Tesla T4G

Turing Tesla T Series Data Center Server GDDR6 2018 NVIDIA

Key Metrics

Architecture: Turing
Launch Date: 2018-09-13
CUDA Cores: 2,560
Memory: 16 GB GDDR6
Memory Bandwidth: 320 GB/s
TDP/TBP: 70 W
Interface: PCIe 3.0 x16

Save this GPU, then open another GPU page to compare them side by side.

GPU Specs

Identity
Entry Type	Data Center GPU
Official Name	NVIDIA Tesla T4G
GPU Codename	TU104
Architecture	Turing
Launch Date	2018-09-13
Market Segment Notes	Data Center in Tesla T Series
Core Specs
GPU Die	TU104
Chip Variant / Package	BGA-2228
Process Node	TSMC 12 nm
Transistor Count	13.6 billion
Die Size	545 mm²
CUDA Cores	2560
SM Count	40
TMUs	160
ROPs	64
AI Units / Tensor Cores	320
RT Units / RT Cores	40
Base Clock	585 MHz
Boost Clock	1,590 MHz
Pixel Fillrate	101.8 GP/s
Texture Fillrate	254.4 GT/s
FP32 Throughput	8.14 TFLOPS
FP16 Throughput	65.13 TFLOPS
FP64 Throughput	0.25 TFLOPS
Memory
Memory Size	16 GB
Memory Type	GDDR6
Memory Bus Width	256-bit
Memory Clock	1,250 MHz
Memory Bandwidth	320 GB/s
L2 Cache	4 MB
L1 Cache	64 KB
Board and Display
TDP / TBP	70 W
Recommended PSU	250 W
Power Connectors	None
Interface	PCIe 3.0 x16
Slot Width	Single-slot, 168 mm length
Card Length	168 mm
Board Partner Model	Reference or OEM specification
Reference Design	NVIDIA reference design
Software
CUDA Compute Capability	7.5
Shader Model	6.8
DirectX Support	DirectX 12.2
Vulkan Support	Vulkan 1.4
OpenGL Support	OpenGL 4.6
OpenCL Support	OpenCL 3.0

Partial entry. Core specifications may be present, but history, variant notes, known issues, and source review may be incomplete.

Why It Mattered

Tesla T4G is a Tesla T Series accelerator from the Turing generation. Its historical context is data-center deployment rather than gaming, so form factor, memory capacity, interconnect, virtualization features, and server compatibility are the details that make this entry useful.

Used-Market Caution

Before buying used, confirm PCIe versus SXM or OAM form factor, passive airflow needs, auxiliary power, HBM size, firmware locks, ECC or retired-page errors, seller proof from nvidia-smi under load, and whether any vGPU or enterprise licensing is required.

Sources

Sources reviewed: 2026-06-20

This entry is still being source-checked.

Specifications are currently based on reference/spec-dump sources. Historical notes require additional source review.

Source Status

Source review pending. Treat narrative notes as incomplete until listed sources are reviewed.

Data confidence and source status

Partial entry. Core specifications may be present, but history, variant notes, known issues, and source review may be incomplete.

Verified: core specifications and historical notes have been reviewed against listed sources.
Partial: core specifications may be present, but history, variant notes, known issues, and source review may be incomplete.
Imported: spec-only imported entry pending editorial source review.
Needs source review: narrative notes are incomplete until reviewed.

Independent reference sources

TechPowerUp GPU SpecsModel-level specification reference for this GPU entryOpen source

Derived datasets and spec dumps

RightNow GPU DatabaseOpen data inventory used for bulk normalization and cross-checking Derived/spec-dump source. Treat as a reference dataset, not independent historical verification.Open source

Found an incorrect spec, missing source, or variant issue?

Submit a correction

Why It Mattered

Used-Market Caution

Tesla T10 16 GB

Tesla T4

CMP 30HX

CMP 40HX

Source Status

Independent reference sources

Derived datasets and spec dumps