Skip to content

TransferBench 1.71.02 #366

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
AtlantaPepsi merged 6 commits into from
Nov 5, 2026
Merged

TransferBench 1.71.02 #366

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
69312fb
range option for LaunchTransferBench.sh
AtlantaPepsi Oct 30, 2026
f3450c1
fix for csh
AtlantaPepsi Oct 30, 2026
29a7637
draft for TB perf sweep
AtlantaPepsi Nov 3, 2026
e31cf98
disable CPU access to small BAR
AtlantaPepsi Nov 4, 2026
26fc5a1
Subiteration time issue fix, LaunchTransferBench dryrun
gilbertlee-amd Nov 5, 2026
532a0db
fix for sweep preset
AtlantaPepsi Nov 5, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 17 additions & 0 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,23 @@
Documentation for TransferBench is available at
[https://rocm.docs.amd.com/projects/TransferBench](https://rocm.docs.amd.com/projects/TransferBench).

## v1.71.02
### Added
- LaunchTransferBench.sh now supports bracketed range wildcards in the host list, e.g. `node[1..4]` or
`rack[01..02]-node[01..04]`. Several ranges in one name combine left to right
- LaunchTransferBench.sh now supports `-h`/`--help` and `-n`/`--dry-run`. `--dry-run` shows the expanded host list
and the exact command that would be run on each rank, without actually running them
### Fixed
- LaunchTransferBench.sh now hands the remote command to bash explicitly, so launching on hosts whose login shell
is csh/tcsh no longer fails on the `VAR=value` environment prefix
- Negative `NUM_ITERATIONS` (time-based run duration) no longer runs `NUM_SUBITERATIONS` times longer than
requested. The per-iteration time accumulated toward the time limit was being divided by `NUM_SUBITERATIONS`
before being compared against the requested duration
- sweep preset no longer generates CPU-executed Transfers to/from GPU memory that isn't host-accessible (Large BAR
disabled) or GPU-executed Transfers to another GPU's memory without peer access; such Transfers are now skipped
with a warning instead of failing. Manually configured Transfers that hit the same cases are now rejected during
validation with a clear fatal error instead of behaving incorrectly

## v1.71.01
### Modified
- `a2a` and `a2asweep` presets now default `A2A_DIRECT=0` on MI4XX (gfx1250) hardware, allowing DPX to route
Expand Down
2 changes: 1 addition & 1 deletion CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -104,7 +104,7 @@ set(ENV{ROCM_PATH} "${ROCM_PATH}")
#==================================================================================================
set(TRANSFERBENCH_VERSION_MAJOR 1)
set(TRANSFERBENCH_VERSION_MINOR 71)
set(TRANSFERBENCH_VERSION_PATCH_FALLBACK "01")
set(TRANSFERBENCH_VERSION_PATCH_FALLBACK "02")

# Auto-compute patch from git: count commits since the last v<MAJOR>.<MINOR>.* tag.
# Falls back to TRANSFERBENCH_VERSION_PATCH_FALLBACK when git is unavailable,
Expand Down
Loading
Loading