer scenarios. For example, if you firewire two macs together, you can boot either of them off of the other.For casually tossing large files around Thunderbolt sounds more attractive than many other ad-hock network solutions I've used.
flyingpants1 - Friday, February 25, 2011 - link
I think Gigabit ethernet has come standard with most computer and laptop motherboards made in the last several years. What do you mean, "spend big money on GbE adapters"? Or did you mean 10GbE?bunnyfubbles - Friday, February 25, 2011 - link
I'm pretty sure he meant 10GbE.But either way, it sounds like this won't replace networking given its disadvantage with any sort of appreciable networking distances.
glugglug - Friday, February 25, 2011 - link
I also don't see why you would use this over a **much cheaper** 10GbE ethernet cable. For the distances its capable of you could go with much cheaper SATA3 for 6Gb/s, or probably SATA4 by the time this becomes available on non-Apple stuff.gt1911 - Friday, February 25, 2011 - link
Yep, Sorry, stuff up in my post. I meant 10GbEAzethoth - Friday, February 25, 2011 - link
My guess for the wire based lengths would be that you could get cat7 performance which in theory means max 40GbE at 50 meters. Much less in practice of course. Lets assume only 2x10GbE at 10 meters or 30' based on currently shipping speeds. Depending on cost and the latency constraints mentioned earlier the lengths may be specced even shorter. Chaining extends the total length.Optical of course should add a nice multiple to the lengths.
I hope we do not waste too many sputnik-broadband $ on thunderbolt. I want the optical goodness!
DanNeely - Friday, February 25, 2011 - link
Assuming you mean 10GBe, it can run upto 100m on quality copper cable before needing a signal booster, thunderbolt's limited to 3m on copper.gt1911 - Friday, February 25, 2011 - link
Thanks guyssoftdrinkviking - Friday, February 25, 2011 - link
Will the optical implementation of thunderbolt continue to supply power?relentlessfocus - Friday, February 25, 2011 - link
No, sorry don't have a link for you but I read this in several places.Shadowself - Friday, February 25, 2011 - link
While the current implementation is copper based having copper wires for both data and power, the future implementations will have fiber for data and copper for power. The power lines will not be eliminated from the fiber implementation. The cables and connectors will be a combination of fiber and copper.The only reasons for the fiber implementation are increased distance and increased speed for the data. Intel has claimed that under the fiber implementation they expect to have 100 Gbps per channel by 2020.
The power supplied is DC power. Within reason, this is independent of the data and the distance of the cables. Therefore no significant change to the power lines will be required when it goes to fiber for the data side of the link.
Guspaz - Friday, February 25, 2011 - link
"The power supplied is DC power. Within reason, this is independent of the data and the distance of the cables. Therefore no significant change to the power lines will be required when it goes to fiber for the data side of the link."The problem of delivering low-voltage power over 3m and over 100m are completely different. PoE has to do the same thing (deliver DC power over 100m of cabling). In order to do so, it requires a relatively high voltage. I believe that PoE puts 57v into the cable, and expects to get at least 37v at the other end after resistance loss is accounted for. If they go for the same 5v as USB, they're going to have major issues at 100m.
flyingpants1 - Friday, February 25, 2011 - link
"Tunnel a PCIe lane over the link, and you can dump it out on a peripheral and use a local SATA, FireWire, USB, or Gigabit ethernet controller to do the heavy lifting. Essentially any PCI Express controller can be combined with the Thunderbolt controller to act like an adapter."So can thunderbolt transport USB and SATA or not? Would it be possible to design an external hard drive with a Thunderbolt interface, and have it implement SATA? Or USB 3.0? Or how is this supposed to work exactly?
