AXIOM Calculator
Calculate Your Compression
See how much memory and cost you save with AXIOM's 15.7x memory compression. Based on verified Llama-2 70B benchmark (Mar 2026).
Configuration
34.4B parameters
164
1.7M GPU-hours — 1.7M hours (Meta disclosed)
Total VRAM:80 GB
AdamW needs:6x H100 80GB
AXIOM needs:1x H100 80GB
This model won't fit with AdamW, but WILL fit with AXIOM! AdamW needs 6 GPUs, AXIOM needs only 1.
15x
Total Compression
384.9 GB
Memory Saved
$29.8M
Cost Saved
0 GB
Optimizer Memory
5950.0 MWh
Energy Saved
2499.0 t
CO₂ Reduced
Memory Breakdown
AdamW412.8 GB (needs 6 GPUs)
AXIOM27.9 GB ✓
Your VRAM: 80 GB (1x H100 80GB)
| Component | AdamW | AXIOM | Compression |
|---|---|---|---|
| Model Weights | 68.8 GB | 25.8 GB | 2.7x |
| Gradients | 68.8 GB | 2.1 GB | 33x |
| Optimizer (m+v) | 275.2 GB | 0 GB | ∞ |
| Total | 412.8 GB | 27.9 GB | 15x |
Cost Analysis (1.7Mh training)
With AdamW (6x H100 80GB)
$35.7M
$21.00/hr × 1.7Mh
With AXIOM (1x H100 80GB)
$6.0M
$3.50/hr × 1.7Mh
Total Savings$29.8M (83%)
Cost based on GPU requirements — AXIOM lets you use fewer GPUs for the same model.
Environmental Impact
AdamW Energy
7140.0 MWh
2998.8 tonnes CO₂
AXIOM Energy
1190.0 MWh
499.8 tonnes CO₂
Carbon Reduction2499.0 tonnes CO₂
119,000
Trees/year equivalent
6,185,644
Car miles avoided
Ready to train larger models?
Get started with AXIOM in under a minute.