Tesla V100
Nvidia Volta GPU for AI and HPC data centres
The Tesla V100 is a graphics processing unit built on the Volta architecture by Nvidia. Announced in 2017, it targets artificial intelligence training and high-performance computing. It introduced Tensor Cores to accelerate matrix math and used high-bandwidth memory for large datasets.
- Built on 12nm TSMC process
- Features 16GB HBM2 memory
- Introduces Volta Tensor Cores
- Successor to Pascal P100
- Delivers 900GB/s memory bandwidth
What it is
The Tesla V100 is a graphics processing unit designed for data centre workloads. Nvidia created it to handle artificial intelligence training and high-performance computing tasks. The chip uses the Volta architecture and was announced in 2017. It serves as the successor to the Pascal-based P100. The V100 focuses on accelerating complex mathematical operations required for modern machine learning models. It supports major AI frameworks and runs on cloud platforms and enterprise servers.
How it came about
Nvidia unveiled the V100 at its GTC 2017 conference. The company positioned it as the next step after the P100, addressing the growing complexity of AI models. Executives stated that general-purpose processors could no longer keep up with these demands. The launch highlighted the shift towards GPU computing for deep learning. Nvidia promised shipments in the third quarter of 2017, with partner systems following in the fourth quarter.
Inside the chip
The V100 introduces Tensor Cores, a new compute construct for the Volta architecture. These units accelerate matrix processing, which is central to deep learning algorithms. The chip features a 20MB register file and 16MB of cache. It connects to 16GB of HBM2 memory, offering 900GB/s of bandwidth. This design allows it to move data quickly during intensive calculations. NVLink technology connects multiple GPUs in servers, replacing standard PCIe links for higher speed.
How it was made
TSMC manufactured the V100 using a 12nm process node. This fabrication technique allowed Nvidia to pack more performance into the chip compared to previous generations. The manufacturing process supports the high transistor density needed for the Volta architecture. The chip is available in various form factors, including a 150W single-slot card for commodity PCIe nodes.
Where it went
Data centres use the V100 for AI training and inference. It powers high-performance computing applications in scientific research, such as molecular simulations. Enterprise software like SAP uses it for brand impact analysis in video. Cloud providers deploy it in HGX-1 systems with eight GPUs. Developers access it through DGX Station workstations. It also supports virtualised desktops and rendering tasks in professional environments.
Versions and family
The V100 comes in several configurations. An HGX-1 board holds eight V100 GPUs for hyperscale clouds. The DGX-1 V server also uses eight NVLink-connected units. A smaller DGX Station includes four GPUs and water cooling. A 150W half-height half-length card fits into standard PCIe slots for inferencing. These variants allow deployment from large server farms to individual developer desks.
Why it matters
The V100 helped establish GPUs as the standard for AI development. Nvidia claimed it could replace hundreds of CPU-based servers with just a few dozen V100 nodes. This efficiency reduced costs for training large models. The introduction of Tensor Cores set a precedent for future AI accelerators. It strengthened Nvidia's position against competitors like Intel in the HPC market.
Questions people ask about Tesla V100
What architecture does the V100 use?
The V100 uses the Volta architecture. This design introduced Tensor Cores specifically to accelerate matrix processing for deep learning and AI workloads.
How much memory does the V100 have?
The chip includes 16GB of HBM2 memory. This high-bandwidth memory provides a data transfer rate of 900GB/s to support large datasets.
Who manufactured the V100?
TSMC manufactured the V100 using a 12nm process node. Nvidia designed the chip and its Volta architecture for data centre applications.
What is a Tensor Core?
A Tensor Core is a compute unit in the Volta architecture. It accelerates matrix processing, which is essential for training and running artificial intelligence models.
Good reading and sources
There is no English Wikipedia article on Tesla V100 yet, so the facts on this page come from the sources listed above: nvidia.com/en-us/data-center/v100/, servethehome.com/nvidia-gtc-2017-keynote-what-you-need-to-know/. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryChip wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
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.