I want 12 Thunderbolt ports and no USB/eSATA/HDMI/DVI/Displayport.
puttersonsale - Friday, February 25, 2011 - link
that is a nifty idea....would be nice to have 1 cable type for all....i bet intel is thinking bout it. they want your $$$softdrinkviking - Friday, February 25, 2011 - link
you can transport USB 3 or whatever else you want over thunderbolt if you "tunnel a PCIe link" over the link.meaning you have to use a PCIe adapter to send the data as PCIe protocol, so thunderbolt can transport it.
then, on the other side, the signal gets dropped into another PCIe adapter and either converted into something else, like USB 3, or maybe SATA, or just sent as PCIe to wherever.
it was originally going to carry ALL of the various protocols that are around today, and then drop data off at whatever adapter was waiting for it with no middle-men, but intel changed that for some reason. i think it must be cost-related, but i'm jaded and distrustful.
Poulsen8r - Friday, February 25, 2011 - link
... at 2 metres. DP 1.2 is 17.28Gbps at the same distance. I love the idea but i think they need more bandwidth to support multi-monitor, high res/refresh setups since if you are sending data and video over the same wire it may become bottlenecked faster than you think. I also don't get why there is a 10Gbps interface when the fastest internal drives are capped at 6Gbps and therefore an eSATA drive on a SATA 3 controller will send data (obviously not video) at the same rate as a 'Thunderbolt' port.As many people are asking though, it will support any other interface you can put the controller on, so you can have Thunderbolt to USB adapters or eSATA or Graphics or RAID cards or Network or whatever you can plug into a normal PCIe slot.
Seems like today 4 more of my PCIe lanes are wasted on a not needed (at this time) technology. As for the future when it moves to fibre and we ditch the crappy SATA interface for internal storage, I think I'll be sold then.
It may be cool that I can plug any usb device into any version of the port and it will work, but backward compatibility slows down forward advancement. Everyone should realise this due to the sandy bridge woes that were caused by a transistor used for some type of legacy operation. New port = more win, sooner.
phatboye - Friday, February 25, 2011 - link
My biggest question is about the licensing of this proprietary link? Will Intel license this technology out to other integrated circuit makers like NEC, AMD, TI, NVIDIA etc? And at what cost will they license this out for? If Intel will be the only company producing Thunderbolt chips then I am not interested at all because you will not see widespread adoption of this interface like we see with USB.SunLord - Friday, February 25, 2011 - link
I have no doubt Intel will license out the controller but I doubt many will use it to integrate it into anything for now like usb 3.0 but worse. That chip is massive and is likely a 45 or 32nm part so even at 28nm it would be kinda large to put on die with a chipset or soc. It would be interesting to put it on a video card though since they could use it to output all the video signals and it kinda would make sense with PCIe 3.0 just around the corner you'd have alot of unused bandiwth on a 16x port to abusesoftdrinkviking - Friday, February 25, 2011 - link
it's supposed to be free, that's it's big deal selling point.intel's making the loot off of selling the controllers and the port hardware.
phatboye - Friday, February 25, 2011 - link
yes it's free to use the port but I am talking about licensing the protocol so that other IC manufactures can make controllers.iwod - Friday, February 25, 2011 - link
IT is actually very UNlikely a 45 or 32nm Chip. It would properly be at best 65nm or 90nm Controller.Anato - Friday, February 25, 2011 - link
Not supporting other protocols is a mistake. Instead of supporting motherboard's quality controller in drivers there is need to support myriad of controllers with different quality and drivers. We will definitely see Thunderbold -> USB3 -> SATA combination for HDD's and thats bad.softdrinkviking - Friday, February 25, 2011 - link
I just read over on pc perspectives that this may not be true.It's possible that this particular implementation of lightpeak (thunderbolt) is an apple only thing, and that there could be others which use different adaptors and support other protocols.
