This sounds like a usefull Thunderbold usage. Does some one remember how much bandwidth normal PCI Express does offer? It will be interesting to see how near normal performance this would offer.
Thing is, thunderbolt is 10 Gb/s, not GB/s. So, thunderbolt is 1.25 GB/s, thus able to handle a 4x electrical lanes of PCI-E 1.x (5x doesn't exist) or 2x electrical lanes of PCI-E 2.x.
Of course, x16 graphics cards will still work, but will only have a 1/8 of the available bandwidth.
Depending on the implementation, Thunderbolt can provide up to four channels of bi-directional data. So at 10Gb/s per channel per direction, you end up with 80Gb/s. That would be 10GB/s in total, however i'm not sure if one device can use all four channels.
You do have a point, but I share your doubt on whether or not a device can properly communicate over the bonded channels.
I can't recall, but doesn't DisplayPort occupy 2 of the 4 channels in the MBP/iMac controller and 1 of the 2 channels in the MBA controller? Or are their connections separate?
I want to say that...okay, the CPU has 16 channels total ('cause Intel's skimping out lately), and half are for the GPU, four are for the display part of Thunderbolt?
So, I need to do my research on this, but the point of the Thunderbolt chip is to actually mux the data channels for transport down the wire, meaning the devices connected via TB will be able to implement more than a single lane, and up to the whole load. So, while a lane is typically reserved for DisplayPort, if a discrete graphics card is being used there will not be any load on that reservation so it should free up it's bandwidth. We'll follow-up on this.
Agreed. I would love to be able to have an external graphics card. We should've had this technology years ago. How hard is it to implement a mobile chipset with a full PCI-E 16x port that has an external plug. Then you get the GPU of your choice and plug it in using a special enclosure. Some of tried in the past but its always been bad at best.
Many laptops today have great CPUs and lots of RAM, but lacking graphics. I don't even care if its hot swappable, just give me the option.
Pci-e 2.0 4x uses 16Gb/s bandwidth. With a single TB port able to use up to 20Gb/s with Lightridge, or 16Gb/s with Eagleridge, it would provide enough bandwidth to support a full 4x slot.
People get way too hung up on bus bandwidth for graphics cards. Even as far back as the AGP days tests showed that the performance loss was not that terrible when using typical GPUs in a slow slot (e.g. putting an AGP 8X card in a 4X slot). PCIe 2.0 16x is an insane amount of bandwidth -- something on the order of 70-80 gigabytes/sec. To give a sense of scale, the fastest DDR3-2133 DIMM you can buy can do something like 16 or 17GB/s.
I wonder what the price will be? Looking at their current products, I have a feeling it would probably make more sense to build your own gaming rig rather than one of their docks.
External graphics cards would be REVOLUTIONARY if they worked well.
Death of the desktop :P - I know everyone here hates to hear that but I wouldn't mind, personally. A notebook form factor is good enough for a CPU, RAM, and hard drive, but not good enough for a good graphics card.
I have several desktops that I leave on all the time (workstations, htpc's, etc..) It would be nice to run my systems off the onboard gfx for 2D tasks and use the external enclosure when I want to game. This also allows me to keep my systems in smaller form factors. Having a portable solution makes a lot of sense to me from a size/heat/energy standpoint as long as the price isn't exorbitant. Another barrier besides pricing would be that none of my systems have Thunderbolt on them...
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
19 Comments
Back to Article
haukionkannel - Wednesday, August 3, 2011 - link
This sounds like a usefull Thunderbold usage. Does some one remember how much bandwidth normal PCI Express does offer?It will be interesting to see how near normal performance this would offer.
Andrew.a.cunningham - Wednesday, August 3, 2011 - link
I think PCIe 2.x gets 8 GB/s either way in an x16 slot, and PCIe 3.x will up that to 16 GB/s. Seems like we could finally start getting close!Zok - Wednesday, August 3, 2011 - link
Thing is, thunderbolt is 10 Gb/s, not GB/s. So, thunderbolt is 1.25 GB/s, thus able to handle a 4x electrical lanes of PCI-E 1.x (5x doesn't exist) or 2x electrical lanes of PCI-E 2.x.Of course, x16 graphics cards will still work, but will only have a 1/8 of the available bandwidth.
superunknown98 - Wednesday, August 3, 2011 - link
Depending on the implementation, Thunderbolt can provide up to four channels of bi-directional data. So at 10Gb/s per channel per direction, you end up with 80Gb/s. That would be 10GB/s in total, however i'm not sure if one device can use all four channels.Zok - Wednesday, August 3, 2011 - link
You do have a point, but I share your doubt on whether or not a device can properly communicate over the bonded channels.I can't recall, but doesn't DisplayPort occupy 2 of the 4 channels in the MBP/iMac controller and 1 of the 2 channels in the MBA controller? Or are their connections separate?
