GPU Configuration
Select from high-performance GPU options based on your computational needs
Select from our high-performance GPU options based on your computational needs.
Available GPU Types
Choose the GPU that best matches your workload requirements:
GPU Model | Performance | Best For |
---|---|---|
H100-SXM5-80GB | Highest performance for large-scale AI training | Large language models, complex AI research |
H100-PCIe-NVLink-80GB | High-performance with NVLink | Multi-GPU workloads, distributed training |
H100-PCIe-80GB | High-performance PCIe interface | Single-GPU inference, model training |
A100-SXM4-80GB-NVLink | Excellent for deep learning and HPC | Deep learning, scientific computing |
A100-PCIe-80GB | High-performance PCIe A100 | ML training, data analytics |
L40 | Balanced performance for diverse workloads | AI inference, graphics workloads |
RTX-A6000 | Cost-effective for smaller workloads | Development, small-scale training |
A40 | Versatile professional GPU | Professional graphics, AI development |
GPU Instance Pricing
USD Pricing
GPU Model | USD per Hour | Performance Level |
---|---|---|
H100-SXM5-80GB | $2.69 | Highest |
H100-PCIe-NVLink-80GB | $2.29 | Very High |
H100-PCIe-80GB | $2.25 | Very High |
A100-SXM4-80GB-NVLink | $1.59 | High |
A100-PCIe-80GB | $1.55 | High |
L40 | $1.19 | Medium-High |
RTX-A6000 | $0.69 | Medium |
A40 | $0.69 | Medium |
EUR Pricing
GPU Model | EUR per Hour | Performance Level |
---|---|---|
H100-SXM5-80GB | €2.49 | Highest |
H100-PCIe-NVLink-80GB | €2.09 | Very High |
H100-PCIe-80GB | €2.05 | Very High |
A100-SXM4-80GB-NVLink | €1.45 | High |
A100-PCIe-80GB | €1.39 | High |
L40 | €1.09 | Medium-High |
RTX-A6000 | €0.45 | Medium |
A40 | €0.45 | Medium |
INR Pricing
GPU Model | INR per Hour | Performance Level |
---|---|---|
H100-SXM5-80GB | ₹239 | Highest |
H100-PCIe-NVLink-80GB | ₹199 | Very High |
H100-PCIe-80GB | ₹195 | Very High |
A100-SXM4-80GB-NVLink | ₹135 | High |
A100-PCIe-80GB | ₹135 | High |
L40 | ₹99 | Medium-High |
RTX-A6000 | ₹49 | Medium |
A40 | ₹49 | Medium |
GPU Detailed Specifications
NVIDIA H100 GPUs
H100-SXM5-80GB
- Memory: 80GB HBM3
- Architecture: Hopper
- Best for: Large language models, transformer training
- Multi-Instance GPU: Yes
- NVLink: 900 GB/s
H100-PCIe-NVLink-80GB
- Memory: 80GB HBM3
- Interface: PCIe with NVLink
- Best for: Multi-GPU distributed training
- Interconnect: High-bandwidth NVLink
H100-PCIe-80GB
- Memory: 80GB HBM3
- Interface: PCIe 5.0
- Best for: Single-GPU inference and training
- Power Efficiency: Optimized for single-node workloads
NVIDIA H100 GPUs
H100-SXM5-80GB
- Memory: 80GB HBM3
- Architecture: Hopper
- Best for: Large language models, transformer training
- Multi-Instance GPU: Yes
- NVLink: 900 GB/s
H100-PCIe-NVLink-80GB
- Memory: 80GB HBM3
- Interface: PCIe with NVLink
- Best for: Multi-GPU distributed training
- Interconnect: High-bandwidth NVLink
H100-PCIe-80GB
- Memory: 80GB HBM3
- Interface: PCIe 5.0
- Best for: Single-GPU inference and training
- Power Efficiency: Optimized for single-node workloads
NVIDIA A100 GPUs
A100-SXM4-80GB-NVLink
- Memory: 80GB HBM2e
- Architecture: Ampere
- Best for: Deep learning, scientific computing
- Multi-Instance GPU: Yes
- NVLink: 600 GB/s
A100-PCIe-80GB
- Memory: 80GB HBM2e
- Interface: PCIe 4.0
- Best for: ML training, data analytics
- Tensor Cores: 3rd generation
Professional Workstation GPUs
L40
- Memory: 48GB GDDR6
- Architecture: Ada Lovelace
- Best for: AI inference, graphics workloads
- RT Cores: 3rd generation
- Tensor Cores: 4th generation
RTX A6000
- Memory: 48GB GDDR6
- Architecture: Ampere
- Best for: Development, visualization
- RT Cores: 2nd generation
- CUDA Cores: 10,752
A40
- Memory: 48GB GDDR6
- Architecture: Ampere
- Best for: Professional graphics, AI development
- Multi-Instance GPU: Yes
GPU Count Selection
Choose the number of GPUs for your virtual machine:
Single GPU
Perfect for:
- Development and testing
- Small to medium model training
- Inference workloads
- Cost-effective computing
Multi-GPU
Ideal for:
- Large model training
- Distributed computing
- High-throughput inference
- Parallel processing workloads
Multi-GPU Configuration
- Available counts are dynamically updated based on current stock
- Higher GPU counts provide more computational power for parallel workloads
- Multi-GPU setups automatically include NVLink for supported GPU types
- GPU availability varies by region and is updated in real-time
NVLink is automatically included for multi-GPU configurations with compatible GPU types (H100 and A100 series with NVLink variants).