This info is in the list of updates they made at the end of the article.
iwod - Friday, February 25, 2011 - link
WHAT? You dont. You get a ThunderBolt > SATA since most SATA chipset support PCI-E natively.lowlymarine - Friday, February 25, 2011 - link
It's all very fancy, but I am 99.99% certain this will see the same fate as FireWire and e-SATA: Technically superior but relegated primarily to large-scale external storage for enthusiast/professional use. Though really, even that is in doubt since USB3 will never bottleneck mechanical storage devices. USB3 is "fast enough," has years of head start, has no associated licensing fees, and most importantly is fully backwards-compatible with the peripherals and cables everyone already owns. Thunderbolt is commercially licensed, currently only available in a handful of very overpriced computers, and is only backwards-compatible with a port that no one but select enthusaists with Eyefinity6 setups and die-hard Mac fans uses anyways.Inertia is an extermely powerful force in consumer electronics. Even Apple's legendary buzzword marketing couldn't get any consumer traction for DisplayPort, Mini-DVI, or Firewire. I really don't see how this will turn out differently.
Mike1111 - Friday, February 25, 2011 - link
Personally, I'm looking most forward to docking stations that one day work with nearly every notebook. Just one cable from my notebook to the docking station and there you have e.g. USB ports plus eSATA, Gigabit Ethernet and DisplayPort etc.Mike1111 - Friday, February 25, 2011 - link
Wow, that controller is huge. No way Apple is putting something this big in an iDevice this year.And I would add one thing to your Thunderbolt article: That it can support multiple topologies, not just daisy chains (according to the Intel PDF).
"A symmetric architecture that supports flexible topologies (star, tree, daisy chaining, etc.) and enables peer-to-peer communication (via software) between devices."
Shadowself - Friday, February 25, 2011 - link
Yes, this is something most people miss -- the variable topologies allowed.I'd bet, in the final analysis, it is like Firewire in the possible topologies allowed. The only topology not allowed in Firewire is a ring of any kind (e.g., you can't create a tree/star combination that can result in a ring).
Penti - Friday, February 25, 2011 - link
Btw, copying the frame-buffer over to the Intel iGFX or using it's connection means there is no DisplayPort 1.2 support right? I.e. no daisy-chaining. Or does it use the discrete gpus DP for that?Since HM67 is PCI-e 2.0 compatible shouldn't it be x2 PCI-e connected to the chip? The Promise Thunderbolt device uses that anyways. It's the cpus x16 lanes PCIe 2.0 plus the chipsets x8 lanes PCIe 2.0. I kinda see it as a successor or competitor to ExpressCard 2.0, twice as fast and not as annoyingly space consuming. A more universal and consumer as well as prosumer friendly connection. Macs will get Native USB 3.0 with Ivy Bridge and the 7-series chipets, until then you could create a Thunderbolt > PCI-e 2.0 x1 NEC USB 3.0 controller converter. I still think it's useful for slightly different uses though, you won't see self powered Thunderbolt harddrives by any means. You already have self powered USB 3.0 devices. I think it would be quite hard with a drive and SATA-controller powered by the mDP cable. Definitively won't happen to memory sticks at least :) Although maybe 10 W is enough. It just won't fit in memory sticks though :) I guess they will move to a pci-express interface though.
Shadowmaster625 - Friday, February 25, 2011 - link
So after integrating the gpu onto the cpu and in theory reducing the chip count by one, they then turn around and add another IC to the motherboard. lolStormyParis - Friday, February 25, 2011 - link
Well, if you don't understand... that IC can replace most I/O ICs currently used ?KitFR - Friday, February 25, 2011 - link
Does the fact that the cabling is copper-based mean that five years down the line I will find myself with a draw full of TB wires yet somehow never with the one I need, be it copper-to-copper, optical-to-optical, or copper-to-optical? If so, I would have rather that Intel held off another six months for a pure optical solution.Shadowself - Friday, February 25, 2011 - link
Not meaning to be rude, but where have you been for the past 50 years in the computing technology world?You ALWAYS end up with drawers and boxes of old cables that are not the exact kind you need. It's just life.
