Buradasın
NVIDIA Ampere Architecture Features
nvidia.com/en-gb/data-center/ampere-architecture/Yapay zekadan makale özeti
- Core Technology
- Largest 7nm chip with 54 billion transistors
- Features third-generation Tensor Cores with TF32 and FP64 precision
- Enables Multi-Instance GPU (MIG) for partitioned workloads
- Includes third-generation NVLink with 600 GB/s bandwidth
- Performance Enhancements
- Provides up to 2X higher performance for sparse AI models
- Offers second-generation RT Cores for photorealistic rendering
- Features 2 TB/sec memory bandwidth and 40 MB level 2 cache
- Supports structural sparsity for improved model training
- Scalability and Security
- A2 GPU operates at 40W power for entry-level servers
- NVIDIA converged accelerators combine compute and network acceleration
- Includes secure boot and rollback protections
- Enables virtual PC access with quad-GPU design