Nistro AI Compute
GPU capacity, on your terms.
NVIDIA and AMD GPUs across clouds and regions. Run on demand or reserve for two or three years, and get a quote for the capacity you need.
- On-demand
- 2 years
- 3 years
- NVIDIAB300288 GB HBM3e · Blackwell UltraThe most memory per GPU in the NVIDIA lineup, for frontier-scale training and long-context inference.View details
- NVIDIAB200180 GB HBM3e · BlackwellBlackwell-generation throughput for large-model training and high-volume inference with FP8 and FP4.View details
- NVIDIAHGX H200141 GB HBM3e · HopperHopper with 141 GB of faster HBM3e per GPU, for larger models and longer contexts than H100.View details
- NVIDIAH100 SXM80 GB HBM3 · HopperThe proven workhorse for distributed training and production inference.View details
- NVIDIAA100 80GB80 GB HBM2e · AmpereA cost-efficient, widely supported choice for training, fine-tuning, and inference.View details
- AMDInstinct MI355X288 GB HBM3E · CDNA 4AMD's flagship accelerator, with 288 GB of memory per GPU on the open ROCm stack.View details
- AMDInstinct MI325X256 GB HBM3E · CDNA 3256 GB of HBM3E per GPU on CDNA 3, for memory-hungry inference and training on ROCm.View details
- AMDInstinct MI300X192 GB HBM3 · CDNA 3A proven 192 GB accelerator for serving large models on a single GPU.View details
- NVIDIARTX PRO 600096 GB GDDR7 · BlackwellA 96 GB Blackwell card for single-GPU inference, fine-tuning, and visual computing.View details
- NVIDIARTX 509032 GB GDDR7 · BlackwellFast, affordable Blackwell compute for small-model inference and generative media.View details
- NVIDIARTX 409024 GB GDDR6X · Ada LovelaceA popular, cost-effective card for inference and image generation.View details
- NVIDIARTX 309024 GB GDDR6X · AmpereThe lowest-cost way to get 24 GB of GPU memory for prototyping and batch jobs.View details