This is a submission for the DEV April Fools Challenge
Abstract
This paper presents BrowserGPU, a novel browser-native framework for the synthesis, configuration, and deployment of fully-functional Graphics Processing Units using JavaScript. I demonstrate that the fundamental building blocks of any GPU, which are transistors, NAND gates, flip-flops, shader cores, and memory buses, are, at their most atomic level, simply things that are either on or off. JavaScript has 1 and 0, doesn't it? I therefore conclude that NVIDIA has been doing it the hard way. My implementation supports 10 preset GPU configurations including the H100, RTX 4090, and AMD RX 7900 XTX, each constructable in a standard Chromium tab running Vanilla JS, in under 800 milliseconds, a fabrication speed that TSMC's 4nm node cannot match. Market implications are discussed.
Keywords: GPU synthesis, browser computing, silicon disruption, Jensen Huang humiliation
1. Introduction
In 1947, three physicists at Bell Labs invented the transistor. They won the Nobel Prize. It was a very big deal. The world celebrated. Textbooks were written. Entire semiconductor empires were built around that event.
What nobody told you is that a transistor is, actually, a switch. It turns on. It turns off. That's it. That is the whole thing. A $30,000 NVIDIA H100 data center GPU is, fundamentally, 80 billion of these switches arranged in a very expensive configuration, cooled by a very expensive fan, inside a very expensive server rack, sold to a very desperate AI startup.
The H100, for example, has 80 billion transistors. JavaScript has let x = 0 and let x = 1. (!!!) That was my eureka moment!!!
I have done the math, and it checks out.
2. Background: A Rigorous Analysis of How GPUs Actually Work
2.1 The Transistor
A transistor is a semiconductor device, typically silicon doped with boron or phosphorus, that acts as an amplifier or a switch. When voltage is applied to the gate terminal, a channel forms between the source and drain, allowing current to flow. This is called "on." When voltage is absent, no channel forms. This is called "off." TSMC charges extra (plus tip) to make this happen on a microscopic level.
In JavaScript: const transistor = (gate) => gate ? 1 : 0. You're welcome, Bell Labs.
2.2 Logic Gates
Combine two transistors and you get a NAND gate. Combine NAND gates and you get everything else: AND, OR, XOR, NOT, all of it. NAND is called a "functionally complete" gate, meaning it is the only gate you ever actually need.
NVIDIA's chip designers spend years optimizing gate-level netlists, running static timing analysis, verifying setup and hold times, simulating process corners, doing Monte Carlo yield analysis across wafers, and arguing in meetings, because they probably haven't taken a Javascript course. They could just do const NAND = (a, b) => !(a && b) ? 1 : 0. You're welcome, NVIDIA.
2.3 The Shader Core (CUDA Core)
A CUDA core is a floating-point arithmetic unit capable of performing one fused multiply-add (FMA) operation per clock cycle. The H100 has 16,896 of them, running at up to 1.98 GHz boost clock, achieving approximately 66.9 TFLOPS of FP32 compute.
In my implementation, each CUDA core is a JavaScript object with a compute() method. We instantiate 16,896 of them. They are stored in an array. This is essentially the same thing except ours does not require a 700-watt power supply.
2.4 The Interconnect
Modern GPUs route signals through a complex mesh of metal layers up to 15 layers of copper interconnect on process nodes separated by dielectric materials to prevent capacitive coupling.
I use the JavaScript event loop and postMessage. It is perhaps slightly slower. I admit this.
3. The Framework: Why It Works
Everything a GPU does is, at bedrock, the movement and transformation of binary information. Electricity moves through a conductor, raises voltage at a gate terminal, inverts or passes a signal, feeds into the next gate, and eventually, after hundreds of millions of these operations per nanosecond, renders your Among Us character running through a corridor.
JavaScript moves bits through logical operations. It raises a value, passes it to a function, inverts or combines it, feeds the result into the next function. The physics are different. The phenomenology is identical.
I am NOT simulating a GPU. I am expressing the mathematical truth that a GPU IS. Physically, NVIDIA builds it from matter. I build it from tokens. One costs $30,000, the other costs a browser tab running JavaScript.