GPU Selection Guide
By Use Case
Large Language Models (LLMs)
Large Language Models (LLMs)
Recommended GPUs:
- H100-SXM5-80GB - Best performance for massive models
- H100-PCIe-80GB - High performance for large models
- A100-SXM4-80GB-NVLink - Excellent for most LLM training
Considerations:
- Memory capacity is crucial for large models
- Multi-GPU setups for distributed training
- NVLink for efficient multi-GPU communication
Deep Learning Research
Deep Learning Research
Recommended GPUs:
- H100 Series - Cutting-edge performance
- A100 Series - Proven performance for research
- L40 - Balanced performance and cost
Considerations:
- Tensor Core performance for mixed precision
- Memory bandwidth for large datasets
- Multi-Instance GPU for resource sharing
AI Inference
AI Inference
Recommended GPUs:
- L40 - Optimized for inference workloads
- RTX A6000 - Cost-effective for inference
- A40 - Professional-grade inference
Considerations:
- Lower latency requirements
- Batch processing capabilities
- Cost optimization for production
Development & Prototyping
Development & Prototyping
Recommended GPUs:
- RTX A6000 - Excellent for development
- A40 - Professional development environment
- L40 - Balanced development platform
Considerations:
- Cost-effective for iterative development
- Sufficient memory for model experimentation
- Good performance for rapid prototyping
Graphics & Visualization
Graphics & Visualization
Recommended GPUs:
- RTX A6000 - Professional graphics workloads
- A40 - High-end visualization
- L40 - Graphics and AI combined workloads
Considerations:
- RT Cores for real-time ray tracing
- CUDA cores for compute workloads
- Professional drivers and support
By Performance Tier
H100 Series
- Latest architecture with highest performance
- 80GB memory for the largest models
- Advanced Tensor Cores for AI workloads
- Best for cutting-edge research and production
Best for: Large-scale AI training, advanced research, production LLM inference
H100 Series
- Latest architecture with highest performance
- 80GB memory for the largest models
- Advanced Tensor Cores for AI workloads
- Best for cutting-edge research and production
Best for: Large-scale AI training, advanced research, production LLM inference
A100 Series
- Proven performance for AI workloads
- 80GB memory variants available
- Multi-Instance GPU capabilities
- Excellent price-performance ratio
Best for: Deep learning, scientific computing, multi-user environments
L40
- Modern architecture with good performance
- 48GB memory for most workloads
- Combined AI and graphics capabilities
- Good balance of performance and cost
Best for: AI inference, mixed workloads, development
RTX A6000 / A40
- Professional-grade performance
- 48GB memory capacity
- Cost-effective for development
- Professional software support
Best for: Development, small-scale training, visualization
Regional Availability
GPU availability and selection varies by region:
NORWAY-1
Europe Region
- All GPU types available
- Low latency for European users
- GDPR compliant infrastructure
CANADA-1
North America Region
- All GPU types available
- High-speed connectivity
- Optimized for North American users
US-1
United States Region
- All GPU types available
- US-based infrastructure
- Low latency for US users
GPU availability is updated in real-time. If your preferred GPU type is not available in your selected region, try another region or check back later.
GPU Selection Tips
Performance Optimization
Memory Considerations
Key Factors:
- Model size requirements
- Batch size optimization
- Dataset memory usage
- Multi-model deployment
Compute Requirements
Key Factors:
- Training time constraints
- Inference latency needs
- Parallel processing requirements
- Throughput expectations
Cost Optimization
Development vs Production
Development vs Production
Development:
- Use cost-effective GPUs (RTX A6000, A40)
- Single GPU configurations
- Pay-per-minute billing for short experiments
Production:
- Higher-performance GPUs for better efficiency
- Multi-GPU for scaling
- Consider total cost of ownership
Workload Matching
Workload Matching
Training Workloads:
- Higher-end GPUs for faster training
- Multi-GPU for large models
- Consider training time vs. cost trade-offs
Inference Workloads:
- Optimize for latency and throughput
- Lower-cost GPUs may be sufficient
- Batch processing for efficiency
Getting Started
Assess Your Needs
- Determine your primary use case
- Estimate memory requirements
- Consider performance needs
- Plan your budget
Select GPU Type
- Choose based on workload requirements
- Consider regional availability
- Review pricing for your currency
- Start with single GPU and scale as needed
Deploy and Test
- Deploy VM with selected GPU
- Test performance with your workload
- Monitor resource utilization
- Optimize configuration as needed
Scale and Optimize
- Add more GPUs if needed
- Optimize software for multi-GPU
- Monitor costs and performance
- Adjust configuration based on results
Start with a single GPU to test your workload, then scale to multi-GPU configurations as needed. This approach helps optimize both performance and costs.