Skip to content

Conversation

@harrism
Copy link
Contributor

@harrism harrism commented Dec 4, 2025

We were calling cudaDeviceGetProperties every time we launch the GaussianRasterizeBackward kernel. This is expensive (> 1ms per call). This definitely impacted our training time.

This PR adds a new utility that uses (faster) cudaDeviceGetAttribute instead, and only ever calls it once per device and caches the result.

Signed-off-by: Mark Harris [email protected]

@harrism harrism requested a review from a team as a code owner December 4, 2025 04:12
@harrism harrism added enhancement New feature or request optimization Performance or memory optimization core library Core fVDB library. i.e. anything in the _Cpp module (C++) or fvdb python module Gaussian Splatting Issues related to Gaussian splattng in the core library labels Dec 4, 2025
@harrism harrism added this to fVDB Dec 4, 2025
Copy link
Contributor

@swahtz swahtz left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice, great catch!

Copy link
Contributor

@blackencino blackencino left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

One spelling change request, one nudge that you can ignore, otherwise LGTM.

@harrism harrism enabled auto-merge (squash) December 8, 2025 23:10
@harrism harrism merged commit 7ce202d into openvdb:main Dec 8, 2025
32 checks passed
@github-project-automation github-project-automation bot moved this to Done in fVDB Dec 8, 2025
@harrism harrism deleted the mjh/fea-max-shared-memory-util branch December 9, 2025 01:02
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

core library Core fVDB library. i.e. anything in the _Cpp module (C++) or fvdb python module enhancement New feature or request Gaussian Splatting Issues related to Gaussian splattng in the core library optimization Performance or memory optimization

Projects

Status: Done

Development

Successfully merging this pull request may close these issues.

3 participants