News
IndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value ChainIndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value Chain
EveryChip
Volta

Volta

Nvidia's 2017 GPU architecture, first with Tensor Cores for deep learning.

Volta is a GPU microarchitecture from Nvidia, the successor to Pascal and the first to include Tensor Cores for neural network work. Announced in May 2017 and built on TSMC's 12 nm FinFET process, it appeared in datacentre and professional products such as the Tesla V100, Titan V and Quadro GV100.

Share on WhatsApp

  • First Nvidia GPU architecture with Tensor Cores
  • TSMC 12 nm FinFET process, 21.1 billion transistors
  • HBM2 memory, about 900 GB/s bandwidth on GV100
  • NVLink 2.0 at an estimated 25 Gbit/s per lane
  • Announced 10 May 2017 with the Tesla V100

What it is

Volta is the codename for a graphics processing unit microarchitecture from Nvidia, sitting between Pascal and Ampere. It is the first Nvidia design to carry Tensor Cores, units built to multiply and accumulate small matrices so that neural networks train faster than on ordinary CUDA cores. The first product, the datacentre Tesla V100, was unveiled at Nvidia's GPU Technology Conference keynote on 10 May 2017. Chips are made on TSMC's 12 nm FinFET process. Volta reached datacentre, professional and enthusiast cards, but no mainstream GeForce board used it.

How it came about

Nvidia put Volta on a public roadmap back in March 2013, years before any silicon appeared. The name honours Alessandro Volta, the Italian chemist and physicist of the late 18th and early 19th centuries. Products arrived in stages: Tesla V100 in May 2017, TITAN V that December, and Quadro GV100 on 27 March 2018. Volta was also slated for the Xavier generation of Tegra system on chips aimed at self-driving cars.

Inside the chip

Volta implements CUDA Compute Capability 7.0 and can run integer and floating-point work at the same time. Its Tensor Core multiplies two 4x4 FP16 matrices, adds a third FP16 or FP32 matrix through fused multiply-add, and returns an FP32 result that may be narrowed to FP16. These first-generation units target neural network training. Memory is High Bandwidth Memory 2, and NVLink 2.0 gives a fast link between CPU and GPU and between GPUs, estimated at 25 Gbit/s per lane, far above PCI Express. NVLink is switched off on Titan V. PureVideo Feature Set I handles video decode.

How it was made

The GV100 GPU is built on TSMC's 12 nm FinFET process and holds 21.1 billion transistors. High Bandwidth Memory 2 sits alongside the die, giving the V100 about 900 GB/s of memory bandwidth.

Where it went

The first card was the datacentre Tesla V100, shipped inside systems such as the Nvidia DGX-1. Professional workstations took the Quadro GV100, while enthusiasts got the Titan V. Volta GPUs were also reported for the Summit and Sierra supercomputers, where they link to POWER9 processors over NVLink 2.0 for general-purpose GPU computing. That link is expected to support cache coherency, which should raise GPGPU performance.

Versions and family

Volta products span the Tesla V100 and V100S PCIe, Titan V and Titan V CEO Edition, and Quadro GV100. Pascal came before it; Turing served consumer and professional markets as a related variant, and Ampere replaced it. Volta's Tensor Cores are first generation, while Ampere carries third-generation units. Support runs to limited help until October 2025 and security updates until October 2028.

Why it matters

Volta introduced Tensor Cores to Nvidia silicon, a design choice that carried forward into later generations and shaped how accelerators handle deep learning. Its use in the Summit and Sierra supercomputers tied the architecture to large-scale GPGPU computing, and NVLink 2.0 raised the bandwidth available between processors. The line stayed at the high end, with no mainstream GeForce cards built on it.

Problems and reception

After two USPTO proceedings, Nvidia lost its Volta trademark application in artificial intelligence on 3 July 2023. The mark remains with Volta Robots, a firm working on AI and vision algorithms for robots and unmanned vehicles. Volta is a codename rather than a trademark for the architecture itself.

Questions people ask about Volta

What is a Tensor Core?

A Tensor Core is a unit that multiplies two 4x4 FP16 matrices, adds a third FP16 or FP32 matrix using fused multiply-add, and produces an FP32 result that can optionally be reduced to FP16. It is meant to speed up neural network training compared with regular CUDA cores.

Which products used Volta?

The datacentre Tesla V100 and V100S PCIe, the Titan V and Titan V CEO Edition, and the Quadro GV100. The V100 also appeared in the Nvidia DGX-1 system. No mainstream GeForce graphics card was built on Volta.

What process made Volta chips?

TSMC's 12 nm FinFET process. The GV100 GPU carries 21.1 billion transistors and pairs with High Bandwidth Memory 2.

What came before and after Volta?

Pascal preceded it and Ampere succeeded it. Turing is listed as a related variant covering consumer and professional products. Ampere's Tensor Cores are third generation, against Volta's first.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "Volta (microarchitecture)", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryChip rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.

It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

Photo credits on this page

The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

The weekly digest

Every Monday: the chip of the day, new pages on EveryChip, and the conferences and anniversaries of the week ahead.