LocalOps

Simulation Lab

Test theoretical model performance on your hardware.

7B
0.1B700B
7B

For dense models, keep this equal to total params. For MoE models, set to active expert count.

My Custom Model

Bottlenecked
VRAM Required17.7 GB
Est. Speed42.4 T/s
KV Cache12.51 GB
Disk Space4.2 GB
32% of layers will offload to system RAM. Expect slower performance.
Selected GPU:NVIDIA RTX 4070 Ti
Available VRAM:12GB
Memory Bandwidth:504 GB/s
Quantization:4-bit Medium (Recommended)