pchardware.org
Categories

NVIDIA · 2018

NVIDIA T4 GPU accelerator

The NVIDIA T4 GPU accelerator is a Turing-based PCIe card released in 2018, designed for data center inference and graphics workloads.

The NVIDIA T4 GPU accelerator, released in September 2018, is based on the Turing architecture with a TU104 chip manufactured on a 12 nm process. It features 2560 shading units and 320 Tensor cores, with base and boost clocks of 585 MHz and 1590 MHz respectively. The card is equipped with 16 GB of GDDR6 memory on a 256-bit bus, providing a memory bandwidth of 320 GB/s. It has a board power (TDP) of 70 W and connects via PCIe 3.0 x16. Notably, it has no display outputs, indicating its use as an accelerator rather than a display adapter. The T4 is intended for server environments, supporting AI inference and virtual GPU workloads.

Specifications

Architecture unverified Turing
GPU chip unverified TU104
Process node unverified 12 nm
Shading units unverified 2560
Tensor / AI cores unverified 320
Base clock unverified 585 MHz
Boost clock unverified 1590 MHz
Memory size unverified 16 GB
Memory type unverified GDDR6
Memory bus unverified 256 bit
Memory bandwidth unverified 320 GB/s
Board power (TDP) unverified 70 W
Bus interface unverified PCIe 3.0 x16
Display outputs unverified No display outputs
Release date unverified 2018-09
Release year unverified 2018