-
**Is your feature request related to a problem? Please describe.**
Both national labs, internal customers, and others are asking for a Mi300 based roofline analysis. In v2.0 release this was temporar…
-
### Problem Description
`make test` or `cmake /opt/rocm/share/rpp/test` command should find and calibration all prerequisites
### Operating System
ALL
### CPU
ALL
### GPU
AMD Instinct MI300
#…
-
Kokkos 4.4 has recently been release (changelog herehttps://github.com/kokkos/kokkos/blob/develop/CHANGELOG.md)
There are some relevant additions that we might want sooner than later (e..g, support f…
-
### What happened?
Some changes during dispatch creation (encountered here https://github.com/iree-org/iree/pull/18341) causes f16 SDXL to produce incorrect results under aggressive fusion. This on…
-
### Description of errors
https://hub.docker.com/r/rocm/pytorch
Can you please add this ? Many are confused as the Overview does not contain mi300 series information.
**Starting from ROCm6.0, …
-
### Your current environment
I am trying out FP8 support on AMD GPUs (MI250, MI300) and the vLLM library does not seem to support AMD GPUs yet for FP8 quantization. Is there any timeline for when thi…
-
**Describe your question**
Hi, Thanks for building omniperf. It is very helpful for kernel profiling. I have a few questions.
1. I'm using omniperf on MI200 on RHEL9 and would like to use the ro…
-
### Problem Description
```
1: -- Found OpenMP_C: -fopenmp (found version "4.5")
1: CMake Error at /usr/local/lib64/python3.6/site-packages/cmake/data/share/cmake-3.25/Modules/FindPackageHandleStan…
-
### Problem Description
Version:
rocm 6.1.1
grpc 1.61.0
rdc rocm-6.1.1
```
1717144799.236 ERROR RdcMetricFetcherImpl.cc(211): PCIe throughput not supported on GPU 1
1717144799.237 ERROR RdcMetr…
-
AFAIK, these opcodes are not supported on GFX940.
Examples of failed tests:
buffer_wbinvl1
buffer_wbinvl1_vol
buffer_store_lds_dword s[4:7], s2 offset:4095 lds
@rampitec, should t…