GPUDatacenterReleased Apr 5, 2016

NVIDIA Tesla P100 SXM2 16GB

16 GB HBM2 · 4096-bit · 3,584 cores

The Tesla P100 SXM2 was the first NVLink-connected NVIDIA accelerator and the engine of the DGX-1, which interconnected eight of these mezzanine modules to build what NVIDIA called the world's first deep-learning supercomputer. The GP100 Pascal GPU carries 3,584 CUDA cores and 16 GB of HBM2 stacked memory on a 4,096-bit interface for 732 GB/s, while NVLink provides 160 GB/s of bidirectional interconnect bandwidth, far beyond what PCIe offered at the time. Native half-precision arithmetic at twice the single-precision rate made it the first Tesla designed around deep-learning throughput. Announced at GTC in April 2016.

General

Manufacturer: NVIDIA
Model: Tesla P100 SXM2 16GB
Segment: Datacenter
GPU chip: GP100
Architecture: Pascal
Release date: Apr 5, 2016
Price: Search on Amazon

Memory

Memory size: 16 GB
Memory type: HBM2
Bus width: 4096-bit
Bandwidth: 732 GB/s
Memory clock: 715 MHz

Clocks & silicon

GPU clock: 1,328 MHz
Boost clock: 1,480 MHz
Shader units: 3,584 CUDA Cores
Tensor / AI cores: —
RT cores: —
Transistors: 15.3 billion
Technology (nm): 16 nm
TSMC 16 nm FinFET
TDP / TBP: 300 W

Interface & APIs

Bus interface: SXM2 mezzanine (NVLink)
SLI / multi-GPU: —
DirectX: N/A (compute-only)
OpenGL: N/A (compute-only)
Vulkan: N/A (compute-only)
Other APIs: CUDA 6.0, OpenCL 3.0, OpenACC

AI support

Precisions: FP8FP16BF16INT8INT4TF32FP64
AI accelerator: None (no matrix units)
Peak AI TOPS: —
Notes: No matrix or tensor hardware. Pascal instead runs half precision at twice the single-precision rate on the CUDA cores, which is what made this the first Tesla designed around deep-learning throughput. NVIDIA rates the module at 21.2 TFLOPS FP16, 10.6 TFLOPS FP32 and 5.3 TFLOPS FP64.

All NVIDIA GPUs →

Datacenter gpus from NVIDIA announced around the same time as the NVIDIA Tesla P100 SXM2 16GB. Add any of them to your compare list to see the specs side by side.

gpu