Compatibility Matrix
This matrix shows how many catalog profiles can run at each VRAM tier, split by workload scenario. Use it as a planning shortcut before checking individual model pages.
| Scenario | 8GB tier | 12GB tier | 16GB tier | 24GB tier | 48GB tier |
|---|---|---|---|---|---|
| Chat | 71 | 102 | 122 | 173 | 205 |
| Coding | 45 | 66 | 81 | 120 | 135 |
| Multimodal | 33 | 48 | 59 | 84 | 94 |
| Reasoning | 16 | 31 | 41 | 65 | 80 |
| Embedding | 12 | 13 | 13 | 14 | 14 |
How to use this matrix
- Pick your actual VRAM ceiling first.
- Find the scenario row that matches your workload.
- Open model pages from that subset and compare measured signals.
Compatibility FAQ
- What does a compatible model count mean?
- A compatible count means the catalog profile has a minimum VRAM requirement at or below that tier. It is a planning filter, not a guarantee of quality, latency, or context length.
- Which VRAM tier should I start with?
- Start with your physical GPU VRAM ceiling, then open model pages to compare quantization, measured status, and workload fit before choosing a runtime.
- Why do some workloads have fewer compatible models?
- Coding, reasoning, multimodal, and embedding workloads can have different model families and memory profiles, so the compatible count changes by scenario.