Eight-GPU training
Evaluate scaling with your model, sequence length and batch size.
8 × H100 SXM rental
Eight GPUs with 80 GB each for distributed training and inference.
$25.536/ hour
Workloads for this GPU configuration
Distributed inference
Choose model sharding around per-GPU memory and interconnect requirements.
Multiple research workloads
Allocate resources across independent experiments with explicit GPU assignment.
Software compatibility
- Memory is per card: 8 × 80 GB HBM3. It is not automatically pooled.
- Eight H100 GPUs with 80 GB each are not one card with a contiguous 640 GB allocation.
- Keeping eight GPUs busy also depends on CPU resources, data, storage and runtime settings.
Rent 8 × H100 SXM
Provide your model, software environment and rental period. We will confirm the rate, availability and connection method before the rental starts.
$25.536/ hour
About this configuration
Is memory pooled across all GPUs?
Memory is per card: 8 × 80 GB HBM3. It is not automatically pooled.
Does my model need all eight H100 GPUs?
Find the smallest device count that fits the model and working data, then compare speed and utilization. More GPUs may not help if data delivery or communication is the bottleneck.
How do I request this configuration?
This configuration is open for rental requests. Share your workload, preferred dates and duration. Hardware availability, pricing and access dates are confirmed individually; a request does not create a confirmed booking.
