Given the implementation on the new Haswell boards it feels as tho thunderbolt is dead. There's no business case for it right now as you'd be insane to bring out a lot of TB based stuff as outside of Apple there's no support.
Agreed. Until it gets to the point of being really, really fast (PCI Express x16-like) to the point where external graphics cards in a docking station are feasible at comparable speeds, I don't see the point.
Seems to me, external graphics card connections would need not just high bandwidth but low latency; I'd almost say that past a certain practical minimum, low latency becomes much more important than higher bandwidth. It's analogous to the characteristics of your internet connection for online gaming: once you get past a certain minimum bandwidth threshold, what starts to matter most is not more bandwidth but lower ping times (i.e. lower 'lag').
Unless an entire rendering pipeline including all data and logic can fit into the GPU's on-board memory (and execute purely on the GPU), any communication with CPU and load/stores to DRAM would probably be significantly slower over Thunderbolt than over plain PCI Express (due to extra encoding/decoding steps, and extra length of cable.)
I've seen somewhere that Intel plans to eventually move Thunderbolt to a fiber optic implementation, which would help with transmission latency over the cable (and allow for longer cables), but will probably add extra latency due to optical transceivers... so in the end might be a wash.
If Apple keeps paying them for the controllers, Intel will keep making and developing the chips. Eventually it will be cheap enough to add in to be relevant in the low-margin PC space.
I completely agree. Once Apple implements Thunderbolt into it's Mac Pro and the prices of the cables / controllers continue to become cheaper, which is already happening, TB will become THE choice when it comes to top-of-the-line Interconnect where getting the fastest and reliable speed is key.
The analogy to Firewire is spot on in nearly every respect.
Formats without enough technical superiority to warrant their tremendous price premium. That few device makers ever adopt due to pricing. Which is nonetheless championed by a small subset of Mac users, while the rest of the world adopt much cheaper rival that is more than good enough for most.
Actually Firewire is/was Apple's IP. They charge a licensing fee for every firewire port regardless of whether it is on a pc or mac. TB is Intel's IP. So it probably has a lower fee (if any). I expect Intel is happy just selling the chips and would rather see wider adoption long term, than short term profits.
Actually, FireWire is a trademark held by Apple and freely licensed to third party developers of IEEE 1394 enabled devices who comply with certain guidelines. IEEE 1394 is a serial bus interface standard maintained by the IEEE P1394 Working Group based on an open host controller interface. The standards essential patents required to implement IEEE 1394 are held by 10 corporations including Apple, which was one of the lead designers of the technology, and are licensed as a package by MPEG LA at a rate of 25¢ per finished product, regardless of the number of physical ports.
Thunderbolt is a proprietary Intel technology developed at the behest of Apple, which continues to hold certain standard essential patents related to the technology. The Thunderbolt trademark, although originally registered by Apple was transferred to Intel, and licensing and certification are dictated by Intel. This is one of the most glaring differences between FireWire and Thunderbolt. While FireWire may not be an obviously popular consumer interface, it's actually rather ubiquitous due to its open nature and reasonable licensing arrangement. The closed nature of Thunderbolt and Intel's refusal to allow it to become the interoperable standard that Apple had hoped it would be may result in its undoing sooner rather than later.
It wouldn't be so expensive if Intel would put support for it in their own freakin CPUs.
Seriously. Haswell may be what I'm going to upgrade to when I retire my almost 5-year old X58 system, but it's under protest. 16 PCI lanes, no Thunderbolt, a GPU solution I'll never use and makes overclocking that much harder because Haswell is a bigger chip. Some nice architectural tuning, but not what I want in the bigger picture.
So the aggregate bandwidth is no different, right? Just individual channels are faster, but there's half as many? That's a good thing I suppose, unless you were daisy chaining multiple devices which would be better on multiple channels.
Well, actually Thunderbolt is generally 5x - 10x more expensive than USB 3.0, which would only be a single order of magnitude, and only in the extreme cases. And Thunderbolt is 2.5x - 8x faster, which makes it worthwhile to the time is money crowd.
