OptiX 6.0 Release

Hi all,

OptiX 6.0 is now available.

Download: https://developer.nvidia.com/designworks/optix/download

Release Notes: https://developer.nvidia.com/designworks/optix/downloads/6.0.0/releasenotes

Documentation: http://raytracing-docs.nvidia.com/optix/index.html

Note this release of OptiX requires a display driver version 418 or higher. For linux, you may need to look in Beta drivers.

Drivers: https://www.nvidia.com/Download/index.aspx

OptiX release notes for drivers 431.02(win)/430.26(linux) released 2019 June 10, since 418.81(win)/418.30(linux) February 2019.

Because OptiX is now getting maintenance releases with every driver update every few weeks, we will begin posting release notes here to the forum, as requested by several of you. Since this is the first one, today’s release notes span multiple driver updates since February. For future release notes, we will keep it limited to individual driver updates.

Release notes for changes to the OptiX API & SDK will still be included in the SDK. These driver update release notes represent bug fixes and backward compatible feature updates in addition to the notes on API version changes included with the OptiX SDK.


Fixed a performance issue where updating transforms triggered a large internal buffer copy & reduced frame rate

Fixed a performance issue where Transform::setChild and Group::setChild were becoming slow with large numbers of children

Fixed many errors in the OptiX Programming Guide document
Document RTX trace depth limit 31 for rtContextSetMaxTraceDepth in optix_host.h

Several improvements/fixes for stack size computation in OptiX
Add API calls for specifying max trace depth and max callable program call depth
Made OptiX multi-level traversal stack size adapt to the scene graph depth

Removed optixSelector sample from the SDK

Print info about nvlink configuration and p2p buffer distribution to usage report

Fixed a bug where cudaDriver().CuMemcpyDtoHAsync returned (Illegal address)

Fixed some inconsistent output between Turing and Pascal boards

Fixed a CUDA error when we recover from Sleep for MultiGPU config in TCC mode

Added a context attribute for watertight traversal and enabled it by default

Fixed a big where print buffer size could not be set

Fixed multi-GPU performance not scaling consistently depending on monitor connection

Fixed GeometryTriangles geometry flags being ignored if not used in combination with multi-material or primitiveIndexOffset

Fixed a bug when having no permission to access cache.db gives “Invalid value” error
Set a default for the cache file size (cache.db)

Move nvml.dll to system32 directory on some OEM Windows machines, following new Microsoft policy

Fixed Kepler gpus showing up in rtDeviceGetDeviceCount

Fixed a small memory leak in the OptiX memory allocator

Fixed rtIsTriangleHit, rtIsTriangleHitFrontFace, rtIsTriangleHitBackFace not working in attribute programs

Fixed denoiser tiling artifacts for resolutions < 32