MarshallG - Friday, February 25, 2011 - link
From the description here, it seems like the DisplayPort pins and the Thunderbolt pins are physically separate; they simply added pins to the DisplayPort connector/cable for Thunderbolt. This would explain why the connector is backwards-compatible with DisplayPort. It seems as though Apple and Intel are concealing this fact with marketing lingo, but it's obvious to me, at least, that the reports of Thunderbolt combining DisplayPort in its protocol are false.At any rate, I think most readers here are missing the key benefit -- it will be very easy to dock your MacBook. Because this is all based on DisplayPort and PCIe, there are many existing controller chips which could be put in a dock or inside a monitor -- SATA, second/faster graphics board, Gig Ethernet, and more USB ports. And this can probably be done with existing chips, or, at most, with only slight tweaks. I don't see this as a new protocol or standard; it's really PCIe in a cable, but that's a good thing.
thewhat - Friday, February 25, 2011 - link
USB is here to stay. Unless the plan is to give people Thunderbolt to USB converters (bad plan). USB devices are simply too widespread to be ignored. USB 3 is simply an evolutionary step we'd be silly not to take. There's even plenty of USB 3 storage out there already.OTOH, I think Thunderbolt offers some distinct advantages over USB and therefore could find its use for some high end applications.
The only problem I have with it is that it's too slow. Think of how awesome it would be to have powerful external GPUs etc.
So I think USB 3 and Thunderbolt will complement each other. Hopefully we'll now get rid of eSata, Firewire and perhaps even various video (and network) connections.
MarshallG - Friday, February 25, 2011 - link
There are no USB 3 drives. They are SATA drives with a SATA-USB 3 bridge. With this technology, you could attach a USB 3 or SATA drive, by having the right host controller on the other side of the cable. The cheapest and fastest way, I think, would be to put a SATA RAID chip on the other side of the cable.muzicman82 - Friday, February 25, 2011 - link
This looks very promising and I tend to trust it considering it was developed by Intel. I imagine at some point Intel will need to break away from Apple on development so that some things may be standardized.I understand that this is just a PCIe and DisplayPort link. However, for all of the various uses and protocols, it sounds like chipsets and controllers are going to need to be embedded in devices or cables for all of that backwards compatibility with Firewire, USB, etc that they claim.
It also seems like this will be quite expensive for some time.
I also think the technology needs to be made available on PCs for the industry to catch on. Or, even ExpressCard to Thunderbolt adapters for PC users may be adequate.
My hesitation right now is that 10Gbps is a burst theoretical claim. Unless you are connecting to a very fast RAID array that is truly 10Gbps internally, you're not going to see those kinds of speeds. Hard drives or SSDs are no where near that fast yet... so don't go out and buy that new Western Digital external drive that only has one or two drives in it. It's not going to be that fast.
I don't see a whole lot of benefit from having DisplayPort in the connector. Perhaps if you could put GPU's inside of displays, you might be onto something. The display's GPU could then send accelerated graphics back to your laptop LCD. Other than that, it seems like you would be paying for it to be there even if you didn't need that aspect of it. Maybe we'll see PCIe only Thunderbolt cables. I don't see a lot of benefit from daisy chaining devices either. It was there in Firewire and I never used that feature.
I would like to see an interface that pulls double duty and connects the same way externally as it does internally, providing both power and data.
glugglug - Friday, February 25, 2011 - link
In 8ns, light only travels 2.4m. Through copper it will go even less. So either the cable needs to be shorter than that (maybe 3 feet instead of 3m?) or the 8ns figure is wrong. Actually the 8ns is crazy anyway -- why would your external peripherals need lower latency than your RAM?jwilliams4200 - Friday, February 25, 2011 - link
Excellent point. And for some applications, the latency would be limited by a round-trip along the cable. Group velocity is likely only around 1x10^8 m/s. That would give a maximum cable length of 40cm (about 16 inches) for 8ns round-trip latency.darckhart - Friday, February 25, 2011 - link
1. didn't see in the article if it used one of the pcie lanes to do its deal. does it?2. it seems the apple implementation is severely neutered. why bother introducing it at all at this point in time? what do they expect it to do? link to (small) external storage? why bother? nothing really needs it currently. big external storage? you probably have it setup through the network on 10GbE. link to display? uh isn't that what the vga (for projectors or monitors) and hdmi (for tvs) currently does just fine?