While it would have been nice for Intel to have sorted out channel bonding and DP 1.2 support from the get go, it's hardly as if those shortcomings slowed adoption. If Intel has done too little, too late, it's in regards to making the Thunderbolt host controller interface specification available so that it could have a fighting chance of becoming a true interoperable standard. At this rate, as soon as Apple stops buying controllers, Thunderbolt goes out like a proprietary punk bitch.
It's often much more than 10 times the price, and that price has to be multiplied for each device, hub, or cable a user may need. The dearth of supported devices is also a major issue.
Combine those factors, and even those with money to burn have very few use cases that could not be handled with USB 3.0, eSATA, or DisplayPort. On paper, TB may have advantages. In the real world, the performance differences may be hardly noticeable.
I fully agree with your conclusion. Whenever Apple tires of driving down this cul de sac, Thunderbolt will suffer a quick death.
Would you care to point me to a single Thunderbolt device that costs more than 10x what a comparable USB 3.0 product does? While there are some very expensive Thunderbolt products on the market, they are often expensive for reasons other than just the interface. Comparing a Thunderbolt 10GbE interface to a generic USB Ethernet adapter is not really a direct comparison of the cost of Thunderbolt vs USB.
And the cost multiplier remains constant regardless of the number of devices and cables one buys: (n*a)+(n*b)+(n*c) = n*(a+b+c).
There isn't so much a dearth of devices as there is a dearth of devices that you personally require. DP is display only and uses a simplex main link, eSATA is storage centric and tops out at 4.8 Gbit/s or less, and SuperSpeed USB currently hits a wall at 4 Gbit/s. If your application requires more than that, you turn to Thunderbolt. There is also a fair amount of PCI/PCIe to x silicon out there for which no USB to x silicon currently exists. You could argue that most PCIe slots in desktop PCs are useless as well since the majority of users are content plugging devices into the other ports available to them.
10x is nothing. How about a 30x price premium? That's how much more you'll pay for Belkin's $300 Thunderbolt hub, or as they say "dock". Sure, it has some add-on features, but most users are never going to use them.
Want a cheaper Thunderbolt hub without the extra ports? The cheapest I've seen is $200.
Face the truth. Most users don't need the features offered. The overwhelming lack of devices and exceptional cost has made Thunderbolt a format with one foot in the grave and the other quickly following.
Ok then, show me a Thunderbolt hub for less than $200? The only hub other than the $300 Belkin seems to be a $200 LaCie. Yes, $200 to $300, for a hub.
If Thunderbolt isn't the dying format I believe it to be, prove me wrong. Why, in the 2 + years since the format's release is there such a dearth of devices? Why does a simple hub cost 20 to 30 times that of a USB hub? Why, two years later are the vast majority of Thunderbolt peripherals a handful of hard drives? Devices that would run equally as fast over far cheaper, existing standards.
Clearly, you don't want to accept the truth, but calling me a troll isn't going to change it. Thunderbolt will be lucky to ever be a 10th as popular as Firewire, and that's saying something.
If you're not trolling, then you have no idea what you're talking about. There are no "Thunderbolt hubs". What LaCie makes is a daisy-chainable Thunderbolt 2-port eSATA 3Gb/s adapter. LaCie, a company known for making high-priced gear, lists it at $199.00. The cheapest 2-port eSATA 3Gb/s adapters I could find with a quick search of Google shopping and Newegg start at $19.99, i.e. the Thunderbolt version is 10 times more expensive. This is an extreme case, and exactly the one I was considering when I said there were no Thunderbolt devices that cost more than 10x the price of a comparable alternative.
The Thunderbolt docking solutions from Belkin and Matrox are not USB hubs. They're designed so you can quickly and easily plug a single cable into a laptop and by doing so connect it to a display, Gigabit Ethernet, speakers, storage devices, keyboard, mouse, etc. There are a lot of USB 3.0 solutions, most of them based on DisplayLink silicon, that provide similar features. The cheapest one of those I could find is the Kensington sd3000v which you can pick up for $122.09. That makes the Belkin less than 2.5x as expensive, and the USB docks can't do FireWire and require real-time compression / decompression to do display output which results in artifacts and noticeable lag. Furthermore, the Belkin dock is currently one of the very few ways to add USB 3.0 capability to a 2011 Mac.
