Compare Q4, Q5, Q6 and Q8 model files
See what changes when you select another exact artifact. Compare memory, headroom and runtime availability at your saved context and concurrency.
Your setup: rtx-4090-24gb · 8,192 context tokens · 1 concurrent session. Change setup in the calculator →
What if I change…
Compare the same model across alternatives. These are calculated memory results; runtime support is checked separately.
| Alternative | Artifact | Calculated peak | Headroom | Memory fit / runtime | Try it |
|---|---|---|---|---|---|
| 4-bit artifactExact catalogue files only; precision is not a measured quality score. | Q4_K_M | 6.5 GiB | 17.0 GiB | comfortableRuntime needs verification | Open setup → |
| 5-bit artifactExact catalogue files only; precision is not a measured quality score. | No supported artifact | — | — | No active, verified artifact and runtime binding matches this setup. | Open setup → |
| 6-bit artifactExact catalogue files only; precision is not a measured quality score. | No supported artifact | — | — | No active, verified artifact and runtime binding matches this setup. | Open setup → |
| 8-bit artifactExact catalogue files only; precision is not a measured quality score. | Q8_0 | 9.9 GiB | 13.6 GiB | comfortableRuntime needs verification | Open setup → |
No compatible artifact means unknown here, not that the model can never run. Multiple GPUs, multi-Spark clusters and hybrid offload need a topology-specific calculation.
Same calculator, exact files
Every row uses the planner’s reviewed model, artifact and runtime records. A missing artifact is shown as unsupported rather than estimated from parameter count. Higher precision does not prove better task quality; a runtime compatibility record does not establish speed.
llama.cpp, vLLM, multi-device inference and hybrid offload are not calculated in this surface. No cross-runtime speed scores are mixed.