Mech0z - Friday, February 25, 2011 - link
For people that think you need 16 PCI-E 2.0 lanes then your very wrong! Look at http://www.techpowerup.com/reviews/AMD/HD_5870_PCI...A HD5870 on x16 Pci-E 2.0 gets 24600 points in 3dmark 06 while it gets 23800 on x1 !
On average x1 gives 75% of the performance of x16 on pci-e 2.0
I use a Vidock and run pci-e 1.0 x1 (Only sandy bridge laptops have pci-e 2.0) and there you get 15000 in 3dmark06
Mech0z - Friday, February 25, 2011 - link
WhoopsI use a DIY Vidock http://forum.notebookreview.com/gaming-software-gr...
Penti - Wednesday, March 2, 2011 - link
Yes of course, and it's not always you need it for gaming.But Thunderbolt isn't a replacement for real external PCIe interfaces for graphics. Having external PCIe 2.0 x2 devices for storage is great though. In servers there might be some merit to x8 cards but here 2 PCI-e 2.0 lanes should be more then fine. By a lot. It's enough for a high-end video work-flow any way. Definitively a more professional setup then to go with NEC or whatever USB 3.0 chip. Which there is still no OS X software stack for any way. From Apple. Lacie has release a third party driver though. For those that can extend their system using ExpressCard or future Thunderbolt > USB 3 converters. (Or by PCIe cards).
FXi - Friday, February 25, 2011 - link
If this was coming on motherboards a month or two from now, I could understand the hoopla. But this won't be in laptops you can BUY for a year, quite likely a bit more than a year. So why the hoopla?It's a delaying tactic. USB 3 is here. It works. And yes, it could transmit video if you so chose, and you wouldn't notice the diffrence between 5Gb/s and 10Gb/s. Quite likely the speed of Intel's solution is sensitive to cable quality and length, so I'm not even sure calling it a 10Gb/s connection is really reasonable in most consumer level solutions and situations. But that's fine.
Intel, come see us in a year or so when you have a product to SELL. And it'll be a year (2 years from now mind you) before you'll see this interface widespread in actual devices.
So nice technology demonstration. The product will be interesting when it gets here, but it is largely irrelevant to this year let alone next.
softdrinkviking - Friday, February 25, 2011 - link
i suggests you read the coverage of the new macbook pro line-up!AnnonymousCoward - Monday, February 28, 2011 - link
Yeah USB3 streams Blu-Ray just fine and it's very hard to saturate its 400MB/s throughput. Matching a displayport + x4 PCIe makes little sense to me.ninjit - Friday, February 25, 2011 - link
In the article you state..."Thunderbolt is dual-channel, with each channel supporting 10 Gbps of bidirectional bandwidth. That’s a potential 20 Gbps of upstream and 20 Gbps of downstream bandwidth"
I don't believe that's correct.
I think it means you can have 10Gbps streams going up and down simultaneously.
There's no mention of 20 Gbps on the intel or apple pages about the tech - if it was capable of pushing that many bits one-way, you think they would be marketing it as such.
This is what intel says about the throughput:
"Dual-channel 10 Gbps per port, Bi-directional"
and Apple states:
"Thunderbolt I/O technology gives you two channels on the same connector with 10 Gbps of throughput in both directions."
To me this means there are 2 channels of 10 Gbps each, one up and one down.
regli - Saturday, February 26, 2011 - link
I believe that he is correct. Intel states the following:A Thunderbolt connector is capable of providing two full-duplex
channels. Each channel provides bi-directional 10 Gbps of bandwidth.