4. BrowserGPU: Implementation
BrowserGPU presents the user with a GPU configuration form of military-grade surgical precision. Users may select from 10 industry presets, or specify a fully custom GPU with the following parameters:
- Core Architecture
- Memory Configuration: VRAM capacity (GB), memory type (GDDR6 / GDDR6X / HBM2e / HBM3 / LPDDR5), memory bus width (bits), memory bandwidth (GB/s), L2 cache size (MB), L3 cache size (MB where applicable).
- Display and Raster Output
- Thermal and Power
Upon clicking "Create and Use in Browser," BrowserGPU instantiates the specified GPU as a structured JavaScript object hierarchy, logs the hardware manifest to the console for professional verification, and proceeds to run a WebGL benchmark demonstrating that the browser is now, effectively, using a GPU of your own specification!
5. Preset Configurations
The following presets ship with BrowserGPU, with approximate current market prices:
| GPU | Segment | Approx. Price |
|---|---|---|
| NVIDIA H100 SXM5 | Data Center | $30,000 |
| NVIDIA RTX 4090 | Enthusiast | $1,599 |
| NVIDIA RTX 4070 Super | Mid-range | $599 |
| NVIDIA RTX 3090 | Previous Gen | $800 (used) |
| AMD Radeon RX 7900 XTX | High-end | $899 |
| AMD Radeon RX 7600 | Budget | $269 |
| Intel Arc B580 | Mid-range | $249 |
| NVIDIA RTX 5090 | Next Gen | $1,999 |
| NVIDIA A100 80GB | Data Center | $10,000 |
| AMD Instinct MI300X | Data Center | $15,000 |
Each preset auto-fills all form fields. Clicking "Create" takes under one second. TSMC takes 3 months. We consider this a competitive advantage. There's also a nice AI feature where you describe your task and the AI (from Google), auto-fills the recommended specifications.
6. Conclusion
NVIDIA Corporation is currently valued at approximately $2.6 trillion (2026), making it one of the most valuable companies in human history. This valuation is built entirely on the act of arranging sand —purified sand, a.k.a silicon.
I have replicated this process in a browser using JavaScript.
My presets are available, and pricing is negotiable. I am open to acquisition offers from Jensen Huang or Sam Altman —whoever calls first. I expect a call sometime this week.
NVIDIA makes $30,000 per H100. I intend to list my BrowserH100 preset at $14,999.99. The margin is, mathematically, more favorable. If 1% of the world's AI researchers switch to BrowserGPU presets, I will be worth more than Jensen. I have already begun designing my leather jacket.
The sand was always free, the same way the browser was always free.
What I Built
BrowserGPU. The world's first browser-native GPU.
I built a full GPU configurator that lets you design, fabricate, benchmark, and USE any GPU in existence,or invent yours right on the spot.
Select a preset (H100, RTX 4090, RTX 5090, RX 7900 XTX, MI300X), tweak every spec from CUDA cores to VRAM bandwidth, then click "BUILD" to open a terminal where your GPU comes alive.
Demo
At BrowserGPU, our mission is simple: Want any GPU? Name it, 'nd get it. So we've got a nice domain for it.
Live Demo: https://gpu.name.ng/home
Screenshots:
How to use:
- Pick a preset or configure your own silicon
- Click "BUILD AND TEST IN BROWSER"
- Run
neofetchto see your GPU identity card - Run
benchmarkand other commands to test your GPU.
Code
Live preview: https://gpus.name.ng/home
GitHub Repository:
BrowserGPU Source Code
How I Built It
Tech Stack:
- Vanilla HTML/CSS/JS, because anyone who wants to use my BrowserGPU shouldn't have to install node.js. No frameworks, no dependencies.
- Google Gemini API for the AI recommender that suggests GPUs based on what tasks/games you want to do/play
- localStorage for saving presets, because why not?
The GPU Architecture:
- 80 billion "transistors" = 80 billion JavaScript
1s and0s - CUDA cores = a number in an input field
- FP32 TFLOPS = a number that gets multiplied by 0.002 to adjust benchmark times
- Ray tracing = a checkbox
The Gemini Integration:
Describe what you want to run ("Crysis at 4K max settings" or "fine-tune Llama L30 90,000B"), and Gemini returns a GPU recommendation with specs. It auto-fills the form. Yes, it actually works.
Prize Category
Primary target: Community Favorite.
My project also has:
- A 419 error "I'm JavaScript, not a GPU", which is my letter to 418.
- Working Gemini integration that recommends, and auto-fills GPUs/specs based on your favourite games or current tasks.
Tag: #418challenge
Top comments (0)