The Thunderbolt Devices Trickle In: Magma's ExpressBox 3T

12 Comments

View All Comments

• cervantesmx - Wednesday, September 7, 2011 - link

  I am actually looking forward to this to be able to use it on my otherwise useless laptop :)

• allometry - Wednesday, September 7, 2011 - link

  Village Instrument's CEO was on Facebook several weeks ago testing the waters of a device that would accomplish external graphics as well. I'm uncertain whether or not it will include support for high end video cards or what the PCI Express length will end up being.
  
  Either way, I'm pretty excited about this.

• dgingeri - Wednesday, September 7, 2011 - link

  let me see:
  
  twice the bandwidth of USB3
  supports external displays
  takes up 4 PCIe lanes
  
  Really, not very impressed. External SAS has much more bandwidth (24Gbs each way) with a 4 lane connector, and can use an adapter with RAID support built in. Display Port is much better for external displays, as it has enough bandwidth for multiple displays and audio. Thunderbolt is just another alternative, but one that requires paying a licensing fee to Intel and supports 2 different things with one connector. It's not that big of a deal.

• DesktopMan - Wednesday, September 7, 2011 - link

  The point here is that Thunderbolt works with existing devices. It doesn't "take" 4 PCIe lanes, it offers them externally. Existing drivers work out of the box. The use cases are many; external graphic cards, sound cards, video capture cards etc.
  
  As I see it the biggest competitor in the future will be the external PCIe standard that's up and coming, but that has the disadvantage of not offering display signaling in the same cable.

• FaaR - Thursday, September 8, 2011 - link

  There's no license fee involved with thunderbolt.
  
  Also, when are you going to pile together a RAID array to saturate 24Gbit/s, and what would its actual use be to any normal person? That's ludicrous, it'd cost thousands, (and most likely probably tens of thousands) of $ just for the controller hardware and drives. Not to mention the cable, which is rather a lot more unwieldy than the slim TB cable.
  
  Also, displayport isn't "much better", TB supports multiple displays (2600*1400 rez and maybe more), and audio, camera/mic, networking and I/O, as evidenced by the Apple Thunderbolt Display.
  
  Not sure why you're being so negative. 40Gbit/s in a tiny connector and at a low cost is unprescedented in the realm of computing up until now.

• DesktopMan - Wednesday, September 7, 2011 - link

  "No word on the width of each slot (x4? x8? x16?)."
  
  As long as it's not x1 and open ended it doesn't really matter.
  
  220W is not bad for a graphics card, you can run a nVidia GTX 560TI on that.

• MySchizoBuddy - Wednesday, September 7, 2011 - link

  thats for just one GPU, what about the other two slots how will you power them as well. This is seriously underpowered.

• repoman27 - Wednesday, September 7, 2011 - link

  A single Thunderbolt channel is only a 10 Gbps link. Therefore the entire enclosure is essentially sharing 2.5 lanes of PCIe 2.0. To put that in perspective, though, that is still half of the bandwidth of the DMI that connects the CPU to the PCH on Sandy Bridge systems.
  
  As for Thunderbolt being NBD, if we look at the block diagram of the Light Ridge chip, on the back end we have connections for 4 x PCIe 2.0 lanes (at a nominal 5 Gbps, full-duplex each) and 2 x DisplayPort 1.1a links (up to 10.8 Gbps apiece for the main links). On the front side are up to two ports which provide 2 x 10 Gbps, full-duplex channels each. As was demonstrated right here at Anandtech, each channel can provide real world throughput of 1000 MB/s. Light Ridge is a 40 Gbps host controller which can operate at 80% efficiency, and even after you account for any 8b/10b encoding, still has 33.48 Gbps of bandwidth behind it.
  
  If we look at pretty much any of the available USB 3.0 host controllers on the market today, they have a PCIe 2.0 x1 connection at the back, and 2 to 4 USB 3.0 ports that provide a nominal symbol rate of 5 Gbps in SuperSpeed mode. Renesas claims their latest host controllers paired with their new UASP driver can achieve real world throughput of 370 MB/s on a single port. USB 3.0 is a 4 Gbps interface, currently only capable of operating at 74% efficiency, with a 4 Gbps back end after accounting for 8b/10b encoding.
  
  I'm not sure about any licensing fees, but at the moment you do have to buy your Thunderbolt controllers from Intel. Which may be a problem, since they have to pump out enough in their first year of production to not impede Apple's sales of Thunderbolt enabled devices (somewhere in the ballpark of 16m for 2011, and similar to the combined total number of USB 3.0 host controllers that shipped in 2010 from multiple vendors). Put another way, by the end of 2011, only 5 times more PCs equipped with USB 3.0 will have been shipped than Thunderbolt, even though Apple is the only OEM shipping TB enabled PCs at this time, has only been doing so for about 8 months, and only has a 10% share of the US PC market.

• ohknarf - Wednesday, September 7, 2011 - link

  Macbook pro's thunderbolt is dual 10Gbps but can it support x16 pci speed? Say if these external box do have x16 pci supports would a Nvidia gtx 590 or Radeon HD 6990 work or we limited?

• The Hardcard - Sunday, September 11, 2011 - link

  Can the channels on Light Ridge be combined. If you have a real world rate of 1000 MB/s could a graphics card use both channels for 2000 MB/s? That is per direction?
  
  So you put a GPU with 2 or 4 GB of GDDR5 in. Is 2000 MB/s enough to cover CPU communications and transfers from main RAM? Could that get you to top-tier gaming? Anywhere near the high end?