I'm not out to prove Thunderbolt isn't withering on the vine; it's a technology that has some serious issues to contend with. It just annoys me when folks such as yourself continue to make ridiculous and erroneous claims. For instance, less than a third of the Thunderbolt devices currently available have anything to do with hard drives. If you took the time to check out some of that gear, you'd see why the manufacturers chose Thunderbolt instead of far cheaper, existing standards.
You sure that math is right? 2560(x-res) * 1440(y-res) * 24(bit-depth) * 60(refresh rate) = 5,308,416,000 bits Thats roughly equal to 4.94 gigabits per display if my math isn't busted.
Yep. You forgot that CVT even with reduced blanking still requires sending more "pixels" than are actually displayed. There's blanking, front porch, sync and back porch in both the horizontal and vertical dimensions which works out to 2720 * 1481 * 24 * 60 = 5,800,780,800 or near as makes no difference 5.8 Gbit/s. There's no dividing by 1024 here either, 1 Gbit = 1,000,000,000 bits.
This still doesn't solve the cost problem that Thunderbolt faces. I remember when Thunderbolt was first introduced, I couldn't help but feel giddy because external GPUs were now within reach. 2 years later, and we are still no closer to reaching that reality.
It's honestly getting hard to get excited about a mainstream tech (well, mainstream enough to be shipping with every Macbook) that can only be utilised by deep-pocketed professionals. I'm starting to become crazy enough to think that we will see external GPUs running on 10 Gbps USB 3.0 or USB 4.0 before Thunderbolt becomes affordable for mainstream users.
A yawn is the best I can muster. When it costs an extra $50-70 per port/device, it's just not going to happen. Intel has to decide to make it ubiquitous on motherboards to even give it a chance... but it seems like they don't really care.
"Current implementations are limited to 10Gbps per channel for data or display, both can't be sent over the same channel."
This... isn't true. You can drive two 2560x1440 displays while writing 440 MB/s to a Thunderbolt RAID with everything daisy chained off of a single port. The two displays alone require 11.6 Gbit/s which is > 10 Gbit/s. Intel has also stated in the past that each direction in each channel can carry display and/or PCIe data.
You sure that math is right? 2560(x-res) * 1440(y-res) * 24(bit-depth) * 60(refresh rate) = 5,308,416,000 bits Thats roughly equal to 4.94 gigabits per display if my math isn't busted.
So actually following the link to Intel's PR spiel yields: "Current versions of Thunderbolt...are limited to an individual 10Gbs channel each for both data and display..." Which is not terribly clear, but I believe is actually trying to say that Thunderbolt devices can currently only make use of a single 10Gbit/s channel at a time.
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.
Facebook
Twitter
RSS
35 Comments
Back to Article
Poik - Tuesday, June 4, 2013 - link
Given the implementation on the new Haswell boards it feels as tho thunderbolt is dead. There's no business case for it right now as you'd be insane to bring out a lot of TB based stuff as outside of Apple there's no support.wicketr - Tuesday, June 4, 2013 - link
Agreed. Until it gets to the point of being really, really fast (PCI Express x16-like) to the point where external graphics cards in a docking station are feasible at comparable speeds, I don't see the point.boeush - Tuesday, June 4, 2013 - link
Seems to me, external graphics card connections would need not just high bandwidth but low latency; I'd almost say that past a certain practical minimum, low latency becomes much more important than higher bandwidth. It's analogous to the characteristics of your internet connection for online gaming: once you get past a certain minimum bandwidth threshold, what starts to matter most is not more bandwidth but lower ping times (i.e. lower 'lag').Unless an entire rendering pipeline including all data and logic can fit into the GPU's on-board memory (and execute purely on the GPU), any communication with CPU and load/stores to DRAM would probably be significantly slower over Thunderbolt than over plain PCI Express (due to extra encoding/decoding steps, and extra length of cable.)
I've seen somewhere that Intel plans to eventually move Thunderbolt to a fiber optic implementation, which would help with transmission latency over the cable (and allow for longer cables), but will probably add extra latency due to optical transceivers... so in the end might be a wash.
