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
NVIDIA@28nm@Maxwell@GM204@GeForce_GTX_970@S_TAIWAN_1433A1_T4N807.M1P_GM204-200-A1___DSC04620

GeForce 900 series

Nvidia graphics cards using the Maxwell architecture on a 28 nm process.

Photo: Fritzchens Fritz, CC0, via Wikimedia Commons

This family of graphics processors launched in September 2014 to introduce the Maxwell design. Built by TSMC on a 28 nanometre node, it aimed for better power efficiency and performance than prior generations. The line includes desktop and mobile chips, ending game-ready driver support in late 2025.

Share on WhatsApp

  • Launched in September 2014 with Maxwell architecture
  • Built on TSMC 28 nm process with up to 8B transistors
  • GTX 970 faced lawsuit over segmented memory design
  • Game-ready driver support ended in November 2025
  • Improved energy efficiency over previous Kepler chips

What it is

The GeForce 900 series represents a lineup of graphics processing units created by Nvidia. These chips debuted in September 2014 as the first high-end consumer products to use the Maxwell microarchitecture. Manufactured by TSMC using a 28 nanometre fabrication process, the series was designed to replace the older Kepler and Fermi designs. The primary goals included significant gains in energy efficiency, improved graphical throughput, and simplified programming models for developers. This family covers a range of performance tiers, from entry-level mobile parts to enthusiast-class desktop cards.

How it came about

Nvidia announced the Maxwell architecture in September 2010, though consumer products did not arrive until early 2014. The 900 series served as the major introduction of this design to the mainstream market, succeeding the GeForce 700 and 600 families. The development team focused on restructuring how the chip handles data to reduce power consumption without sacrificing speed. By changing the internal layout of processing units, they aimed to deliver better performance per watt. The series eventually paved the way for the subsequent GeForce 10 generation.

Inside the chip

Maxwell introduced a new streaming multiprocessor called SMM, which replaced the previous SMX design. Each SMM contains four warp schedulers that manage independent instructions more efficiently. Unlike earlier architectures, Maxwell removed the power-hungry crossbar switch, allowing resources to be allocated in finer grains. This change reduced the need for wide memory buses, as a larger L2 cache helped store data locally. The second generation added features like Dynamic Super Resolution and improved video encoding capabilities. However, asynchronous compute tasks are handled via software queues rather than dedicated hardware, which can limit performance in specific workloads.

How it was made

TSMC produced these chips using its 28 nanometre process technology. The transistor counts varied across the different dies within the family. The GM206 chip contained approximately 2.94 billion transistors, while the larger GM204 held 5.2 billion. The biggest die, GM200, packed in 8.0 billion transistors. This manufacturing approach allowed Nvidia to balance cost and performance across various market segments, from mid-range gaming cards to high-end enthusiast models.

GeForce 900 series
A drawing made by EveryChip from what this page says. It is a diagram, not a picture of the real chip.

Where it went

These processors powered a wide array of desktop graphics cards and laptop GPUs. Desktop models included the GTX 950 and 960 for mid-range users, while the GTX 970 and 980 targeted high-end gamers. The enthusiast segment featured the GTX 980 Ti and the Titan X. Mobile variants, such as the 900M series, brought Maxwell efficiency to notebooks. Some lower-end mobile chips still relied on older Fermi or Kepler cores. The cards supported modern APIs like DirectX 12 and Vulkan, making them suitable for contemporary gaming and media playback.

Versions and family

The family is split into first and second generation Maxwell designs. Early chips used GM107 and GM108 dies, focusing on basic efficiency gains. Later models utilized GM200, GM204, and GM206 dies, adding advanced features like HEVC encoding and HDMI 2.0 support. The mobile 9xxM series included some non-Maxwell parts based on GF117 and GK208. The series succeeded the GeForce 700 line and was eventually replaced by the GeForce 10 series, which moved to a newer manufacturing node.

Why it matters

The launch of the GTX 970 sparked significant controversy regarding its memory architecture. Users discovered that the card’s 4 GB of memory was split into a fast 3.5 GB section and a slower 0.5 GB section. Nvidia admitted the specifications had changed before release and apologized for the confusion. This led to a class-action lawsuit in the United States, which settled with refunds for affected buyers. Despite this issue, the Maxwell architecture set a new standard for power efficiency in graphics processing.

Problems and reception

Critics noted that Maxwell struggled with asynchronous compute tasks compared to competing AMD hardware. The driver implemented this feature through software shims, which incurred a performance penalty. Oxide Games highlighted these limitations during the development of Ashes of the Singularity. Nvidia later clarified that the driver would force serial execution for such tasks to maintain stability. Game-ready driver support for the series ended in December 2025, with security updates scheduled to continue until October 2028.

Questions people ask about GeForce 900 series

What manufacturing process was used for the GeForce 900 series?

TSMC manufactured these chips using a 28 nanometre process. This node allowed Nvidia to balance transistor density and power efficiency across the various GPU dies in the family.

Why was the GTX 970 controversial when it launched?

Users found its 4 GB memory was split into a fast 3.5 GB segment and a slow 0.5 GB segment. Nvidia apologized for not disclosing this initially, leading to a lawsuit and refunds for buyers.

Does the GeForce 900 series support asynchronous compute?

It supports it partially through driver-based software queues rather than full hardware implementation. This approach can cause performance penalties compared to GPUs with dedicated async compute hardware.

When did driver support end for these graphics cards?

Nvidia discontinued Game Ready Driver support in December 2025. Critical security updates for the architecture are scheduled to continue until October 2028.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "GeForce 900 series", 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.