From: Anders Gabrielsson (
Date: Wed Aug 26 2015 - 10:56:20 CDT

Thanks John!
For my issue, making the interactive rendering mode honor the set device mask would suffice.
(Im in the somewhat odd situation where the older card is actually faster than the newer..)

> On 26 Aug 2015, at 17:37, John Stone <> wrote:
> Hi,
> The TachyonL-OptiX ray tracer in VMD can use multiple GPUs, but
> they have to have a matching major compute capability, and the
> minor compute capability can only differ by 0.2. If enough people
> want, it might be possible for me to support multi-GPU rendering
> with mismatched GPUs, but that would primarily be beneficial for
> batch mode rendering. At present, I depend on the built-in
> multi-GPU work scheduling in OptiX itself, and that's where
> the limitation requiring GPUs with matching compute capabilities
> comes from currently.
> Let me know if that addresses your question fully.
> Cheers,
> John Stone
> On Wed, Aug 26, 2015 at 05:19:26PM +0200, Anders Gabrielsson wrote:
>> BTW, Running on a compute node with a K80, both on-board GPUs are used:
>> OptiXRenderer) VMD TachyonL-OptiX Interactive Ray Tracer help:
>> OptiXRenderer) ===============================================
>> OptiXRenderer) Using 2 devices:
>> OptiXRenderer) [0] Tesla K80 CUDA[0], 11.2GB RAM
>> OptiXRenderer) [1] Tesla K80 CUDA[1], 11.2GB RAM
>> Perhaps the issue on my workstation (as described below) is a mismatch of device generations? (mixed SM 3.0 & 5.0)
>> I did further testing with an even older card (SM2.1), and it seems the optix renderer in interactive mode always chooses the device with higher compute capability, always ignoring VMDOPTIXDEVICEMASK.
>> /Cheers
>>> On 25 Aug 2015, at 16:08, Anders Gabrielsson <> wrote:
>>> Regarding the interactive ray tracer in VMD 1.9.2. (Linux64 binary release)
>>> I've got two cards installed in my machine, but it seems I can only use a single device as optix accelerator for the interactive rendering mode. (the one connected to a display -- ie the OpenGL renderer).
>>> VMD reports two accelerators present at launch, but setting "VMDOPTIXDEVICEMASK" does not to influence what device is used for the interactive rendering. (I don't know how to check what devices are used for batch rendering.)
>>> The interface can become very unresponsive for complicated scenes, so offloading some work to the other card would be useful.
>>> Is it actually possible to use both devices for interactive rendering -- or only the headless device?
>>> Are there any hacks to increase responsiveness during rendering when using the display device? (at the cost of reduced rendering speed, I suppose..)
>>> Cheers!
> --
> NIH Center for Macromolecular Modeling and Bioinformatics
> Beckman Institute for Advanced Science and Technology
> University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
> Phone: 217-244-3349