Fermi
Nvidia GPU microarchitecture from 2010
Fermi is a graphics processing unit microarchitecture developed by Nvidia and released in April 2010. It succeeded the Tesla design and powered the GeForce 400 and 500 series cards. Manufactured by TSMC, it introduced support for Direct3D 12 feature level 11.
- Released in April 2010 by Nvidia
- Contains up to 3.0 billion transistors
- Supports Direct3D 12 feature level 11
- Manufactured on 40 nm and 28 nm nodes
- Named after physicist Enrico Fermi
What it is
Fermi is a graphics processing unit microarchitecture created by Nvidia. It launched in April 2010 as the successor to the Tesla architecture. This design powered the GeForce 400 and 500 series of consumer graphics cards. It also served professional markets through Quadro workstations and Tesla computing modules. Fermi was manufactured using 40 nanometre and 28 nanometre processes by TSMC. It remains the oldest Nvidia architecture to support Microsoft Direct3D 12 at feature level 11.
How it came about
Nvidia developed Fermi to replace the earlier Tesla microarchitecture. The team named it after Enrico Fermi, an Italian physicist. Retail availability began in April 2010. The design aimed to improve parallel processing capabilities for both graphics and general computing tasks. It introduced significant changes to how threads were scheduled and how memory was managed on the chip.
Inside the chip
Each Streaming Multiprocessor contains 32 CUDA cores for parallel calculations. A dual warp scheduler allows two warps to issue instructions concurrently. The architecture includes special function units for complex math like sine and square root. Memory handling features a configurable 64 KB block per multiprocessor. This can act as shared memory for thread cooperation or as an L1 cache. A unified 768 KB L2 cache services all load and store operations across the chip.
How it was made
TSMC manufactured the Fermi chips. Desktop versions used a 40 nanometre process node. Mobile variants utilized both 40 nm and 28 nm technologies. The full GF100 implementation contained approximately 3.0 billion transistors. This high transistor count supported the increased core counts and cache structures compared to previous generations.
Where it went
Consumer gamers used Fermi in GeForce 400 and 500 series cards. Professional users employed it in Quadro x000 series workstations for design tasks. Data centres utilized Nvidia Tesla computing modules based on this architecture for scientific calculations. It also appeared in mobile GPUs within the GeForce 600, 700, and 800 series laptops.
Versions and family
The family includes several chip variants such as GF100, GF104, GF106, GF108, and GF110. Later revisions included GF114, GF116, GF117, and GF119. Kepler succeeded Fermi as the primary architecture. However, Fermi cores continued to appear alongside Kepler in certain mobile GPU models for several product cycles.
Why it matters
Fermi established support for modern rendering APIs that persist in older hardware today. It was the first Nvidia architecture to fully support double precision floating point operations in professional cards. The introduction of configurable shared memory improved data reuse efficiency. This design choice influenced subsequent generations of parallel computing hardware.
Problems and reception
Some consumer GeForce cards had artificially limited double precision performance compared to professional models. A citation note in records questions whether some 28 nm mobile parts were truly Fermi or rebranded Kepler chips. Despite these quirks, it provided a stable platform for early DirectX 12 adoption on legacy systems.


Questions people ask about Fermi
What is the Fermi microarchitecture?
Fermi is a GPU microarchitecture developed by Nvidia. It was released in April 2010 and succeeded the Tesla design. It powers the GeForce 400 and 500 series graphics cards and supports Direct3D 12 feature level 11.
Who manufactured Fermi chips?
TSMC manufactured the Fermi GPUs. Desktop models were built on a 40 nanometre process. Mobile versions used both 40 nm and 28 nm process technologies depending on the specific chip variant.
How many transistors does Fermi have?
The Fermi architecture features approximately 3.0 billion transistors. This high count allowed for increased parallel processing units and larger cache structures compared to the previous Tesla generation.
Does Fermi support modern graphics APIs?
Yes, Fermi is the oldest Nvidia microarchitecture that supports Microsoft Direct3D 12 at feature level 11. This allows older hardware to run some modern games and applications that require this API standard.
Good reading and sources
The facts on this page come from the Wikipedia article "Fermi (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.
Like Fermi
All graphics processorsMore from Nvidia
All of them
WorldGeForce 500 series
Nvidia Fermi-based graphics processor family from 2010GPU · GPU architecture · 2010
🇺🇸 United StatesTILE-Gx
VLIW multicore processor family by Tilera with mesh network architectureCPU · VLIW · 2010
WorldGeForce 300 series
OEM-only rebranded Tesla GPUs from late 2009GPU · GPU architecture · 2009