The hash rate of the bitcoin network has exceeded (or is very close to) 1 Petahash per second (PH/s), according to various charts online.

TradeBlock, which runs its own mining dashboard, reported that the computational power on the bitcoin network exceeded 1,000,000GH/sec over the weekend. As of last night, it was showing that the hash rate had crept down again to just under 914,000GH/s (914TH/s).

Blockchain.info had a lower estimate, suggesting a hash rate of a little under 950TH/s last night, while Bitwisdom put it at 955TH/s at the time of writing. Bitcoincharts had it at less - around 847.25TH/s (dropping to 802 by the time we finished writing). This in-depth chart set the hash rate's upper boundary at 881.7TH/s on Sunday.

All of these charts show two things: firstly, how difficult it is to arrive at an exact consensus of performance on a highly decentralised network of autonomous devices. And secondly, how quickly the hash rate can fluctuate.

While 1PH/s is a psychologically significant number, the important thing is the general upward trend of network hashing power, which has accelerated at breakneck speed. TGB’s mining dashboard shows the hash rate soaring through 420-440TH/s a month ago, and continuing to rise.

It can’t go on forever, says Tuur Demeester, editor of economic newsletter Macrotrends, who has also been handling PR for ASIC mining firm CoinTerra during its launch. He said:

The question worrying most people who are still waiting for their miners is what the increasing hash rate will do to network difficulty.

“It's a crap shoot,” said Charles Hoskinson, founder of the Bitcoin Education Network. The network difficulty doesn't change in a linear fashion. Instead it increases in a single jump every time 2,016 blocks have been mined.

“They want six blocks per hour, or 2,016 every two weeks,” he says of the core development team that manages the Bitcoin protocol. “But we have seen blocks produced every five minutes.” This happens when the difficulty adjustment consistently lags the increase in network hash rate.

We saw network difficulty more than double in August, reaching 65,750. On Sunday 15th September, it leaped again, rising to 112,600,000.

Sam Cole, co-founder of CoinTerra competitor KnCMiner, argues that this can't continue forever. He said:

He says this because network difficulty is calculated based on the amount of time that it took to solve the last 2,016 blocks (also here). To halve the time taken to solve those blocks means adding the same amount of hashing power to the network again.

That’s fine when you’re increasing the total hash rate from 500TH/s to 1PH/s, or even to 2PH/s, he says. “But to jump to 20 to 40PH/s, I have to release 10,000 boxes. You can’t do that in two weeks. So it can't double this often forever, because there isn’t that much manufacturing in the world dedicated to ASICs.”

This means that while we’re seeing huge increases in difficulty now, we can expect to see it slow over time, say experts. But when?

“Its almost impossible to predict more than a month or two into the future,” said Cole. “We have simulations where all competition ships, and we have simulations where half of them ship. The numbers vary way too much, and long-term approaches are very varied to say the least.”

But we do know that large amounts of hashing capacity are due to hit the market in the next couple of months. Websites are littered with stories of how much hashing capacity people expect to ship. KnCMiner was originally looking at 450TH/s by the end of next month, and that figure may have grown (Cole isn’t saying).

CoinTerra expects to ship 2PH/s of ASICs in December, which would triple the current network hash rate. Bitmine reckons it’ll drop 4PH/s onto the network by March. Then, there's BitFury, and Alydian, which hopes to offer up to 1PH/s in single blocks to bit-ticket miners as of next month. All of this would push network difficulty substantially at least through the end of the year, and probably early into next.

There are unknown variables, however. For example, it's possible that people could turn off significant amounts of computing power between them for various reasons. If prices drop substantially, and people are spending expensive energy mining bitcoins, it may make sense to stop doing so until prices rise again, for example. That is a distinct possibility, because the bitcoin market is still relatively volatile thanks to a need for more liquidity.

One thing is pretty certain, said Demeester: GPU miners at this point are basically running their cards for nothing, at least on the bitcoin network. He explained:

Perhaps he is speaking of FPGA miners, which some are configuring to mine Scrypt-based coins. KnCMiner is planning one of these.

To confuse matters, it isn't just the number of ASIC mining units being shipped that affects hashing power; it's also how well these units perform. 28nm ASICs may be the new 'norm' this month, but things are moving quickly, and sub-20nm is just round the corner, say commentators.

“I think all ASIC manufacturers are looking at sub-20 in some form that will produce another spike in hash rate, maybe by mid 2014,” says Cole. HashFast, which will itself dump significant amounts of hashing power onto the network this quarter, is also taping out its next-generation FinFet chip this quarter, which will up the ante yet again.

So, while exact difficulty is hard to predict, we can expect significant jumps for the next few months, and a probable spike again when next-generation ASICs ship.

Feature image: bitcoin.sipa.be