[Update: according to Nathaniel Popper’s book, Digital Gold (as well as in The Age of Cryptocurrency), Laszlo Hanycez is the first publicly known individual who actually utilized a GPU to mine bitcoins, two months prior to ArtForz.]

[Note: in conducting research for a new paper (pdf) on Bitcoin, I culled together the following interesting information on the junction between solo mining, GPU mining and pooled mining]

ArtForz is arguably one of the most interesting individuals in Bitcoin in that he abruptly appeared out of nowhere and simultaneously understood an incredible amount of hands-on knowledge about how to tweak mining hardware performance. He did not make his first forum post (assuming it is not a pseudonym) until July 24, 2010, nearly two weeks after Slashdot announced the version 0.3 bitcoind release and a week after his farm (“ArtFarm”) purportedly found its first block. On July 25 he claimed to have generated 1,700 bitcoins in the previous 6 days which translates into 4% marketshare (43,200 total bitcoins mined over six days / 1,700 = 4%). One user estimate the hashrate necessary to do that at 80,000 KHash/s (or the equivalent of 8 quadcore server CPUs).  In his very last post on Bitcoin Talk two years ago, he mentioned that his farm comprised of 24 Radeon 5970s. If tweaked properly (which he may have been capable of), these would each net a theoretical maximum 800 Mhash/s creating a farm of 19,200 Mhash/s (an upperbound).

Getting to the bottom of the numbers

On August 25, 2011 he stated his farm was less than 1% of the total network hashrate. For comparison, during August 2011 the difficulty rating was around 1.8 million thus ignoring the increased hashrate of competitors, this theoretical upperbound setup was capable of generating 8.38 bitcoins per day. Theymos (creator of Bitcoin Talk, moderator at reddit and now “owner” of the Bitcoin Wiki) noted on October 3, 2010 that ArtForz had a system that held roughly 20-30% of the network hashrate. For comparison, user ‘tcatm’ (Nils Schneider) built a system with three Radeon 5870s operating at 983 Mhash/s, announcing it on October 3, 2010.  Based on the current difficulty at that time tcatm was generating 749 bitcoins per day (~10% of the network).  Schneider later that month went on to build one of the first publicly usable distributed hash farms among other projects.

Theymos also stated that ArtForz was the first person to GPU mine (writing the code himself, which based on his forum posts is very plausible) and in January 2011 explained on the Bitcoin Wiki that ArtForz comprises about 25% of the hashrate.  Yet it is unlikely that this system was the one utilized in July 2010.  And I will describe how and when it likely was.

According to the Bitcoin difficulty rating, the largest jump ever (300% at block 68544) took place several days after the announcement on Slashdot that version 0.3 of bitcoind was release which brought in a large amount of new adopters and miners. At the time, the network hashrate was roughly 1,300 Mhash/s or an entire order of magnitude less than the capability of ArtForz’s farm. However one clue he provided in December 2010 is that the cards he was using were overclocked 5970s hashing at 625 Mhash/s (20% slower than that upperbound estimate).

If this is the case, then the collective hashrate of his system was 15,000 Mhash/s which at the 12252 difficulty level he cites would generate 1,231 bitcoins per day which is approximately 17% of the network hashrate. If the upperbound is used instead, his hash farm would have generated 1,477 bitcoins per day, roughly 20.5% of the network hashrate. While speculative, based on the jump in difficulty and hashrate between blocks 94572 and 9678 the previous week, if this was his farm coming online, it would have generated 2,241 bitcoins per day, or 31.1% of network hashrate.

Taking this forensics a step further, by looking at IRC logs, on September 23, 2010 ArtForz noted he had around 2 Ghash/s and was targeting 15-20% of the total network hashrate. This was comprised of 2 overclocked Radeon 5970s each generating more than 650 Mhash/s and also 4 Radeon 5770s – based on a difficulty rating of 918 this farm was generating 2,191 bitcoins per day (~30% of network hashrate). On September 28, he mentioned he had ~2.1 Ghash/s and another 850 Mhash/s “on and off” and the following day that the collective power was just over 3 Ghash/s. On December 15, 2010 ArtForz noted on IRC that he had 15.75 Ghash/s which at the current difficulty of 12252 would generate 1,292 bitcoins per day and this farm still used Radeon 5770s.

Or in short, ArtForz figured out months before anyone else, how to not only leverage the capability of GPUs, but write the code and deploy a small scale farm (which was initially powered by a generator).  He also later used FPGAs and an sASIC (structured ASIC).  This type of economies of scale spurred the hashrate arms race that continues today.

Contrast this with Satoshi Nakamoto who noted early on in December 2009 that, “We should have a gentleman’s agreement to postpone the GPU arms race as long as we can for the good of the network.” Satoshi was thus surprised that these types of systems could be built, stating in October 2010, “Seriously? What hardware is that?”  ArtForz’s farm was also discussed by Charles Lee (creator of Litecoin) in February 2013.

While I have learned of his probable location during this time (which is unimportant) two other notable facts about ArtForz:

1) on September 23, 2010 he mentioned he had 26,650 bitcoins all from mining from the previous 9 weeks and that this was left over from a much larger batch; in his words he “sold off about 2/3.”  Assuming he began mining on July 18th (based on his forum post stating that) that would make 66 days since his first block which is just over 9 weeks.  Based on block rewards that would mean of the 475,200 bitcoins that were mined during that period his own pool (~75,000) represented 15.78%.  In contrast, roughly 5,000 people were mining by mid-September (his estimate), typically on laptops and older desktop equipment.  The only other publicly known person during this same time frame moving and trading this amount of tokens (and with similar hashrate) was William Pitock (Nenolod).

2) he also tested the limits on a variety of early altcoins such as SolidCoin, i0coin and GeistGeld.   Charlie Lee gave a great presentation on the early history of alts back in January at the BTC Miami Conference (video) (slides).

His identity was seemingly retired two years ago and I am not aware of his next project(s).  Next however, was the first mining pool, Bitcoin Pooled Mining operated by Marek Palatinus (slush), which began public operations on November 27, 2010.  And consolidation and centralization has been happening ever since.

Coda: a few more timeline mysteries solved, looks like a few guesstimates above were very accurate.  Tonight someone from the Bitcoin Talk forum sent me a message with a few more details about the numbers regarding the early days of ArtForz’s farm.  According to IRC logs, on August 9, 2010, ArtForz notes that, “I’m still limping along at ~ 76000khash/s.”

Four days later he notes the following in IRC:

[13/08/2010 15:07:27] <ArtForz> currently I can do up to ~450000kh/s
[13/08/2010 15:11:13] <ArtForz> cpu? 6 ati HD 5770s at 77-78000kh/s a pop
[13/08/2010 15:13:17] <ArtForz> a 5770 is ~75Mh/s at stock clocks
[13/08/2010 15:15:33] <ArtForz> well, I’d say total hours for the software… about 40 or so, but most of that was optimizing to get another +20% performance or so
[13/08/2010 15:16:42] <ArtForz> the first modified client + opencl miner were about 12h of work
[13/08/2010 15:22:53] <ArtForz> If I sold the BTC I have sitting here right now at $0.04, I’d have made back the initial investment for GPUs, power costs and a few $100 for my time
[13/08/2010 15:24:35] <ArtForz> well, guess what, I’m not really planning on selling any large number of these any time soon