Energy Efficiency (TOPS/W)

• The Nvidia H100 GPUs have an INT8 Tensor Core performance of up to 3,958 TOPS
• The power consumption of the H100 GPUs can be configured up to 700 watts

$$ TOPS \text{ } W = \frac{TOPS}{Watts} = \frac{3958 \text{ } TOPS}{700 \text{ } Watts} = 5.654 \text{ } TOPS \text{ } W $$

HRS/LRS (Memristors)

$$ \frac{37000000}{826} $$

Resistive RAM (memristor) Modeling and In-memory Computing using Majority Logic

Power Used for Memory

Target: 50% Reduction In Memory Power

Typical Power Consumption of PC Components - Power Draw in Watts

Accuracy (AI)

NeuroPack and ArC One

Space Required for Equal Performance

https://www.sciencedirect.com/science/article/pii/S1319157817302963

Cost of running a data center for 10 hours (only computational cost)

Cost Reduction for Data Center Usage