What is Nvidia Ampere?

When it comes to GPUs, Nvidia is a leading expert. Being in the business since 1993, Nvidia has produced a wide range of top-quality professional and consumer GPUs.  The esteemed architectures of their processors are behind their product’s success, especially of their primary line, the GeForce. Nvidia’s GPUs went beyond the gaming space, expanding their presence in deep learning, AI (Artificial Intelligence), and accelerated analytics. Nvidia infiltrated the data center market a decade ago, starting with the Fermi chips. Subsequent iterations were made, and the company continually released GPUs that fill the insatiable demand for faster processing speed in data centers. Nvidia’s GPUs have evolved over the years, and the architecture designs primarily drive evolution.   

Nvidia’s Ampere Architecture

The competition is seriously tight among GPU manufacturers, but Nvidia does not hold anything back to take the top spot. In fact, Nvidia has dominated the AI market in the past decade. In 2020, news of Nvidia’s first 7nm (8nm for consumer parts) GPU with 54 billion transistors compressed in such a small die made quite a buzz. Codenamed Ampere after the French mathematician André-Marie Ampère, Nvidia’s processor architecture boosts a massive improvement over its predecessors, Turing and Volta, promising more functionalities, better efficiency, and higher performance at lower power levels. Ampere is behind Nvidia’s second generation of RTX GPUs, the RTX 30 series, and is said to be twice as fast as their RTX 20 series counterparts. The Ampere architecture is also the force behind Nvidia A100 data center GPUs.

Ampere’s Specifications

Ampere is Nvidia’s first 7nm/8nm GPU, the second generation of consumer ray tracing, and has the third generation of tensor cores. Ampere is the underlying architecture of GA100, GA102, and GA104 GPUs which are embedded in GeForce RTX 3090, RTX 3080, RTX 3070, and most recently, the RTX 3060. Nvidia is also slated to release RTX 3050 in the following months. Meanwhile, the enterprise GA100 GPU provides 20x more computing power than previous generations of data center GPUs. Here’s an overview of Nvidia’s Ampere-based professional and consumer GPUs [1]:

GPU GA100 GA102 GA102 GA104
Graphics Card Nvidia A100 GeForce RTX 3090 GeForce RTX 3080 GeForce RTX 3070
Process (nm) TSMC N7 Samsung 8N Samsung 8N Samsung 8N
Transistors (billion) 54 28.3 28.3 17.4
Die Size (mm^2) 826 628.4 628.4 392.5
GPC Configuration 8×16 7×12 6×12 6×8
SMs 108 82 68 46
CUDA Cores 6912 10496 8704 5888
RT Cores None 82 68 46
Tensor Cores 432 328 272 184
Boost Clock (MHz) 1410 1700 1710 1730
VRAM Speed (Gbps) 2.43 19.5 (GDDR6X) 19 (GDDR6X) 14 (GDDR6)
VRAM (GB) 40 (48 max) 24 10 8
Bus Width 5120 (6144 max) 384 320 256
ROPs 128 112 96 96
TMUs 864 656 544 368
GFLOPS FP32 19492 35686 29768 20372
RT TFLOPS N/A 69 58 40
Tensor TFLOPS FP16 (sparsity) 312 (628) 143 (285) 119 (238) 81 (163)
Bandwidth (GB/s) 1555 936 760 448
TBP (watts) 400 (250 PCIe) 350 320 220

Nvidia A100 

The Nvidia A100 is the first GPU that used the Ampere architecture. The enterprise chip is targeted at data centers and designed for GPU-intensive tasks such as deep learning and AI. Built using TSMC’s 7nm process, it houses a tremendous amount of 54 billion transistors. It has outperformed the previous generations of enterprise GPUs with a 20x increase in performance featuring 6,912 CUDA cores, 432 texture mapping units, 160 ROPs, third-gen Tensor cores, and a 40GB VRAM with memory bandwidth that tops out at 1.6TB/sec. DGX A100 is the world’s first AI system with a cluster of eight A100s and a hefty price tag of $199,000.

GeForce RTX 30 Series

Aside from dominating the enterprise market, Nvidia always has the consumers, especially gamers and creators, in mind. The Ampere architecture is also adopted in Nvidia’s consumer graphics cards. The second generation of GeForce RTX carries the Ampere-based GPUs boosting the performance to twice as much as the previous generation.

In most aspects, Ampere has more than twice the processing power of Turing. It has doubled Turing’s shader performance with twice the number of FP32 CUDA cores. Specifically, it has 30 Shader-TFLOPS, 2.7x more than Turing’s which is only 11 Shader-TFLOPS. Likewise, Turing’s Tensor Core measures at 89 Tensor-TFLOPS, but Ampere has more than doubled this rate with its 238 Tensor-TFLOPS. Not to forget its Ray Tracing Core rates, which is 58 RT-TFLOPS, 1.7x faster than Turing’s 34 RT-TFLOPS, and to make GPUs run even quicker, the Ampere chip connects to the world’s fastest memory, the Micron G6X.

The RTX 30 series is made using Samsung 8N Nvidia Custom Process with 28 billion transistors for GA102 and 17 billion transistors for GA104. The Titan class RTX 3090 is powered by the GA102 GPU and delivers extremely powerful consumer card performance. Unlike the previous generations, RTX 3090 is open to third parties for custom design.

RTX 3080 is also using a GA102 GPU, delivering twice the performance of RTX 2080 and an impressive 4K gaming prowess. RTX 3070, powered by GA104 GPU, is on par with RTX 2080 Ti at half the price. The recently released RTX 3060 also never fails to impress with its staggering performance driven by enhanced Ray Tracing Cores, Tensor Cores, new streaming multiprocessors, and high-speed G6 memory.

Despite the massive increase in performance, the new RTX GPUs don’t go heavy on the pocket. The much faster graphics cards are easily accessible when it comes to pricing. The Nvidia GeForce’s flagship, RTX 3080, starts at $699, the RTX 3070 sells at $499, and the RTX 3060 has a price tag of $329. The higher-end RTX 3090 costs $1,499; still considerably low-cost considering its Titan-level performance.

Just months after its release, there’s reportedly been a shortage of Ampere-based graphics cards, and it’s no surprise with Ampere’s stunning performance at a reasonable price. For those considering a GPU upgrade, now’s the best time to get your hands on Nvidia’s Ampere-based GPUs.


[1] Walton, Jarred. “Nvidia RTX 30-Series Ampere Architecture Deep Dive: Everything We Know”. https://www.tomshardware.com/features/nvidia-ampere-architecture-deep-dive. October 13, 2020.

About the author

Glynis Navarrete

A freelance blogger who loves to write about anything related to technology. Born and raised in the Philippines and worked in Singapore for eight years as Technical Support for a wide range of IT equipment. Took a dive into the world of freelancing and now enjoying doing what I’m passionate about while not losing touch with technology.