Nvidia launched the first GPU in 1999 with the GeForce 256, pioneering hardware T&L. In 2001, the GeForce 3 introduced programmable shaders, marking the shader era. Over 24 years, GPUs advanced massively, from 57 million transistors in NV20 to 92 billion in Blackwell (B100). Shader counts exploded—from 16 in 2007 to over 21,000 in 2025. Unified shaders appeared in 2007, and AI-focused Tensor Cores began in 2017. Despite huge performance gains, GPU prices rose modestly: a high-end card cost $800 in 2007 vs. $1,900 in 2025—only 57% more after inflation.
Nvidia was the first company to offer a fully integrated graphics processor for PCs, in August 1999, which they named GPU, and the acronym has been with us since. The primary differentiating element of Nvidia’s pioneering GeForce 256 was its hardware transform and lighting (T&L) engine, which offloaded geometry calculations from the CPU. The T&L engine then fed four fixed pixel pipelines, each capable of applying a single texture at a time. Nvidia introduced its first GPU with programmable shaders with the GeForce 3 series (NV20) in February 2001, a significant advancement over the GeForce 256’s fixed-function pipelines. And so, the shader era began in 2001. The GeForce 3 featured four pixel shaders and a vertex shader.
Since then, 18 years later, the programmable shader architecture has expanded, gotten more plentiful, faster, and relatively smaller. The GeForce 3 was built using a 150 nm process. That allowed Nvidia to pack 57 million transistors onto the NV20 die, which had a die area of 128 mm². Today’s Blackwell (B100) design has 92 billion transistors, incorporating 24,064 shaders in a 750 mm² die.
Die MM2 M Transistors Process nm NV20 128 57 150 B100 750 92,000 4 difference 486% 161,304% -97%
Table 1. Moore’s law at work—comparison of 2001 GPU to 2024 GPU (SOURCE: Nvidia)
After the GPU was introduced Nvidia started branching out into different PC segments such as workstation, mobile, and computational. But the gaming market had always been the big one for from 2002 to 2005 graphics AIB’s GPUs were primarily differentiated by clock frequency.
With the introduction of the GeForce 7 Series and GeForce 8 Series, Nvidia began distinguishing between entry-level, mid-range, and high-end gaming GPUs by shader count. Between August 2005 and July 2007 Nvidia introduced 26 versions of the GeForce 7 series with pixel shader counts ranging from 2 to 24, and a variety of vertex shaders. It was pretty chaotic as Nvidia tried to fill every market segment niche.
In April 2007, Nvidia introduced its unified shaders architecture in the GeForce 8600 series. Unified shaders eliminated the fixed function vertex and pixel shader distinction and made a sea of shaders available through DirectX 3D 10 to whatever workload needed one. It was a breakthrough year and established the era of unified shaders which all GPU builders would adopt.
Then, and thereafter, Nvidia (and its competitors) differentiated GPUs by shader count for the main gaming segments, which at that time were Value (or Entry), Low-end (or Mainstream), Midrange (or High-end), and High-end. (AKA Enthusiast) The Value segment would get wiped out by ever more powerful iGPUs in the CPU between 2011 and 2015. Ironically, AMD which made some of the first iGPUs (they called APUs) was the last company to abandon the Value segment.
From 2007 to 2025, the shader count went from 16 (low end) to 128 (Enthusiast high-end), to 3,840 (Entry-level) to 21,760 (High-end), a 2,3900% to 16,900% respectable. Or stated another way, 1,406% to 994% per year. That rate is of course much faster than Moore’s law and one of the reasons GPU vendors said (and still do) that they are increasing performance faster than Moore’s law (which by the way, is not dead).
With so many shaders in modern GPUs the block diagram changes drastically,
Nvidia introduced Tensor Cores as part of their Volta GPU architecture during 2017. This early version of the technology was incorporated into the Tesla V100, a GPU developed with machine learning and artificial intelligence applications in mind. Following this, Tensor Cores became a common component within Nvidia’s GeForce RTX line of graphics processing units intended for gaming. Their inclusion began with the initial RTX cards, which were built upon the Turing architecture and released in 2018, and thus began the era of AI GPUs.
AI will use every shader available to it and ask for more.
In the dedicated AI GPUs, like Nvidia’s Blackwell or AMD’s MI300 series, shader count hits 16,896 and 16,384, respectively, and AI cores are 1024 for AMD and 528 for Nvidia.
With all the technological development in hardware, the price of the GPU add-in boards (AIBs), hasn’t risen nearly as fast of as much. In 2007 a high-end enthusiast, AIB, sold for $600 to $800. In 2025 the high-end RTX 5090 sold for $1,900, just 150% more in 18 years. The inflation rate from 2007 to today is 159%, so an $800 2007 AIB would cost $1,270 today, adjusted for inflation, the cost of an AIB with 994% more shaders would only cost 57% more.
More bang for the buck—Moore’s law at work.