Pick the model.See the GPUs that proved it.
This is not a static VRAM chart. Choose a workload and we rank GPUs by real benchmark evidence, trust status, and hardware fit.
Tested GPUs
57
Consumer, pro, and datacenter paths
Model presets
8
Mapped to actual workloads in the benchmark DB
Freshest evidence
N/A
Latest tested date among ranked devices
Choose a target model
Selected workload
Llama 3 8B Q4
The mainstream local-LLM baseline. Great for chat, RAG, and lightweight agents.
Llama 3 8B Q4
8GB works, 12GB+ feels comfortable.
general
0 tested
Recommendation logic
We rank devices only when records match the workload, runtime version, quantization, context and batch size. Missing comparison evidence produces an empty ranking.
Decision shortcuts
Search your GPU and we'll show whether this specific model target is already benchmarked on it.
Ranked hardware