Thunderbolt PCIe expansion?

[Joseph Owens]

[Andrew Sableton] “Thunderbolt PCI e expansion”

The early word that we were getting back from manufacturers is not to expect PCIe/ Thunderbolt implementation or interface as T-Bolt is a motherboard sandwich-layer process and doesn’t lend itself at all to this concept.

jPo

You mean “Old Ben”? Ben Kenobi?

[Andrew Sableton]

So the PCI e expansion chassis that have been announced would work for things like PCI e fibre cards, PCI e video I/O etc but not for graphics cards?

A.S.

[Juan Francisco calero]

Sonnet say: “Imagine using full-size professional video capture cards” in the page about his Echo Express PCIe 2.0 Expansion Chassis with Thunderbolt Ports.

More information in his web: https://www.sonnettech.com/product/thunderbolt/index.html

Available fall 2011… if it works with Davinci will be great.

[Clayton Burkhart]

I suspect that what you are looking for won’t actually be implemented until the Thunderbolt v2 comes out. Currently the speed of Thunderbolt (10 Gb/s) still can’t match the speed of your PCIe slots. With v2 that would become more of a reality as it will exponentially increase in speed (100 Gb/s).

[Andrew Sableton]

Well….that refers to video I/O which has nothing to do with the graphics card capability……

[Andrew Sableton]

Yes I have read that the speed of Thunderbolt being less than PCI e in the current iteration, but am not clear on whether than in itself would prevent a graphics card from working, if the drivers were able to recognize it, or just hinder the performance to some degree. Obviously I am not a highly technical guy but try to understand these things to the extent I can and need to in order to get a working system going and functioning reliably.

[Chris Kenny]

[Clayton Burkhart] “I suspect that what you are looking for won’t actually be implemented until the Thunderbolt v2 comes out. Currently the speed of Thunderbolt (10 Gb/s) still can’t match the speed of your PCIe slots. With v2 that would become more of a reality as it will exponentially increase in speed (100 Gb/s).”

Even 3D games don’t suffer as much of a performance penalty as might be imagined when moving from 16x PCIe to, say, 4x. Perhaps a 10-20% hit to frame rate. If I understand the way Resolve uses GPUs, there should be even less impact there, at least with 1080p/2K. All you need, I think, is enough bandwidth to send an uncompressed video stream out to the GPU and get another one back, and Thunderbolt provides more than enough bandwidth to do that.

Thunderbolt PCIe devices even show up the same as internal PCIe devices to the host operating system, meaning graphics cards wouldn’t even require special drivers for this.

There may be some more subtle issue that prevents this from working well with Resolve, or I may be misunderstanding how Resolve uses the GPU. But at a first glance external GPUs via Thunderbolt seem quite plausible.

—
Digital Workflow/Colorist, Nice Dissolve.

You should follow me on Twitter here. Or read our blog.

[Paul Jay]

Cubix announced a thunderbolt chassis.

[Eric Fiegehen]

Chris,

I’ve been steering clients towards the 16-channel GPU-Xpander Desktop & Rackmount configurations ever since Resolve 8 was released. There are several new GPU-accelerated features in Resolve 8 which seem to saturate the PCIe bus very quickly. Not even an 8-channel, 40Gbps PCIe connection is enough bandwidth for many users.

Thunderbolt’s current 10Gbps performance spec appears to be more than adequate for a variety of OSX applications, including Resolve Lite. I recommend contacting Blackmagic Design first before attempting to run the full version of Resolve on a Thunderbolt-equiped Macbook Pro or iMac.

Eric Fiegehen
Cubix Corporation