GPUDatacenterReleased Jun 21, 2017

NVIDIA Tesla V100 PCIe 16GB

16 GB HBM2 · 4096-bit · 5,120 cores

The PCI Express add-in card version of NVIDIA's Volta data center accelerator, for standard servers that cannot take an SXM2 module. Volta introduced Tensor Cores, which made mixed-precision deep learning training practical, and this card became a mainstay of research clusters and cloud instances. It runs at lower clocks and a lower power limit than the SXM2 module and has no NVLink, and as a compute-only product it has no display outputs. Released in June 2017; a 32 GB version followed in 2018.

General

Manufacturer: NVIDIA
Model: Tesla V100 PCIe 16GB
Segment: Datacenter
GPU chip: GV100
Architecture: Volta
Release date: Jun 21, 2017
Price: Search on Amazon

Memory

Memory size: 16 GB
Memory type: HBM2
Bus width: 4096-bit
Bandwidth: 900 GB/s
Memory clock: 875 MHz

Clocks & silicon

GPU clock: 1,245 MHz
Boost clock: 1,370 MHz
Shader units: 5,120 CUDA Cores
Tensor / AI cores: 640
RT cores: —
Transistors: 21.1 billion
Technology (nm): 12 nm
TSMC 12 nm FFN
TDP / TBP: 250 W

Interface & APIs

Bus interface: PCIe 3.0 x16
SLI / multi-GPU: None (no NVLink)
DirectX: N/A (compute-only)
OpenGL: N/A (compute-only)
Vulkan: N/A (compute-only)
Other APIs: CUDA 7.0, OpenCL 3.0

AI support

Precisions: FP8FP16BF16INT8INT4TF32FP64
AI accelerator: 1st Gen Tensor Cores (640)
Peak AI TOPS: —
Notes: First-generation Tensor Cores do FP16 multiply with FP32 accumulate for up to 112 TFLOPS of deep learning throughput. FP64 runs at half rate (7 TFLOPS) and FP32 at 14 TFLOPS. Volta has no BFLOAT16, TF32, FP8 or structured sparsity.

All NVIDIA GPUs →

Datacenter gpus from NVIDIA announced around the same time as the NVIDIA Tesla V100 PCIe 16GB. Add any of them to your compare list to see the specs side by side.

gpu