We are excited to announce a new open-source release of our GIM SR-IOV Driver. Along with regular bug fixes and improvements, this release is aligned with ROCm 10.0,
- AMD-SMI/GIM reliability — Bug fixes: GPU-reset event reporting and power-cap data behavior,
- RHEL 10.2 support — New OS enablement for MI325X and MI300X host/guest combinations.
Hardware / Features Supported for this Release
| Hardware | Host OS | Guest OS | ROCm/Graphics version | Number of VFs per GPU | Infinity Fabric (XGMI) Configuration |
|---|---|---|---|---|---|
| AMD Instinct MI355X / AMD Instinct MI350X | Ubuntu 24.04 Kernel 6.14 | RHEL 9.6, RHEL 10.0, Ubuntu 24.04 | ROCm 10.0 | 1, 2*, or 8* | between 8 GPUs |
| AMD Instinct MI325X | Ubuntu 22.04 Kernel 6.8 | Ubuntu 22.04 | ROCm 10.0 | 1 | between 8 GPUs |
| RHEL 10.2** kernel 6.12.0-211 | RHEL 10.2** | ROCm 10.0 | 1 | between 8 GPUs | |
| AMD Instinct MI300X | Ubuntu 22.04 Kernel 6.8, Ubuntu 24.04 Kernel 6.14 | Ubuntu 22.04, Ubuntu 24.04 | ROCm 10.0 | 1 or 8* | between 8 GPUs |
| RHEL 9.4 Kernel 5.14.0-427, RHEL 10.2** kernel 6.12.0-211 |
RHEL 9.4, RHEL 10.2**, Ubuntu 24.04 |
ROCm 10.0 | 1 or 8* | between 8 GPUs | |
| AMD Instinct MI210X | RHEL 9.4 Kernel 5.14.0-427 | RHEL 9.4, Ubuntu 22.04 | ROCm 10.0 | 1 | between 4/8 GPUs |
| AMD Radeon PRO V710 | Ubuntu 24.04 Kernel 6.14 | ROCm: RHEL 9.6, Ubuntu 24.04 Graphics: Ubuntu 24.04 |
ROCm 10.0 amdgpu 6.4.2.2 |
1 to 12 | N/A |
Partition Mode Support
| Number of VFs per GPU | Dynamic Compute Partitioning | NPS1 | NPS2 | NPS4 |
|---|---|---|---|---|
| 1 | SPX (Default) | MI300X, MI325X, MI35XX | ||
| 1 | DPX | MI35XX | ||
| 1 | CPX | MI35XX | ||
| 2* | DPX | MI35XX | ||
| 8* | CPX | MI35XX | MI300X |
For the full list of supported partition modes and the relationship between partition mode and number of VFs per GPU, see the GPU Partitioning.
* Multi-VF support requires a compatible firmware. Please refer to the ROCm Firmware Compatibility Table for the list of required firmware versions and supported configurations.
** RHEL 10.1 is EOLed and replaced by RHEL 10.2
Host AMD SMI Versions
- Library: 64.0.1
- Tool: 37.0.5
Please checkout out SMI doc for more details.
Getting Started
Please check out our User Guide for instructions on how to set up GIM and example configurations to run SR-IOV enabled VMs.
Known Issues
- On VMs with multiple VF assigned, detaching one of the VF (e.g. with virsh detach-device) while a workload is running will cause the guest kernel to refuse further GPU work. The VM need to be restarted before GPU work can be issued again.
- MI350X: In 1 VF mode, there is a small chance VF FLRs will fail due to an SMU timeout if a reset is triggered concurrently with an active GFX workload. The VF FLR will be promoted to a whole GPU reset in these cases. This reset may interrupt workloads running on other VFs in a multi-tenancy setup.