A5 - Tuesday, June 4, 2013 - link
If Apple keeps paying them for the controllers, Intel will keep making and developing the chips. Eventually it will be cheap enough to add in to be relevant in the low-margin PC space.Dman23 - Wednesday, June 5, 2013 - link
I completely agree. Once Apple implements Thunderbolt into it's Mac Pro and the prices of the cables / controllers continue to become cheaper, which is already happening, TB will become THE choice when it comes to top-of-the-line Interconnect where getting the fastest and reliable speed is key.Death666Angel - Wednesday, June 5, 2013 - link
So, it goes the Firewire road as many have predicted before. Nothing new. Nothing for me.Dentons - Wednesday, June 5, 2013 - link
The analogy to Firewire is spot on in nearly every respect.Formats without enough technical superiority to warrant their tremendous price premium. That few device makers ever adopt due to pricing. Which is nonetheless championed by a small subset of Mac users, while the rest of the world adopt much cheaper rival that is more than good enough for most.
enealDC - Friday, June 7, 2013 - link
This!CalaverasGrande - Friday, June 7, 2013 - link
Actually Firewire is/was Apple's IP. They charge a licensing fee for every firewire port regardless of whether it is on a pc or mac. TB is Intel's IP. So it probably has a lower fee (if any). I expect Intel is happy just selling the chips and would rather see wider adoption long term, than short term profits.repoman27 - Saturday, June 8, 2013 - link
Actually, FireWire is a trademark held by Apple and freely licensed to third party developers of IEEE 1394 enabled devices who comply with certain guidelines. IEEE 1394 is a serial bus interface standard maintained by the IEEE P1394 Working Group based on an open host controller interface. The standards essential patents required to implement IEEE 1394 are held by 10 corporations including Apple, which was one of the lead designers of the technology, and are licensed as a package by MPEG LA at a rate of 25¢ per finished product, regardless of the number of physical ports.Thunderbolt is a proprietary Intel technology developed at the behest of Apple, which continues to hold certain standard essential patents related to the technology. The Thunderbolt trademark, although originally registered by Apple was transferred to Intel, and licensing and certification are dictated by Intel. This is one of the most glaring differences between FireWire and Thunderbolt. While FireWire may not be an obviously popular consumer interface, it's actually rather ubiquitous due to its open nature and reasonable licensing arrangement. The closed nature of Thunderbolt and Intel's refusal to allow it to become the interoperable standard that Apple had hoped it would be may result in its undoing sooner rather than later.
Hector2 - Tuesday, June 4, 2013 - link
You really don't have a cluerangerdavid - Tuesday, June 4, 2013 - link
Drive-by trolling? I'm tempted to agree with every unsubstantiated word! And I'll give you a free "." to end your hastily-typed phrase! Good job!BMNify - Wednesday, June 5, 2013 - link
They're still trying to revive this stillborn? Amazing! Say a prayer, bury it and move on already!USB 3.0 and DisplayPort have won, just stop already Intel!
axien86 - Tuesday, June 4, 2013 - link
It is too expensive to implement on board and for external interfaces, unless you're part of the 1% where money has no boundaries.Intel should go back to the drawing board...
Sabresiberian - Friday, June 7, 2013 - link
It wouldn't be so expensive if Intel would put support for it in their own freakin CPUs.Seriously. Haswell may be what I'm going to upgrade to when I retire my almost 5-year old X58 system, but it's under protest. 16 PCI lanes, no Thunderbolt, a GPU solution I'll never use and makes overclocking that much harder because Haswell is a bigger chip. Some nice architectural tuning, but not what I want in the bigger picture.
CalaverasGrande - Friday, June 7, 2013 - link
Funny, we have an Asus laptop running an external PCIe box over TB.tipoo - Tuesday, June 4, 2013 - link
So the aggregate bandwidth is no different, right? Just individual channels are faster, but there's half as many? That's a good thing I suppose, unless you were daisy chaining multiple devices which would be better on multiple channels.Dentons - Tuesday, June 4, 2013 - link
Too little, too late, too expensive.USB 3.0 is fast enough for nearly every use case and is orders of magnitude cheaper.
repoman27 - Wednesday, June 5, 2013 - link
Well, actually Thunderbolt is generally 5x - 10x more expensive than USB 3.0, which would only be a single order of magnitude, and only in the extreme cases. And Thunderbolt is 2.5x - 8x faster, which makes it worthwhile to the time is money crowd.While it would have been nice for Intel to have sorted out channel bonding and DP 1.2 support from the get go, it's hardly as if those shortcomings slowed adoption. If Intel has done too little, too late, it's in regards to making the Thunderbolt host controller interface specification available so that it could have a fighting chance of becoming a true interoperable standard. At this rate, as soon as Apple stops buying controllers, Thunderbolt goes out like a proprietary punk bitch.