This adds up to 20 Gb/s each way. Based on Stephen Foskett's input
http://blog.fosketts.net/2011/02/24/thunderbolt-li...
you can even add "plus up to two DisplayPort v1.1a connections with 8.64 Gb/s each".
iwod - Saturday, February 26, 2011 - link
Since Thunderbolt provide enough power for both SSD, HDD and DVDs, wouldn't it be better to used Thunderbolt internally as well?Once PCI - Express 3.0 lands and we have enough bandwidth head room, this is truely one port to rule them all.
jcandle - Monday, February 28, 2011 - link
That really isn't the issue. If you look at Intel's datasheet you see that devices use their native PCIe drivers. That means each device would essentially be a PCIe device. Thunderbolt essentially wraps the PCIe signal making it transparent to the end device. The device simply thinks its directly connected to the PCIe bus, just like an internal PCIe add-on card is connected today. Thus, this would be similar to the existing PCIe SSD units already being sold. And would likely require drivers for every manufacturer's controller unless already bundled with the OS. This also inflates the cost of each component. The internal SATA controller on your chipset provides multiple port defraying the cost of implementing a device. Having an internal DVD on thunderbolt serves no purpose than to inflate its cost and implementation. That said, future IO devices after current optical drives have been phased out, may use a more evolved version of thunderbolt to internally connect.Thunderbolt is most useful in external situations because computers are shrinking for the large workstation towers of old to new slimmer form factors. Thunderbolt makes it practical for machines with less internal space to host a variety of high performance peripherals and add-ons.
newonanand - Monday, February 28, 2011 - link
if thunderbolt can run as networking and it has native driver supports, it will definitely benefits server products for small business/ home business.Think about setup like Mac mini Server -> Promise RAID -> Mac mini Server running Parallel Server for Mac.
The point is the Mac comes with ThunderBolt FREE. 10Gb Ethernet is still VERY EXPENSIVE for now.
jcandle - Monday, February 28, 2011 - link
Direct connection is likely out of the question, but it could be proxied through thunderbolt to thunderbolt transfer cable; similar to the usb to usb transfer cables already available. However, its cost would be similar to the existing usb solution and thus make it costly to link multiple computers together; costly, in comparison to existing GbE.Still I agree it would be cheaper than current 10GbE. However, if for no other reason that simple demand-- if consumers desire greater bandwidth than current GbE it would drive down the cost of 10GbE. Currently the majority of 10GbE cost lies in its switches; "dumb" switches would need to be developed to fill this need.
There is one note about your example. You have two computers sharing one RAID. That doesn't work. The same issue already occurs with enterprise customers with FC RAIDs that want to share it between two workstations. You need a special file system and locks to prevent each computer from corrupting data on the device. Or you have to set up two separate "partitions" thus making two separate RAIDs. The simplest solution for consumers is to have the RAID be a special NAS with each computer having a driver to access the NAS in a network like fashion. I don't expect such a solution to come cheaply.
Nentor - Monday, February 28, 2011 - link
What do they want to imply with it? Speed? Thunderbolts sure are fast, but also destructive, loud, bright, zigzaggy.It is like calling a small round candy "choke" to me.
Kakureru - Monday, February 28, 2011 - link
price = adoption leveli wont be buying bridge devices that hold a very high cost for a very long time.
first seeing them in apple products is bad since this adds a perceived value to
it which will make other companies tie an un necessarily high price to its use
hoping that they can ride apple's coat tails and make higher profits.
fortunately for me USB 3 dropped in price dramatically since inception..
I can buy a 2 port host controller for 20$ and a SATA 2 controller for 15$ (and falling)
if "thunderbolt" can carry these numbers, Ill be all aboard and be waiting.
(ppshopping) - Wednesday, March 2, 2011 - link
welcomejb510 - Saturday, March 5, 2011 - link
I understand the beauty of this solution to be that manufacturers can essentially take any existing PCIe card out of desktop (a USB3 PCIe card for example), stick a Thunder Bolt chip on it (PCIe to TB bridge) and TB connector on it and boom... you have an external USB 3 hub.. or external whatever hub you want. Now it might be even simplier to bypass USB3 and just go straight from SATA to TB, but my point is creating a external hub for any connection that already exists in PCIe card format should be nearly trivial, no?I'm not entirely clear however on the evolution of LightPeak into Thunder Bolt. Was LightPeak always running over PCIe, or is that something entirely new?
lili53 - Thursday, March 10, 2011 - link
welcomelili94 - Wednesday, March 23, 2011 - link
welcome