10/10 requests left
Sibacus Workbench
Every response is powered by the Sibacus Transform — bit-shifts and integer adds only. No GPU. No floating-point multiply.
Max 80 tokens per response • 10 requests/min • Powered by Sibacus Transform
Cost & Energy Comparison
Session Energy Saved
0.0 mJ
0 tokens • 0 requests
⚡ Power Draw
BSA (4× Graviton4) 48W
GPU (1× H100) 300W
6.3× less power
💰 Cost / Hour
$0.80
4× Graviton4 spot
vs
$3.50
1× H100 on-demand
4.4× cheaper
🎯 Cost / Million Tokens
$0.15
BSA Performance
vs
$1.50
H100 GPU
10× cheaper per token
🏗️ Hardware Capex
$0
Commodity ARM
vs
$30K
Per H100 GPU
⚙️ Energy per Token
~10.2 µJ
Sibacus BSA
vs
~67 µJ
GPU (H100)
6.6× more efficient
📊 Annual TCO (100 Servers)
BSA Cluster (spot) $65,900
GPU Cluster (H100) $6,112,000
Annual savings $6,046,100
93× cost reduction
Current Model
🇺🇸
Phi-3 Mini
Microsoft • 3.8B params
Compute Shifts + Adds only
Multipliers Used Zero
Quantization K=2 (2.7× compression)
Cooling Required Passive (fanless)
Export Restricted No