NVIDIA Tesla T4 AI Inferencing GPU Benchmarks and Review

1
NVIDIA Tesla T4
NVIDIA Tesla T4

Today we have our benchmarks and review results of the NVIDIA Tesla T4 AI inferencing GPU. The Tesla T4 is an extraordinarily popular GPU for AI inferencing solution adopted by every major vendor and many cloud providers. Using a single low profile PCIe slot, 70watts of power, and 16GB of memory it puts GPU power in servers that otherwise could not take GPUs. Sometimes multiple GPUs in a 1U or 2U server. No additional power cables are needed with the Tesla T4 as it draws power from the PCIe slots. This helps to reduce cable clutter inside the server box and system integration. As the Tesla T4 is passively cooled, less cable clutter provides better air-flow. In addition, the Tesla T4 adds INT4 capabilities for even faster inferencing needs. In our review, we are going to fun the GPU through our normal battery of tests to see how well it performs.

NVIDIA Tesla T4 Overview

The NVIDIA Tesla T4 is a single-slot low-profile GPU which is only 6.6” long. No power connections are needed enabling the Tesla T4 to fit in tight spaces inside the server. The Tesla T4 size enables one to fit two Tesla T4’s inside the same space as a double-slot full-sized GPU.

NVIDIA Tesla T4 Angle View
NVIDIA Tesla T4 Angle View

We have seen this small design packed into 1U servers. We have seen it paired with embedded CPUs like the Intel Xeon D-2141I as well as high-end dual-socket CPUs like the Intel Xeon Platinum 8280. When the T4 came out, one of the key value propositions was its small form factor making it easy to deploy.

As the Tesla T4 is passively cooled both the front and back of the GPU are taken up by air-flow inlets and outlets.

NVIDIA Tesla T4 Front
NVIDIA Tesla T4 Front
NVIDIA Tesla T4 Back
NVIDIA Tesla T4 Back

There are no video inputs/ outputs here, unlike the similarly sized NVIDIA Quadro P620. If you are looking to power digital signage, this is not the GPU for you.

When we compare sizes of different graphics cards we have tested we see the NVIDIA Tesla T4 is indeed small compared to the full-size RTX 2080 SUPER. The NVIDIA Tesla T4 is the same size as the AMD Radeon Pro WX4100 and only slightly longer than the NVIDIA Quadro P620.

NVIDIA Tesla T4 Size Comparison
NVIDIA Tesla T4 Size Comparison

Next, let us take a look at the NVIDIA Tesla T4 key specifications and continue on with our performance testing.

1 COMMENT

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.