Quantum GPU & LLM Optimization Visualizer

Model & Topology Config Dual RTX 3060

Target LLM Architecture

Partitioning Algorithm

Transformer Layers (N) 28

VRAM per Layer (MB) 200 MB

GPU 0 Thermal Pressure 47°C

GPU 1 Thermal Pressure 43°C

PCIe Transfer Weight 10.0 MB/cut

⚛ Quantum Objective: Min H(x) = ∑ w_uv(x_u - x_v)² + λ(∑ x_i · v_i - C)² + Thermal_Bias

📷 Isometric 3D View

GPU 0 (Primary)

GPU 1 (Cooler Target)

PCIe Cut Activation

Real-Time Performance QUBO Optimal

0.94 ms

Slicing Latency

632.2

Prompt Tok/s

65.76

Gen Tok/s

-8.75

QUBO Energy Cost

GPU 0 VRAM (0 layers) 0.0 / 12.2 GB

GPU 1 VRAM (28 layers) 5.5 / 12.2 GB

2D Hamiltonian Energy Landscape Min Ground State H(x)

Inference Insight: GPU 1 is running 4°C cooler. QUBO dynamically shifted 28 layers (5.5GB) to GPU 1, eliminating cross-GPU PCIe bus transfer penalties.