Win32 builds

I know there are plans of removing 32bit optix builds. I understand the extra work in maintenance and testing that is caused by keeping both 32bit and 64bit.

Our one reason for using 32bit optix, is for the, IMHO, flagship feature of Visual Studio, edit and continue. For some reason, they have yet to add in support for edit and continue in native 64 bit c++ builds.

For graphics programming, edit and continue is nothing short of amazing. I can break, edit code, continue, and immediately see changes.

I’m also of course a fan of using statics, break, edit the static in the debugger and continue. These give you a “poor mans” edit and continue.

I’m hoping Microsoft does something to update edit and continue. It is the only reason we internally keep a 32 bit build ourselves.

That already happened with OptiX 3.8.0 beta.
See the OptiX 3.8.0 Release Notes. Access as described here:
https://devtalk.nvidia.com/default/topic/820766/optix/optix-3-8-beta-is-posted-with-quadro-vca-support/

That is not a decision isolated to OptiX. Since it’s based on CUDA it has the same support limitations.
Please read Chapter 3 “Unsupported Features” of the CUDA 7.0 release notes:
http://developer.download.nvidia.com/compute/cuda/7_0/Prod/doc/CUDA_Toolkit_Release_Notes.pdf

Furthermore it’s not only about maintenance and testing, it’s about the actual functionality.
Anything which requires more than 4GB address space wouldn’t work with a 32-bit application. That includes being able to fully use all resources of a graphics boards with more than 4GB of VRAM. Professional applications easily exceed that.

Agreed on the limitations of 32bit. Lets just hope Microsoft gets Edit and continue working in 64bit! I’ll miss it.

Hello!
We have a 32-bit product that is hard to re-build to 64-bit architecture. So, if Optix is going to be only 64-bit, we will have to do something: resign Optix or make some tricks with 64-32 bit interoperation. Both solutions are undesirable for us.

I understand the motivation with 4-GB memory and overal progress direction, but:

  • Most of existing GPUs have less than 4 GB of memory.
  • A lot of people are still running 32-bit Windows, some of our customers among them.

Are these arguments not enough to support 32-bit versions of CUDA and Optix? For example, even Windows 10 still has 32-bit versions (due to compatibility with older software, I think), although the RAM is always much larger than GPU memory.

I replied in another thread, but here’s more detail.

Yes, the plan is that after OptiX 3.8 we will finish removing 32-bit support, which currently only exists on Windows. The release notes say it’s totally gone from 3.8, but we actually did ship the 32-bit DLLs in 3.8, just for you. :) We have massively renovated the OptiX code base since 3.X. Some details are in my GTC talk: http://bit.ly/1BJ7TsM The new code base simply will not support 32-bit.

I really recommend biting the bullet and rebuilding your app for 64-bit. More and more of the infrastructure you depend on will stop being supported on 32-bit, so in most cases that value of keeping it alive is too low.

FYI, only 14% of users are still on 32-bit Windows, at least according to the Steam survey.

Thank you for building 32-bit version for us! :)

Thanks again from us as well. Although we don’t ship a 32 bit build, that edit and continue, when it comes to testing graphics code, making small changes, and seeing the results, is just amazing.