Dentons - Wednesday, June 5, 2013 - link
It's often much more than 10 times the price, and that price has to be multiplied for each device, hub, or cable a user may need. The dearth of supported devices is also a major issue.Combine those factors, and even those with money to burn have very few use cases that could not be handled with USB 3.0, eSATA, or DisplayPort. On paper, TB may have advantages. In the real world, the performance differences may be hardly noticeable.
I fully agree with your conclusion. Whenever Apple tires of driving down this cul de sac, Thunderbolt will suffer a quick death.
repoman27 - Wednesday, June 5, 2013 - link
Would you care to point me to a single Thunderbolt device that costs more than 10x what a comparable USB 3.0 product does? While there are some very expensive Thunderbolt products on the market, they are often expensive for reasons other than just the interface. Comparing a Thunderbolt 10GbE interface to a generic USB Ethernet adapter is not really a direct comparison of the cost of Thunderbolt vs USB.And the cost multiplier remains constant regardless of the number of devices and cables one buys: (n*a)+(n*b)+(n*c) = n*(a+b+c).
There isn't so much a dearth of devices as there is a dearth of devices that you personally require. DP is display only and uses a simplex main link, eSATA is storage centric and tops out at 4.8 Gbit/s or less, and SuperSpeed USB currently hits a wall at 4 Gbit/s. If your application requires more than that, you turn to Thunderbolt. There is also a fair amount of PCI/PCIe to x silicon out there for which no USB to x silicon currently exists. You could argue that most PCIe slots in desktop PCs are useless as well since the majority of users are content plugging devices into the other ports available to them.
Dentons - Wednesday, June 5, 2013 - link
10x is nothing. How about a 30x price premium? That's how much more you'll pay for Belkin's $300 Thunderbolt hub, or as they say "dock". Sure, it has some add-on features, but most users are never going to use them.Want a cheaper Thunderbolt hub without the extra ports? The cheapest I've seen is $200.
Face the truth. Most users don't need the features offered. The overwhelming lack of devices and exceptional cost has made Thunderbolt a format with one foot in the grave and the other quickly following.
repoman27 - Thursday, June 6, 2013 - link
And what is it exactly that one is supposed to buy for $10 that is comparable in any way to the Belkin Thunderbolt Express Dock?Well, in any event, good luck with your continued trolling.
Dentons - Thursday, June 6, 2013 - link
Ok then, show me a Thunderbolt hub for less than $200? The only hub other than the $300 Belkin seems to be a $200 LaCie. Yes, $200 to $300, for a hub.If Thunderbolt isn't the dying format I believe it to be, prove me wrong. Why, in the 2 + years since the format's release is there such a dearth of devices? Why does a simple hub cost 20 to 30 times that of a USB hub? Why, two years later are the vast majority of Thunderbolt peripherals a handful of hard drives? Devices that would run equally as fast over far cheaper, existing standards.
Clearly, you don't want to accept the truth, but calling me a troll isn't going to change it. Thunderbolt will be lucky to ever be a 10th as popular as Firewire, and that's saying something.
