Parallel GPU buffer writes #22314
Merged
Parallel GPU buffer writes #22314
+129
−82
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
alice-i-cecile
added
A-Rendering
james7132
approved these changes
Dec 30, 2025
kfc35
reviewed
Jan 2, 2026
Co-authored-by: Kevin Chen <chen.kevin.f@gmail.com>
Member
Author
|
Revisted benchmarks on latest main, and improvements are still reproducible. 
Bottle eck is mesh collection, which is improved in #22297. |
james7132
added
S-Ready-For-Final-Review
cart
pushed a commit
that referenced
this pull request
Jan 8, 2026
# Objective - After a series of optimizations making render and postupdate more parallel, `write_batched_instance_buffers` was regularly one of the largest spans with very low thread use, sitting at 4ms in 1 4ms frame. This makes it an ideal target to improve throughput. Note this screenshot doesn't include some visibility system optimizations: <img width="650" height="718" alt="image" src="https://github.com/user-attachments/assets/bbd6762b-5145-48f8-a427-5da3cb11a04a" /> ## Solution - Spawn tasks for writing buffers to the GPU. This is especially helpful for `current_input_buffer` and `previous_input_buffer`, which take about the same time and are the longest buffer writes - moving these to tasks effectively halves the time spent in the system. <img width="588" height="251" alt="image" src="https://github.com/user-attachments/assets/0a086e7a-1d3c-4c17-9d66-eff94196943d" /> - In the 250k bevymark_3d stress test, this saves 1.7ms in the system, and 2.8ms in frame time frametime <img width="620" height="376" alt="image" src="https://github.com/user-attachments/assets/a4c106ac-7668-4f8a-970f-71cbb8be851c" /> system <img width="1384" height="744" alt="image" src="https://github.com/user-attachments/assets/5c42227d-8ee5-4b84-bc1a-c04768356255" /> ## Testing - `cargo rer bevymark_3d --features=debug,trace_tracy -- --benchmark --waves 250 --per-wave 1000` --------- Co-authored-by: Kevin Chen <chen.kevin.f@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Objective
write_batched_instance_bufferswas regularly one of the largest spans with very low thread use, sitting at 4ms in 1 4ms frame. This makes it an ideal target to improve throughput. Note this screenshot doesn't include some visibility system optimizations:Solution
current_input_bufferandprevious_input_buffer, which take about the same time and are the longest buffer writes - moving these to tasks effectively halves the time spent in the system.frametime
system
Testing
cargo rer bevymark_3d --features=debug,trace_tracy -- --benchmark --waves 250 --per-wave 1000