The NVIDIA Tesla T4 16GB GPU delivers exceptional performance for AI inference, deep learning, data analytics, and virtual desktop infrastructure (VDI).
Built on the Turing architecture, it’s designed for energy-efficient scalability across cloud, enterprise, and edge deployments.
Featuring 2,560 CUDA cores, 320 Tensor Cores, and 16GB GDDR6 memory, the Tesla T4 brings versatile acceleration for AI workloads — from natural language processing and recommendation systems to video analytics and high-density virtualization.
Its low-profile PCIe form factor and 70W power consumption make it ideal for high-density servers and edge AI environments.
Compatible Part Numbers:
900-2G183-0000-001, P09571-001, R0W29C, 490-BFHN, 490-BFLB, UCSC-GPU-T4-16 — all share identical specifications and performance.
| Specification |
Details |
| GPU Architecture |
NVIDIA Turing |
| CUDA Cores |
2,560 |
| Tensor Cores |
320 |
| GPU Memory |
16GB GDDR6 |
| Memory Interface |
256-bit |
| Memory Bandwidth |
320 GB/s |
| FP32 Performance |
8.1 TFLOPS |
| FP16 Performance |
65 TFLOPS |
| INT8 Performance |
130 TOPS |
| INT4 Performance |
260 TOPS |
| Form Factor |
Low-profile, PCIe Gen3 x16 |
| Power Consumption |
70W |
| Cooling Type |
Passive or active (OEM dependent) |
| NVENC / NVDEC Support |
Yes (Video Encoding/Decoding) |
| Virtualization Support |
NVIDIA GRID / vGPU Ready |
| Thermal Solution |
Server airflow required |
| Use Cases |
AI Inference, Deep Learning, Virtual Desktops, HPC Edge Computing |
| Supported APIs |
CUDA, cuDNN, TensorRT, DirectX 12, OpenGL 4.6, Vulkan |
| Operating Temperature |
0°C – 55°C (typical) |
❓ Frequently Asked Questions (FAQs)
Q1: What are the main workloads supported by the NVIDIA Tesla T4?
A1: The T4 is optimized for AI inference, deep learning, data analytics, and virtual desktop infrastructure (VDI).
Q2: Can the T4 GPU be used in a standard server chassis?
A2: Yes. Its low-profile PCIe design fits easily into standard and high-density rack servers.
Q3: Does the T4 require active cooling?
A3: It’s passively cooled by server airflow, though active-cooled versions exist depending on OEM design.
Q4: How does the T4 compare to the A100 or A30 GPUs?
A4: The Tesla T4 is optimized for inference and virtualization - offering excellent efficiency at lower power, while A100/A30 target high-end training and HPC workloads.
Q5: Are all listed part numbers identical in performance?
A5: Yes - all listed part numbers (900-2G183-0000-001, P09571-001, R0W29C, 490-BFHN, 490-BFLB, UCSC-GPU-T4-16) share the same T4 GPU architecture, memory, and capabilities.