repoman27 - Friday, June 7, 2013 - link
If you're not trolling, then you have no idea what you're talking about. There are no "Thunderbolt hubs". What LaCie makes is a daisy-chainable Thunderbolt 2-port eSATA 3Gb/s adapter. LaCie, a company known for making high-priced gear, lists it at $199.00. The cheapest 2-port eSATA 3Gb/s adapters I could find with a quick search of Google shopping and Newegg start at $19.99, i.e. the Thunderbolt version is 10 times more expensive. This is an extreme case, and exactly the one I was considering when I said there were no Thunderbolt devices that cost more than 10x the price of a comparable alternative.The Thunderbolt docking solutions from Belkin and Matrox are not USB hubs. They're designed so you can quickly and easily plug a single cable into a laptop and by doing so connect it to a display, Gigabit Ethernet, speakers, storage devices, keyboard, mouse, etc. There are a lot of USB 3.0 solutions, most of them based on DisplayLink silicon, that provide similar features. The cheapest one of those I could find is the Kensington sd3000v which you can pick up for $122.09. That makes the Belkin less than 2.5x as expensive, and the USB docks can't do FireWire and require real-time compression / decompression to do display output which results in artifacts and noticeable lag. Furthermore, the Belkin dock is currently one of the very few ways to add USB 3.0 capability to a 2011 Mac.
I'm not out to prove Thunderbolt isn't withering on the vine; it's a technology that has some serious issues to contend with. It just annoys me when folks such as yourself continue to make ridiculous and erroneous claims. For instance, less than a third of the Thunderbolt devices currently available have anything to do with hard drives. If you took the time to check out some of that gear, you'd see why the manufacturers chose Thunderbolt instead of far cheaper, existing standards.
DarkXale - Wednesday, June 5, 2013 - link
You sure that math is right?2560(x-res) * 1440(y-res) * 24(bit-depth) * 60(refresh rate) = 5,308,416,000 bits
Thats roughly equal to 4.94 gigabits per display if my math isn't busted.
DarkXale - Wednesday, June 5, 2013 - link
oh damn, where'd this reply run off to?repoman27 - Wednesday, June 5, 2013 - link
Yep. You forgot that CVT even with reduced blanking still requires sending more "pixels" than are actually displayed. There's blanking, front porch, sync and back porch in both the horizontal and vertical dimensions which works out to 2720 * 1481 * 24 * 60 = 5,800,780,800 or near as makes no difference 5.8 Gbit/s. There's no dividing by 1024 here either, 1 Gbit = 1,000,000,000 bits.wizfactor - Tuesday, June 4, 2013 - link
This still doesn't solve the cost problem that Thunderbolt faces. I remember when Thunderbolt was first introduced, I couldn't help but feel giddy because external GPUs were now within reach. 2 years later, and we are still no closer to reaching that reality.It's honestly getting hard to get excited about a mainstream tech (well, mainstream enough to be shipping with every Macbook) that can only be utilised by deep-pocketed professionals. I'm starting to become crazy enough to think that we will see external GPUs running on 10 Gbps USB 3.0 or USB 4.0 before Thunderbolt becomes affordable for mainstream users.
Joel Kleppinger - Tuesday, June 4, 2013 - link
A yawn is the best I can muster. When it costs an extra $50-70 per port/device, it's just not going to happen. Intel has to decide to make it ubiquitous on motherboards to even give it a chance... but it seems like they don't really care.And if they don't, why should we?
repoman27 - Wednesday, June 5, 2013 - link
"Current implementations are limited to 10Gbps per channel for data or display, both can't be sent over the same channel."This... isn't true. You can drive two 2560x1440 displays while writing 440 MB/s to a Thunderbolt RAID with everything daisy chained off of a single port. The two displays alone require 11.6 Gbit/s which is > 10 Gbit/s. Intel has also stated in the past that each direction in each channel can carry display and/or PCIe data.
DarkXale - Wednesday, June 5, 2013 - link
You sure that math is right?2560(x-res) * 1440(y-res) * 24(bit-depth) * 60(refresh rate) = 5,308,416,000 bits
Thats roughly equal to 4.94 gigabits per display if my math isn't busted.
repoman27 - Wednesday, June 5, 2013 - link
So actually following the link to Intel's PR spiel yields: "Current versions of Thunderbolt...are limited to an individual 10Gbs channel each for both data and display..." Which is not terribly clear, but I believe is actually trying to say that Thunderbolt devices can currently only make use of a single 10Gbit/s channel at a time.Zandros - Wednesday, June 5, 2013 - link
Yeah, I was wondering about that too.Also, apparently Anand is expecting Apple to go 2560x1440@2x instead of 4k. 20 Gbit/s isn't enough for that, so…?
Hrel - Wednesday, June 5, 2013 - link
What's the licensing fee? Infinity bagillion dollars Intel?