You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[FEA]: Evaluate performance impact of graph capturable segmented sort #10800
This issue can be closed with nvbench compare result run on upstream device segmented sort vs graph capturable device segmented sort from #10799 on B200 and 6000 Pro.
Activity
nanan-nvidia commentedon Aug 14, 2026
nvbench compare of upstream (
f32edf2c69) v the implementation #10799The regressions concentrate in the same two bands on both devices:
B200 benchmark data
## [0] NVIDIA B200small
[0] NVIDIA B200
large
[0] NVIDIA B200
Summary
nanan-nvidia commentedon Aug 14, 2026
RTX 6000 benchmark data
small
[0] NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition
large
[0] NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition
Summary
gevtushenko commentedon Oct 7, 2026
@anikaj-eng please evaluate performance impact to see if the issue can be closed.