Available GPUs

Private Cloud clusters are available with the latest NVIDIA data center GPUs, optimized for AI training and inference at scale.

GPU Options

The industry standard for large-scale AI training.

Spec	Value
GPU Memory	80 GB HBM3
Memory Bandwidth	3.35 TB/s
FP16 Performance	989 TFLOPS
Interconnect	NVLink 4.0 (900 GB/s)
Best For	LLM training, fine-tuning, high-throughput inference

Enhanced H100 with more memory and bandwidth for memory-bound workloads.

Spec	Value
GPU Memory	141 GB HBM3e
Memory Bandwidth	4.8 TB/s
FP16 Performance	989 TFLOPS
Interconnect	NVLink 4.0 (900 GB/s)
Best For	Large model inference, long-context training, models that need more VRAM

Next-generation Blackwell architecture with massive compute gains.

Spec	Value
GPU Memory	192 GB HBM3e
Memory Bandwidth	8 TB/s
FP16 Performance	2,250 TFLOPS
FP4 Performance	9,000 TFLOPS
Interconnect	NVLink 5.0 (1,800 GB/s)
Best For	Frontier model training, next-gen inference, FP4 quantized workloads

The latest Blackwell Ultra with maximum memory and performance.

Spec	Value
GPU Memory	288 GB HBM3e
Memory Bandwidth	12 TB/s
FP16 Performance	2,250 TFLOPS
FP4 Performance	9,000 TFLOPS
Interconnect	NVLink 5.0 (1,800 GB/s)
Best For	Largest-scale training, trillion-parameter models, maximum memory capacity

GPU	Memory	Best For	Availability
H100	80 GB	General AI training, proven and widely supported	High
H200	141 GB	Memory-hungry models, large batch inference	Moderate
B200	192 GB	Next-gen training, 2x compute over H100	Growing
B300	288 GB	Maximum scale, highest memory capacity	Limited

Not sure which GPU is right for your workload? Contact us at support@runcrate.ai and our team will help you choose the optimal configuration.

All Private Cloud clusters include high-speed interconnect: