OptiX 3.7 final release is posted!

Dear OptiX Developers,

The OptiX team is heads-down working on huge things for next year, but we know we need to keep your tools current. So, we are pleased to announce the beta release of OptiX 3.7. This release enables new GPUs and new platforms, improves stability and performance, and enables very fast, flexible geometry instancing in OptiX Prime.

OptiX 3.7 is already integrated into Iray 2015 and mental ray. Check out this great video of instancing in mental ray at their blog: http://blog.mentalray.com/. (The instancing is toward the end.)

As always, you can download this OptiX release using your credentials that you received by email when you were accepted to the registered developer program. If you are not in the program you must join to download OptiX 3.7. Go to NVIDIA OptiX™ Ray Tracing Engine | NVIDIA Developer. If you forgot your password, email OptiX-Help@nvidia.com.

Sincerely,

David McAllister
OptiX Manager | NVIDIA Corporation

Enhancements in OptiX 3.7 beta 1
• Instancing support in OptiX Prime. The implementation is very fast. The typical performance hit vs. non-instanced geometry is only 5%.
• Tesla K80 support with SM 3.7. This is the highest performing GPU for ray tracing.
• Visual Studio 2013 support.
• CUDA 6.5 support.
• OpenGL libraries are now dynamically loaded, allowing easier linking on Linux, especially for HPC installations.
• Samples have updated model loading code.
• OptiX API Capture bug fixes for OptiX Prime.
• rtpQuerySetCudaStream API added to OptiX Prime to for increased performance on asynchronous queries.
• RTP_BUFFER_FORMAT_INDICES_INT3_MASK_INT is an additional way to suppress intersections without rebuilding the Model’s acceleration structure.
• Bug fixes to Trbvh.
• Bug fix for when RT_EXCEPTION_PROGRAM_ID_INVALID is used with callable programs.
• Bug fix Only dirty device-side storage on rtBufferGetDevicePtr.
• Bug fix Prevent Collision sample window resizing when using the fake rendered background, since it doesn’t scale.
• Bug fix Add more detail to the “Unable to set the CUDA device” error message.
• Documentation improvements in OptiX Prime API reference and Buffer of Buffer IDs in Programming Guide, and RTtimeoutcallback in OptiX API reference.

Hi

I don’t see no 32-bit version of Optix 3.7.0 beta, could you add a 32-bit build on ftp ?

Thank you

edit: I mean a Windows 32-bit

Hello.

The same: build 32-bit version, please.

We just released beta 2. Here’s what we fixed:

  • Support PTX copysign instruction.
  • Don’t ship 32-bit OptiX Prime libraries, which was never a supported configuration.
  • Additional enhancements in OpenGL library and glew linking.
  • Bug fix for PROGRAM_ID exception check.
  • Bug fix for OAC trace capture.
  • Bug fix while parsing input programs for Fermi.
  • Bug fix with callable programs in Win32 build.
  • Bug fix with chunked Trbvh builds in OptiX.

To the two of you who requested Windows 32-bit, please look in the 64-bit distro for the 32-bit libraries. We haven’t shipped 32-bit precompiled samples. We don’t expect this to be a problem, but please let me know if you feel otherwise.

We are currently planning to not support 32-bit Windows starting mid-2015. Please let me know how big of a problem this might be. We ended 32-bit Mac and Linux over a year ago, with no sad faces, so we’re hoping for the same with Windows.

Thank you very much, it works like a charm

I should be able to compile in 64-bit, so it’s not a problem for me. I just didn’t know you planned to drop 32-bit support, but now that you announced it it’s ok.

Hi DaveMc,

The bug I found here is still a problem: https://devtalk.nvidia.com/default/topic/795148/optix/optix-prime-3-7-bug-report/

DaveMC,

Thank you for including 32-bit version into a new distribution.
There is no problen with 64-bit samples, but we need the 32-bit version of library binaries, because our product is still 32-bit and we are unable to rebuild it. The end of 32-bit Optix support will be a big problem for us.

