NVIDIA H100
The NVIDIA H100 is a Data Center graphics card based on the Hopper architecture. Released on 2023-03-21 with a launch price of $30,000.00, it features 14,592 CUDA cores and 138 GB of HBM3 memory.
General info about H100
| Architecture | Hopper (GH100) |
| Market Segment | Data Center |
| Release Date | 2023-03-21 |
| Launch Price | $30,000.00 |
| Manufacturing Process | 4nm |
Technical specs of H100
| CUDA Cores | 14,592 |
| Base Clock Speed | 1410 MHz |
| Boost Clock Speed | 1830 MHz |
| Transistor Count | 80B |
| VRAM Capacity | 138 GB HBM3 |
| Memory Bus Width | 5120 bits |
| Memory Bandwidth | 3000.0 GB/s |
| TDP | 700 W |
Key Technical Parameters
Key technical parameters of the NVIDIA H100 include its 14,592 CUDA cores, 1410 MHz base clock, and 1830 MHz boost clock, delivering high performance for data center applications.
Benchmark performance of H100
FP32 Performance
67
TFLOPS
FP16 Performance
2000
TFLOPS
The NVIDIA H100 achieves 67 TFLOPS in FP32 performance, making it ideal for data center workloads.
Gaming performance of H100
Gaming Performance
The NVIDIA H100 is not designed for gaming. It is optimized for AI training, machine learning, and high-performance computing workloads.
Features and connectivity of H100
Tensor Cores
Gen 4
Transformer Engine
Enabled
Flash Attention
Supported
Connectivity
NVLink 4, PCIe Gen5
Physical dimensions of H100
| Length | 10.5in |
| Width | 4.4in |
| Height | 2-slot |
Rankings and efficiency of H100
| Ranking Position | #1 |
| Popularity Ranking | #5 |
| Cost Effectiveness | 0.8/5 |
| Power Efficiency | 1.2/5 |
Cloud Availability
Available Instances
1
Starting Price
$2.80/hr
Top GPUs for Training and Inference
| Category | Rank 1 | Rank 2 | Rank 3 |
|---|---|---|---|
| Best for Training | NVIDIA H200 | NVIDIA H100 | NVIDIA B200 |
| Best for Inference | NVIDIA A40 | NVIDIA A100 | NVIDIA A10 |
Compare GPU specifications and cloud instances to find the best GPU for your workload.