github/ROCm
1
1
Fork 0
mirror of https://github.com/ROCm/ROCm.git synced 2026-01-08 22:28:06 -05:00
Code Issues Packages Projects Releases Wiki Activity

710 RCCL Known Issues and CRIU note update (#5647)

Browse Source 
* RCCL ALltoALL known issue added

* CRIU note added

* Minor change

* Review feedback and AMDSMI detailed changelog link added

* Github issue link added
This commit is contained in:
Pratik Basyal
2025-11-11 16:54:36 -05:00
committed by GitHub
parent eb956cfc5c
commit 09c6a9fdef
2 changed files with 12 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
CHANGELOG.md
View File

@@ -48,6 +48,10 @@ for a complete overview of this release.
* Fixed certain output in `amd-smi monitor` when GPUs are partitioned. It fixes the issue with amd-smi monitor such as: `amd-smi monitor -Vqt`, `amd-smi monitor -g 0 -Vqt -w 1`, and `amd-smi monitor -Vqt --file /tmp/test1`. These commands will now be able to display as normal in partitioned GPU scenarios.
```{note}
See the full [AMD SMI changelog](https://github.com/ROCm/amdsmi/blob/release/rocm-rel-7.1/CHANGELOG.md#amd_smi_lib-for-rocm-710) for details, examples, and in-depth descriptions.
```
### **Composable Kernel** (1.1.0)
#### Added

8
RELEASE.md
View File

@@ -735,6 +735,10 @@ For a historical overview of ROCm component updates, see the {doc}`ROCm consolid
* Fixed certain output in `amd-smi monitor` when GPUs are partitioned. It fixes the issue with amd-smi monitor such as: `amd-smi monitor -Vqt`, `amd-smi monitor -g 0 -Vqt -w 1`, and `amd-smi monitor -Vqt --file /tmp/test1`. These commands will now be able to display as normal in partitioned GPU scenarios.
```{note}
See the full [AMD SMI changelog](https://github.com/ROCm/amdsmi/blob/release/rocm-rel-7.1/CHANGELOG.md#amd_smi_lib-for-rocm-710) for details, examples, and in-depth descriptions.
```
### **Composable Kernel** (1.1.0)
#### Added
@@ -1513,6 +1517,10 @@ As a workaround, first install `libopenblas-dev` or `libopenblas-deve`, dependin
Some `gemm_ex` operations with `half` or `f32_r` data types might yield 16-bit precision results instead of the expected 32-bit precision when matrix dimensions are m=1 or n=1. The issue results from the optimization that enables `_ex` APIs to use lower precision multiples. It limits the high-precision matrix operations performed in PyTorch with rocBLAS and hipBLAS. The issue will be fixed in a future ROCm release. See [GitHub issue #5640](https://github.com/ROCm/ROCm/issues/5640).
### RCCL profiler plugin failure with AllToAll operations
The RCCL profiler plugin `librccl-profiler.so` might fail with a segmentation fault during `AllToAll` collective operations due to improperly assigned point-to-point task function pointers. This leads to invalid memory access and prevents profiling of `AllToAll` performance. Other operations, like `AllReduce`, are unaffected. It's recommended to avoid using the RCCL profiler plugin with `AllToAll` operations until the fix is available. This issue is resolved in the {fab}`github`[RCCL `develop` branch](https://github.com/ROCm/rccl/tree/develop) and will be part of a future ROCm release. See [GitHub issue #5653](https://github.com/ROCm/ROCm/issues/5653).
## ROCm resolved issues
The following are previously known issues resolved in this release. For resolved issues related to