V100

V100 refers to Nvidia Tesla V100, Nvidia’s flagship data-center graphics processing unit based on the Volta architecture. Introduced in 2017, it is designed for AI training and inference, high-performance computing, and data analytics, offering high throughput for both traditional HPC workloads and deep learning tasks.

The V100 centers on the GV100 GPU, which features 5,120 CUDA cores and 640 Tensor Cores dedicated to accelerating mixed-precision matrix operations. Memory is delivered via high-bandwidth memory 2 (HBM2) in 16 GB or 32 GB configurations, providing strong bandwidth for large datasets and models. The device is manufactured on a 12 nm FinFET process and can be deployed in PCIe 3.0 form factors or in NVLink-connected SXM2 modules to enable multi-GPU configurations.

In terms of performance, the V100 offers up to about 14 TFLOPS of FP32 performance, up to 112 TFLOPS for FP16 calculations when using Tensor Cores, and around 7 TFLOPS for FP64. Tensor Cores are a defining feature, delivering substantial acceleration for deep learning workloads, particularly for training large neural networks and performing inference with mixed precision.

Deployment and impact: the V100 served as Nvidia’s flagship Volta accelerator and saw widespread use in data centers, research HPC clusters, and enterprise AI environments. It was eventually succeeded by newer Nvidia data-center GPUs, such as the Ampere-based A100, but remains a notable milestone in the progression of GPU-accelerated computing.