Opportunities
Meridian Motion Labs
ValidateActiveEXAScalerAI400X3MNorth AmericaOn-premises
AI training data-loading & checkpoint delays
$950.0K
85
0
0
0
0
Industry
Automotive
Country / Region
United States / North America
Close Date
—
Sales Owner
—
SE Owner
—
Probability (SE judgement)
60%
Quote Status
—
Contract Value
$2.85M
Next Action
Execute benchmark with representative dataset
Owner: Jason Reed · Due: 2026-11-30
Business Challenge & Consequences
Training runs spend significant time on data loading and checkpointing.
GPU idle time; longer model iteration cycles.
Desired Business Outcomes
Reduce data-loading and checkpoint time; increase GPU utilisation.
Compelling Event
Autonomous driving model deadline
Workloads & Solution Direction
AI training and fine-tuningDigital twins and engineering simulation
Primary Direction
parallel-file
Application / Framework
—
Current Environment
Compute / GPU
16× H100 nodes, Slurm
GPU / Node Counts
128× H100
Storage / Data Platform
NFS, bottleneck at scale
Network
200Gb InfiniBand
Workload Characteristics
Data Capacity & Growth
—
Active Working Set
80 TB
Current Performance
12h training run, 50 min checkpoint, GPU util 55%
Target Performance
9h training run, 15 min checkpoint, GPU util 80%
Access Patterns
Sequential large-file, high-concurrency
Read/Write, Concurrency
—
Required Protocols
POSIX, Lustre
Checkpoint / Restart
—
Availability / Recovery
—
Security / Residency
—
Known Technical Blockers
None identified yet
Success Criteria (agreed)
GPU utilisation > 75%; checkpoint < 20 min
Competition & Alternatives
Competitors
Incumbent NAS vendor
Alternatives
Buy more GPUs