Wolfpup - Thursday, August 4, 2011 - link
I want to say that...okay, the CPU has 16 channels total ('cause Intel's skimping out lately), and half are for the GPU, four are for the display part of Thunderbolt?JasonInofuentes - Thursday, August 4, 2011 - link
So, I need to do my research on this, but the point of the Thunderbolt chip is to actually mux the data channels for transport down the wire, meaning the devices connected via TB will be able to implement more than a single lane, and up to the whole load. So, while a lane is typically reserved for DisplayPort, if a discrete graphics card is being used there will not be any load on that reservation so it should free up it's bandwidth. We'll follow-up on this.Jason
Andrew.a.cunningham - Wednesday, August 3, 2011 - link
That lowercase B gets me every time. Article updated for accuracy. Let my comment stand as a testament to my error.FITCamaro - Wednesday, August 3, 2011 - link
Agreed. I would love to be able to have an external graphics card. We should've had this technology years ago. How hard is it to implement a mobile chipset with a full PCI-E 16x port that has an external plug. Then you get the GPU of your choice and plug it in using a special enclosure. Some of tried in the past but its always been bad at best.Many laptops today have great CPUs and lots of RAM, but lacking graphics. I don't even care if its hot swappable, just give me the option.
A5 - Wednesday, August 3, 2011 - link
"We should've had this technology years ago.""Some of tried in the past but its always been bad at best."
There's your answer. Turns out that it's hard to do. I don't think it'll really be practical until the optical version of Light Peak is out.
sean.crees - Thursday, August 4, 2011 - link
Hard OCP did a comparison between 16x and 4x pci-e 2.0 slots. They show how a 4x slot does not bottleneck performance.http://www.hardocp.com/article/2010/08/25/gtx_480_...
Pci-e 2.0 4x uses 16Gb/s bandwidth. With a single TB port able to use up to 20Gb/s with Lightridge, or 16Gb/s with Eagleridge, it would provide enough bandwidth to support a full 4x slot.
DesktopMan - Wednesday, August 3, 2011 - link
Current graphic cards work surprisingly well on limited bandwidth. At PCI Express 2.0 4x (Thunderbolt) they're already quite close to 100%. See http://www.techpowerup.com/reviews/AMD/HD_5870_PCI... etc.I have a GTX 285 in an ExpressCard Vidock on PCI Express 2.0 1x, which still performs better than most mobile GPUs available. 60-80% of a full 16x.
Metaluna - Wednesday, August 3, 2011 - link
People get way too hung up on bus bandwidth for graphics cards. Even as far back as the AGP days tests showed that the performance loss was not that terrible when using typical GPUs in a slow slot (e.g. putting an AGP 8X card in a 4X slot). PCIe 2.0 16x is an insane amount of bandwidth -- something on the order of 70-80 gigabytes/sec. To give a sense of scale, the fastest DDR3-2133 DIMM you can buy can do something like 16 or 17GB/s.KeithP - Wednesday, August 3, 2011 - link
I wonder what the price will be? Looking at their current products, I have a feeling it would probably make more sense to build your own gaming rig rather than one of their docks.-KeithP
zorxd - Wednesday, August 3, 2011 - link
Great idea in theory. In practice, we all know that this will be way too expensive to be useful.peacemyfriends - Wednesday, August 3, 2011 - link
Isn't the problem NOT bandwidth but latency?External graphics cards would be REVOLUTIONARY if they worked well.
Death of the desktop :P - I know everyone here hates to hear that but I wouldn't mind, personally. A notebook form factor is good enough for a CPU, RAM, and hard drive, but not good enough for a good graphics card.
groundhogdaze - Thursday, August 4, 2011 - link
I have several desktops that I leave on all the time (workstations, htpc's, etc..) It would be nice to run my systems off the onboard gfx for 2D tasks and use the external enclosure when I want to game. This also allows me to keep my systems in smaller form factors. Having a portable solution makes a lot of sense to me from a size/heat/energy standpoint as long as the price isn't exorbitant. Another barrier besides pricing would be that none of my systems have Thunderbolt on them...sean.crees - Thursday, August 4, 2011 - link
"since Thunderbolt gives devices 10 Gb/s of bandwidth to work with"Isn't it 10Gb/s per channel, with 2 channels per physical TB port. So wouldn't it actually be 20Gb/s bandwidth for a single TB port?
archer75 - Tuesday, January 17, 2012 - link
I would love this for my imac. It could use a video card upgrade.