Commit ec526bd
feat(mcv): Add unified container with NVIDIA and AMD support
Add mcv:unified container variant that includes both NVIDIA (CUDA/NVML)
and AMD (ROCm) GPU support. The container automatically detects the GPU
vendor at runtime, simplifying deployment across mixed GPU clusters.
Key changes:
- New mcv-unified target in amd64.dockerfile (CUDA 12.6.3 + ROCm 6.2.4)
- Updated Makefile: 'make image-unified' builds unified image
- Unified image tagged as both :unified and :latest (~1.2 GB)
- Comprehensive documentation with Kubernetes examples
- Auto-detection via MCV client library (dlopen for NVML, CLI for ROCm)
Runtime behavior:
- NVIDIA nodes: Uses NVML (libnvidia-ml.so.1) for GPU detection
- AMD nodes: Uses rocm-smi for GPU detection
- CPU nodes: Gracefully falls back to no-GPU mode
Container variants now available:
- mcv:unified (NEW DEFAULT) - NVIDIA + AMD support (~1.2 GB)
- mcv:minimal - No GPU libs (~176 MB)
- mcv:nvidia - NVIDIA only (~356 MB)
- mcv:amd - AMD only (~923 MB)
This change is backward-compatible. Existing mcv:nvidia and mcv:amd
deployments continue to work unchanged.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Signed-off-by: Maryam Tahhan <mtahhan@redhat.com>1 parent 947823f commit ec526bd
4 files changed
Lines changed: 523 additions & 8 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
139 | 139 | | |
140 | 140 | | |
141 | 141 | | |
142 | | - | |
| 142 | + | |
143 | 143 | | |
144 | 144 | | |
145 | 145 | | |
| |||
148 | 148 | | |
149 | 149 | | |
150 | 150 | | |
151 | | - | |
152 | 151 | | |
153 | 152 | | |
154 | 153 | | |
| |||
178 | 177 | | |
179 | 178 | | |
180 | 179 | | |
| 180 | + | |
| 181 | + | |
| 182 | + | |
| 183 | + | |
| 184 | + | |
| 185 | + | |
| 186 | + | |
| 187 | + | |
| 188 | + | |
| 189 | + | |
| 190 | + | |
| 191 | + | |
| 192 | + | |
| 193 | + | |
181 | 194 | | |
182 | | - | |
| 195 | + | |
183 | 196 | | |
184 | 197 | | |
185 | 198 | | |
| |||
0 commit comments