Super User is a question and answer site for computer enthusiasts and power users. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I have a workstation with a recent graphics adapter: Nvidia Gefore FX3700. As I want to use that adapter for long running CUDA programms, it might be very useful not to run also my desktop graphics on the same device. Unfortunately I cannot plug in further graphics boards into that machine. So I've searched and found that display link might be a solution: Not using the FX3700 for graphics but for CUDA and then using the DisplayLink for desktop. As I've read this I thought: "wait!, it would be awesome to just not plug a normal graphics board into the computer but a high performance CUDA device" which would have the enormous advantage of being portable.

But I guess this is not possible since I assume that no external interface has a memory bandwidth which capable of high performance computing (e.g. similar to PCIe), right?

share|improve this question
    
Thunderbolt can handle PCIe - don't know how many x's though. Also don't think there are any external PCIe cards yet but it seems to be something that might happen in the future. – LawrenceC Mar 12 '12 at 16:27
    
It is now the future and it has happened. Still bloody expensive though (about EUR 200 for an external PCIe case with power, connected via thunderbolt) – Hennes Dec 10 '15 at 15:58
up vote 3 down vote accepted

There is ViDock (and it's DIY clones), which probably might enable using CUDA-capable GPU via ExpressCard, which provides 2.5Gb/s bandwidth (PCIExpress 1.0x8 has 2Gb/s). However, ExpressCard slot is common for laptops, but would require installing additional PCIExpress board into desktop.

It is reported to provide a descent performance in games, but when new textures are loaded the lags begin, so the CPU<->GPU bandwidth is indeed a bottleneck. However, its severity depends on what computations you are going to run.

Also, nVidia Tesla S-series (pic) IS an external CUDA-capable GPU, though probbably not of a kind you are looking for.

share|improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .