GPUDatacenterReleased Sep 13, 2016

NVIDIA Tesla P4

8 GB GDDR5 · 256-bit · 2,560 cores

The Tesla P4 was NVIDIA's low-profile inference accelerator, designed to drop into scale-out servers without an auxiliary power connector. A GP104 Pascal GPU provides 2,560 CUDA cores and 8 GB of GDDR5 at 192 GB/s in a single-slot, half-height card drawing up to 75 W, with a 50 W mode available to system builders. Its 8-bit integer instructions gave it 22 TOPS of INT8 throughput, which NVIDIA positioned as a far denser alternative to CPU-only inference. Announced in September 2016 alongside the larger Tesla P40.

General

Manufacturer: NVIDIA
Model: Tesla P4
Segment: Datacenter
GPU chip: GP104
Architecture: Pascal
Release date: Sep 13, 2016
Price: Search on Amazon

Memory

Memory size: 8 GB
Memory type: GDDR5
Bus width: 256-bit
Bandwidth: 192 GB/s
Memory clock: 1,500 MHz

Clocks & silicon

GPU clock: 810 MHz
Boost clock: 1,063 MHz
Shader units: 2,560 CUDA Cores
Tensor / AI cores: —
RT cores: —
Transistors: 7.2 billion
Technology (nm): 16 nm
TSMC 16 nm FinFET
TDP / TBP: 75 W

Interface & APIs

Bus interface: PCIe 3.0 x16
SLI / multi-GPU: —
DirectX: 12 (12_1)
OpenGL: 4.6
Vulkan: 1.3
Other APIs: CUDA Compute Capability 6.1, OpenCL 3.0

AI support

Precisions: FP8FP16BF16INT8INT4TF32FP64
AI accelerator: None (no matrix units)
Peak AI TOPS: 22 TOPS
Notes: No matrix or tensor hardware. Pascal GP104 adds DP4A 8-bit integer dot-product instructions for inference, and NVIDIA rates the card at 22 TOPS INT8 and 5.5 TFLOPS FP32.

All NVIDIA GPUs →

Datacenter gpus from NVIDIA announced around the same time as the NVIDIA Tesla P4. Add any of them to your compare list to see the specs side by side.

gpu