Nvidia Tesla T4 vs Nvidia A100


Name	Nvidia Tesla T4	Nvidia A100
Release Date	13 September 2018	22 June 2020
Price	1,991 USD	10,342 USD
GPU Score
Brand	Nvidia	Nvidia
Series	Tesla T4	A100

Key Differences

The meteoric rise of computer technology and Artificial Intelligence in the past decade has been a source of inspiration and competition for many companies. This consistent competition has led many companies to develop and market more powerful tools for their users, requiring extraordinarily novel and robust hardware.

Nvidia’s Tesla T4 and A100 are two graphics accelerators that allow you to harness the parallel computing capabilities of a graphics card at its peak. The A100 is much more robust, expensive, and effective than the T4, but the T4 offers a good balance of features, price, and performance that easily garners attention.

Why you should consider Nvidia Tesla T4

It offers supreme clock speeds and features an excellent set of features.
Its attributes are available at a much smaller price than the A100.
It does its work at a very trivial power consumption.

Why you should consider Nvidia A100

The A100 is a more recent product with a highly optimized infrastructure.
Its 40GB VRAM capacity substantially improves over the T4’s 16 gig memory.
Furthermore, it relies on a faster memory with extreme bandwidths.

Graphics Processor

The Tesla T4 and the A100 comprise a much different working pipeline and graphics processors. The T4 has a 12nm process-size chip consisting of 13,600 million transistors and utilizes the Turing microarchitecture. Most people are familiar with this chip and architectural design from Nvidia’s RTX 20 series GPUs, which were then superseded by the Ampere architecture. Its 562mm2 die size is also nominal, corresponding directly to its high power efficiency and compact size.

The A100 is a GPU powerhouse that utilizes the latest Ampere architecture to deliver top-notch performance and power efficiency. The chip is manufactured with a cutting-edge 7nm process size and contains a staggering 54 billion transistors. The GPU’s large die size and optimized cores result in higher power consumption, but its effectiveness is significantly improved.

GPU Name	Nvidia Tesla T4	Nvidia A100
GPU Variant	TU104-895-A1	-
Architecture	Turing	Ampere
Manufacturing Size	12nm	7nm
Numbers of Transistors	13,600 million	54,200 million
Die Size	545 mm²	826 mm²

GPU Clocks

The more optimized working conditions of the A100 come at the cost of slightly lower clock frequencies than the T4. Where the T4 can attain a boost core frequency of 1590 MHz, the A100 shortly falls behind at only 1410 MHz. However, the A100 offers a better base clock of 765 MHz than the T4’s 585 MHz.

It’s worth noting that the T4 has a slightly more effective memory clock, with a rating of 1250 MHz, while the A100 has a memory frequency of only 1215 MHz. However, this difference is insignificant, and the A100 compensates for it with its faster and highly efficient memory.

Base Clock	585 MHz	765 MHz
Boost Clock	1590 MHz	1410 MHz
Memory Clock	1250 MHz 10 Gbps effective	1215 MHz 2.4 Gbps effective

Memory Specifications

The A100 uses the HBM2e memory with a 40-gigabyte capacity. With this memory, the A100 also features a substantially wide memory bus of 5120 bits, allowing exclusively high rates of data transfer. Its high data transfer rates are reflected in its bandwidth of over 1500 GB/sec, which is nearly unattainable with other memory types.

The Tesla T4 looks pale in comparison. However, with its 16 Gigs of memory along with a GDDR6 type, it can achieve good performance and results that are truly remarkable on their own. Its memory bus is a more trivial type with only a 256-bit width and attains a commendable bandwidth of around 320 GB/sec.

Memory Size	16GB	40GB
Memory Type	GDDR6	HBM2e
Memory Bus	256 bit	5120 bit
Bandwidth	320.0 GB/s	1,555 GB/s

Board Specifications

The T4 and A100 have a similar look and design with an identical length of 167mm. Also, they don’t feature display ports.

Furthermore, the T4 requires only a tiny amount of power to run. It has a TDP of only 70 Watts and doesn’t need any external PCIe Power connector, meaning it will take its required power directly from the motherboard without any supplement from outside. The A100, on the other hand, needs power like a mid-tier gaming GPU and has a TDP of 250 Watts. Also, it requires a CPU-type 8-pin connector to get the additional power directly from the power supply.

Number of Slots	Nvidia RTX 3090	AMD RX 5700 XT
Lenght	168 mm	168 mm
Width	-	-
Height	-	-
TDP	70W	250W
Power Connectors	None	8-pin EPS
Suggested PSU	250W	600W
Display Outputs	N/A	N/A

Graphics Features

Lastly, the T4 supports all kinds of development APIs, which can be used to write code or develop/render graphical outputs based on Direct X 12, Open GL 4.6, OpenCL 3.0, Vulkan 1.3, and Shader Model 6.7. With this, users have the versatility to develop programs and pipelines in correspondence to their more favored platform. The A100 only supports OpenCL 3.0.