Photo Credit: Jasper Reichardt

Laszlo Hanecz is the Hungarian engineer who is immortalized in Bitcoin history for buying two Papa John's pizzas with 10,000 BTC  in May 2010—making him the first person to publicly initiate a real-world transaction with the fledgling digital currency. Today, 10,000 BTC is worth north of $82 million.

Screenshot 2019-05-15 09.52.25.png
Laszlo's BitcoinTalk post with his offer of Bitcoin for pizza. source: Bitcoin Talk

What many don't know is the story of how Hanecz stockpiled such a large quantity of Bitcoin in the first place. In April that year, Hanecz had figured out how to route mining through the GPU of his computer. Mining previously had occurred through a computer's CPU, which was responsible for performing a computer's other functions. GPUs, which are designed to process and render images and video, were an order of magnitude more efficient for repetitively computing large blocks of data, making them far better suited for hashing.

With his CPU, Hanecz was solving around one block per day. With his new GPU setup, he was winning one to two blocks per hour, earning a reward of 50 BTC per block. By the time of the infamous pizza purchase, Hanecz had amassed 70,000 BTC, or a cool $378 million at today's prices.

Over the past few years, the mining landscape has shifted rapidly. The development of specialized ASIC chips have dampened GPU mining profits for the largest currencies like Bitcoin and Ethereum. But ASICs have to be purpose-built around a specific mining algorithm, which means that GPU mining remains more versatile and valuable for speculative mining. In this article, we'll walk you through a brief timeline of key developments in GPU mining, explain how it works, and highlight opportunities for GPU mining today.

How does GPU mining work

A GPU is the graphics processing unit of a computer, which is designed to help render videos and images. This involves manipulating the same pixels on a screen over and over again, which means GPUs are optimized for performing repetitive tasks and processing large blocks of data.

That's what makes GPUs so much better than CPUs at mining cryptocurrency. In proof-of-work mining, miners compete to solve a cryptographic puzzle to earn a block reward. The only way to solve this cryptographic puzzle is to guess a bunch of different solutions by hashing the same data over and over again with a different random number, until a miner guesses the right answer.

The Bitcoin wiki gives the following analogy for comparing CPUs to GPUs:

One way to visualize it is a CPU works like a small group of very smart people who can quickly do any task given to them. A GPU is a large group of relatively dumb people who aren't individually very fast or smart, but who can be trained to do repetitive tasks, and collectively can be more productive just due to the sheer number of people.

On any computer chip, there's a limited amount of space. A CPU is structured to frequently switch between tasks, and it uses what are called Arithmetic Logic Units (ALUs)  that perform mathematical and logical functions. A GPU doesn't need to frequently switch between tasks, which leaves room for many more ALUs and the ability to process many more hashes simultaneously.

CPU-vs-GPU-Architecture.png
source

A Radeon 5970 GPU can achieve 863 MH/s, or 863 million hashes a second. Meanwhile an Intel i7 CPU might only clock 18 MH/s, or 18 million hashes per second. That's because a GPU has a ton more “dumb workers” or ALUs than a CPU, which can be used to calculated millions of hashes simultaneously.

A Timeline of GPU Mining

While Hanecz was one of the first people to mine with a GPU, he certainly wasn't the last. Back then, the Bitcoin network had been maintaining a “gentleman's agreement” not to hasten GPU mining for the good of the network. Satoshi Nakamoto knew that GPU mining was inevitable, but was concerned that the rise of GPUs would make it harder for new users to mine Bitcoin.

In an email to Hanecz in April 2010, Satoshi Nakamoto wrote:

I don’t mean to sound like a socialist, I don’t care if wealth is concentrated, but for now, we get more growth by giving that money to 100% of the people than giving it to 20%. Also, the longer we can delay the GPU arms race, the more mature the OpenCL libraries get, and the more people will have OpenCL compatible video cards. If we see from the difficulty factor that someone is using too much GPU, we can certainly pick this OpenCL stuff up again then. Maybe my effort to maintain GPU innocence is running out of time. It’s worked out so far.

As Nakamoto knew, capitalism is at the core of Bitcoin, which made the GPU arms race inevitable. Later that year, GPU mining would skyrocket on the network. Here are some of the key events in GPU mining history.

