# Enterprise AI with H100 GPUs

Deploy NVIDIA H100 Tensor Core GPUs with Intel TDX + NVIDIA Confidential Computing. 80GB HBM3 memory, proven performance, complete hardware protection.

## Specifications

- **80GB** HBM3 Memory  
- **3.35 TB/s** Bandwidth  
- **9x Faster vs A100**  
- **Full TEE** Intel + NVIDIA

### Pricing

- Starting at $3.08/GPU/hr on-demand  
- $2.38/GPU/hr with 12-month commitment (23% savings)

## Proven Enterprise Performance

MLPerf-certified AI acceleration with Full-Stack TEE security

- **21.8K tokens/sec** Llama 2 70B Inference  
- **700W Thermal Design Power**

### Full-Stack TEE

- Intel TDX + NVIDIA Confidential Computing protection with cryptographic attestation  
- ✓ Hardware-enforced isolation  
- ✓ <5% performance overhead

### MLPerf Certified

- GPT-3 Training Baseline  
- ResNet-50 3.9x vs V100  
- Available Now in US-East & US-West

## H100 vs H200

Official benchmarks from MLPerf v5.0 and NVIDIA Technical Labs

| Metric                       | H100   | H200   | B200   |
|------------------------------|--------|--------|--------|
| Tokens per second            | 21.8K  | 33.0K  | 75.0K  |
| Inference Performance         | Baseline | 1.51x  | 3.4x   |

### Source

Source: MLPerf Inference v5.0 (2024)

## H100 Pricing

Flexible pricing for any workload. All prices include Full-Stack TEE protection.

### On-Demand
- Pay only for what you use  
- $3.08/GPU/hr  
- 24-hour minimum billing

### Reserved
- $2.38/GPU/hr  
- Best long-term value

## Available H100 Configurations

### 1x H100
- Memory: 80GB HBM3  
- Bandwidth: 3 TB/s  
- TEE Protection: Full-Stack TEE  
- Perfect for enterprise AI workloads.

### 2x H100
- Memory: 160GB HBM3  
- Bandwidth: 6 TB/s  
- TEE Protection: Full-Stack TEE  
- Perfect for enterprise AI workloads.

## Full-Stack TEE Architecture

- Complete hardware protection from CPU to GPU. 
- Intel TDX protects CPU, memory, and VM from host access.  
- NVIDIA CC encrypts all GPU memory.

### End-to-End Protection

- Data encrypted in transit (TLS), at rest (AES-256), and during processing (TEE).

### Compliance Ready

- SOC 2 Type I certified, HIPAA compliant.

## What You Can Build with GPU TEE

Real-world applications running on Phala Cloud with complete Intel TDX + NVIDIA Confidential Computing protection

1. Train and deploy models on sensitive healthcare, financial, or legal data  
2. Build autonomous AI agents managing cryptographic keys and digital assets.  
3. Accelerate zkVM and zkRollup proof generation with GPU TEE.  
4. Use GPU TEE as 2FA for FHE and MPC systems.  
5. Combine ZK proofs with TEE attestation for double security.

## Three Ways to Deploy GPU TEE

- **CVM + GPU**: Full control, deploy your own Docker containers  
- **Confidential AI Models**: Instant deployment with pre-configured templates  
- **Enterprise Solutions**: White-glove service with dedicated clusters and custom SLAs

### CVM + GPU: Maximum Flexibility

- Deploy your own Docker containers with SSH access to TEE-protected GPUs.