GPU FLEET
OVERBOOKING CONTROL
Real-time provisioning intelligence across 392 physical GPUs. Model overcommitment ratios and predict contention risk.
OVERBOOKING RATIO2×
1×2×3×4×5×
2× (Conservative)
PHYSICAL GPUs
392
Across all models
VIRTUAL SLOTS
784
+392 overcommit
FLEET VRAM
35.7 TB
Total HBM capacity
AVG UTILISATION
73%
24h rolling average
NVIDIA
H200 SXM
Hopper · 141 GB VRAM · 700W TDP
PHYSICAL32
VIRTUAL64
OVERCOMMIT+32
EFF. UTIL100.0%
CAPACITY UTILISATION100.0%
CRITICAL$8.5/hr · $362/hr est.
NVIDIA
H100 SXM
Hopper · 80 GB VRAM · 700W TDP
PHYSICAL64
VIRTUAL128
OVERCOMMIT+64
EFF. UTIL100.0%
CAPACITY UTILISATION100.0%
CRITICAL$5.5/hr · $453/hr est.
AMD
MI325X
CDNA 3 · 256 GB VRAM · 750W TDP
PHYSICAL24
VIRTUAL48
OVERCOMMIT+24
EFF. UTIL100.0%
CAPACITY UTILISATION100.0%
CRITICAL$7.8/hr · $257/hr est.
AMD
MI300X
CDNA 3 · 192 GB VRAM · 750W TDP
PHYSICAL48
VIRTUAL96
OVERCOMMIT+48
EFF. UTIL100.0%
CAPACITY UTILISATION100.0%
CRITICAL$6.2/hr · $391/hr est.
NVIDIA
L40
Ada Lovelace · 48 GB VRAM · 300W TDP
PHYSICAL96
VIRTUAL192
OVERCOMMIT+96
EFF. UTIL100.0%
CAPACITY UTILISATION100.0%
CRITICAL$2.2/hr · $265/hr est.
NVIDIA
RTX 6000 Ada
Ada Lovelace · 48 GB VRAM · 300W TDP
PHYSICAL128
VIRTUAL256
OVERCOMMIT+128
EFF. UTIL100.0%
CAPACITY UTILISATION100.0%
CRITICAL$1.8/hr · $282/hr est.
PHYSICAL vs OVERCOMMIT CAPACITY
Stacked view of physical GPU count (solid) and overcommitted virtual slots (red) at 2× ratio
SIMULATION DATA · UTILISATION PATTERNS BASED ON REALISTIC AI WORKLOAD PROFILES · REVENUE ESTIMATES EXCLUDE NETWORK & STORAGE