  • May 2010:  Laszlo Hanecz figures out how to route mining through his GPU. He spends 10,000 BTC of his earnings on the first real-world purchase with Bitcoin: two Papa John's pizzas.
  • July 2010: Bitcoin Talk user ArtForz claims to have generated 1,700 BTC in six days, translating to 4% of the market. ArtForz built out what was perhaps the first mining farm for crypto, which would grow to 24 Radeon 5970s in a year.
  • September 2010: Bitcoin developer Jeff Garzik offers 10,000 BTC for an open-source CUDA mining client to BitcoinTalk user Puddinpop, who accepts and releases the code for his GPU mining client later that month.
  • October 2010: The first OpenCL mining client is released, opening up GPU mining to those with OpenCL compatible video cards.
  • October 2011: Litecoin, a Bitcoin clone is launched using the Scrypt mining algorithm, designed to make it harder to mine the currency with FPGAs and ASICs.
  • July 2015: Ethereum launches, employing a mining algorithm called ethash which is designed specifically to bolster GPU mining and stymie ASICs.
  • January 2018: Demand for GPUs for cryptocurrency mining dramatically raises the price of GPUs. In May 2017, an AMD Radeon RX 580 cost around $250—by January 2018, the price has increased to over $400.
  • April 2018: Following the development of Cryptonight ASICs by Bitmain, Monero initiates a hard fork to stymie the growth of ASICs on the network and favor CPU and GPU miners.
  • July 2018: Bitmain releases the first batch of ASICs designed specifically for the Ethereum network, although it's likely that the company used these ASICs to mine Ethereum well before.
  • August 2018: Nvidia announces that following the decline in cryptocurrency prices, GPU sales from crypto have fallen drastically. Nvidia had earned $289 million from  “crypto-specific GPUs” the first quarter of 2018. By the second quarter, that number had fallen to $18 million.

Back in the early days of cryptocurrency, those who wanted to mine Bitcoin with their GPUs had to create their own code to do it. Today, GPU mining is the easiest way for new miners to get started, with open-source GPU miners available for most coins and countless tutorials on the web for selecting the best hardware and building a mining rig.

GPU Mining in an age of ASICs

600px-1_VcJKmqav8tcSRZLksc0jLA.jpeg
source: BitcoinWiki

While GPUs are better at processing large blocks of data and repetitive tasks to CPUs, like CPUs they are designed to perform as general purpose hardware. This makes GPUs useful in applications from video rendering to deep learning and cryptocurrency mining.

The trade-off is that GPUs make a number of compromises that make them less efficient at mining than specialized chipsets. In 2013, the first ASIC chips were developed for the Bitcoin network, and quickly dominated the network's hashpower. Most GPUs produced under 1GH/s. The latest Bitminer S9i can produce 13 TH/s.

Although the rise of ASICs have substantially cut into profits for GPU miners, there remain several opportunities for GPU mining within the space.

Speculative Mining or “SPEC Mining”

ASIC chips purpose-built to mine specific coins will always be more efficient than GPUs, but ASIC miners can only target coins that use the same mining algorithms.

An Antminer S9 can be used to mine Bitcoin and Bitcoin Cash, which both are based around Sha-256, but it won't work on Ethereum, which uses Ethash. While anyone can buy a GPU from manufacturers like Nvidia or Radeon, ASICs take time to design and manufacture. That means that GPU mining remains the most feasible way of speculatively mining of new coins that don't have ASICs yet.

Ethereum launched in 2015, but it was three years before Ethereum ASICs were available to purchase for the public. Monero was launched in 2014, and it took four years before Monero ASICs were publicly released. Today, Ethereum has a market cap north of $17B, and Monero's market cap is north of $1B. While ASICs were ultimately released on both chains,  people still mine XMR and ETH with GPUs. Early GPU miners, of course, raked in the lion's share of the profit.

ASICs may be inevitable on most high marketcap currencies, but GPU mining remains the best way to mine newer coins and potentially get in early on the next Bitcoin.

ASIC-Resistant Algorithms

Although it's possible to develop ASIC chips for any mining algorithm, some developer teams are committed to making their coins resistant to the growth of ASICs.Monero is a case-in-point. Monero has initiated two hard forks specifically designed to tweak its Cryptonight mining algorithm and make it more difficult for ASICs to gain traction, making CPU and GPU mining the most competitive ways of mining the currency. Most people acknowledge that ASIC resistance isn't a feasible long-term solution.

As Riccardo Spagni, Lead Developer of Monero tweeted:

While ASIC resistance won't last, that doesn't mean there isn't a place for it. ASIC resistance provides an incentive for early adopters to join the network, contribute, and earn incentives by GPU or CPU mining. In return, the network gets a stable base layer of users with skin in the game. You can think of ASIC resistance as a subsidy to bootstrap adopters to the network. The cost is that the overall security of the network is lower—but the cost of entry for miners is lower too.

What's Next?

In the early days of cryptocurrency, many were wary of GPU mining because they believe that it would lead to greater mining centralization, and make it harder for new miners to make a profit. Ultimately, the capitalist nature of Bitcoin prevailed along with greater investment into mining. With the dominance of ASICs in mining, the GPU script has flipped. Many small mining outfits and hobbyists can't afford to invest in building out ASIC farms, so for smaller miners GPU mining is typically the most feasible option. While specialized ASIC mining will likely dominate most cryptocurrencies, GPU mining allows miners to get in on the ground and earn money while a coin is still new.

Today, a handful of ASIC manufacturers such as Bitmain, Ebang, and Canaan benefit from economies of scale and dominate the mining business by undercutting the competition. Over time, we'll eventually see increased competition in the industry that commoditizes ASICs and makes them more available to the masses—at which point ASICs might become the new GPUs.