OptiX 3.7 beta 3 is released. Registered developers can download it in the normal way. If you’re not a registered developer you can apply for the program here: NVIDIA OptiX™ Ray Tracing Engine | NVIDIA Developer

The bug fixes in this release:

Bug fix in both OptiX Prime and OptiX with Trbvh. The compiler in the R346 driver has a regression that caused kernel timeouts.
Bug fix for OAC trace capture in OptiX Prime.
Bug fix for st.local.v* instructions targeting a local pointer.
Bug fix with Trbvh in locales that use “,” as the decimal point.
Bug fix in OptiX Prime with chunked acceleration structure build on rtpModelCopy.
Bug fix for missing inline decorator in optix_primepp.h.

DaveMC,
I’m sorry, but somehow I don’t receive comments from this topic via e-mail (your last comment), though, I pressed “Subscribe” button and now I can see “Unsubscribe”. Maybe I understand something wrong?

Also, when will the stable version of Optix 3.7 be released?

Try if setting up the RSS feed on the main OptiX forum site (one hierarchy up) in your e-mail client works better.
That should report new posts once a day, though formatting gets lost.

Seems like there is a problem with sticky threads and the RSS feeds though. I didn’t get anything from this thread either. I’ll inform the forum software team.

OptiX 3.7 final release is out. The major enhancements in OptiX 3.7 are:
• Instancing support in OptiX Prime. The implementation is very fast. The typical performance hit vs. non-instanced geometry is only 5%. See a demo in this post: http://blog.mentalray.com/.
• Tesla K80 support with SM 3.7. This is the highest performing GPU for ray tracing.
• Visual Studio 2013 support.
• CUDA 6.5 support.
• OpenGL libraries are now dynamically loaded and link more easily with the glew library, allowing easier linking on Linux, especially for HPC installations.
• Samples have updated model loading code.
• PTX copysign instruction support.
• rtpQuerySetCudaStream API added to OptiX Prime for increased performance on asynchronous queries.
• RTP_BUFFER_FORMAT_INDICES_INT3_MASK_INT is a way to suppress intersections without rebuilding the Model’s acceleration structure.
• Improved concurrency in OptiX Prime with multiple GPUs.

Enhancements in OptiX 3.7 final (since beta 3)
• Trbvh builder is the default in several OptiX samples.
• Improved concurrency in OptiX Prime with multiple GPUs.
• Improved error messages for invalid children of Selector nodes.
• Bug fix with bindless texture IDs getting deleted and reused.
• Bug fix in samples. Some textures were not being loaded.

Hi.
Thank NVidia for building 32-bit version of Optix 3.7 final.

But in the version 3.8 release notes I can see:

Operating System:
-Windows 7/8/8.1 64-bit; Linux RHEL 4.8 or Ubuntu 10.10 - 64-bit
Note: 32-bit operating systems are no longer supported.

Is it a plan for all new versions of Optix in the future? If it’s true, we’ll be unable to use Optix in our product.

In OptiX 3.7 release notes, we see CUDA R343 or later driver is required. Is there a way to check whether a required driver is installed? Tried cudaDriverGetVersion() but it returns CUDA version instead of driver release version. Thank you.

Hello,
Is any one alive here? Please answer my question about 32-bit support in future Optix releases, it’s important for us.

Yes, the plan is that after OptiX 3.8 we will finish removing 32-bit support, which currently only exists on Windows.

I’m sure you’re working on a 64-bit migration plan as only 14% of users are still on 32-bit Windows, at least according to the Steam survey. You’re probably aware that CUDA has dropped most 32-bit support as well.

Could you tell us more about your use case so we can help you figure out a good path forward?

Thank you for the reply, I just wanted to understand the plans and of Nvidia regarding 32-bit support. We have several solutions that are, however, more or less hard for us.