From the perspective of the card manufacturers, they don't want to chase the mining market and ramp up card production to match it as they view the end of mining demand as a potential risk.
From the perspective of the card buyer, they don't want to buy a 'mining only' card for much the same reason, with one addition. Mining profitability for the individual drops over time, due to rising difficulty. Which means any card you buy has a finite profitable shelf-life. This makes both ASICs and special-purpose mining GPU cards (basically cards with regular GPUs but no video outputs) unattractive -- they aren't worth much on the second-hand market. At least with a gaming card, once you're done mining with it you have a card you can game with that has essentially paid itself off.
I personally think it would be silly and dangerous for graphics card manufacturers to gimp their own hardware in an attempt to appease the gaming market. GPUs are useful for a whole range of things now, which could be impacted negatively by such a move (including things that might be important for gaming performance in the future). To be clear, I say this as an avid gamer myself.
The near future is going to be very interesting, especially if the sentiment that 'cryptocurrencies might go away' disappears. There is always the ever-looming 'Proof-of-Stake' also. If that gets popular, the dynamic will shift again and you might see a flood of cheap graphics cards hitting the second-hand market. That would be nice. I think that PoS is fairly risky though; I keep hearing people saying that Ethereum won't be able to deliver on it.
Cards which have been very much used. Not sure if that's a problem, seen conflicting info for each side on that one.
As a miner, you want to use as little power and generate as little heat as possible. This means in my case that my cards are run underclocked and undervolted.
Two cards in my rig run at 1250MHz core clock and 1085MHz memory clock, at a voltage setting of 0.810V.
I haven't been quite as lucky with the remaining third card in my rig - I can only run that at a core clock of 1220MHz and a memory clock of 950MHz, at a voltage setting of 0.850V.
I have what some might consider rather aggressively low temperature targets for the cards - 54 degrees C. My two good cards can achieve this temperature with fan speeds around 3650 RPM. The fan on the not-so-lucky card needs to spin around 4500-4650 RPM to achieve this. This is on a hot day and these fan speeds are lower if I'm running my air-conditioner. I haven't seen what the rig is like in winter yet. These numbers are likely the worst-case, since it's summer right now.
In my mind, these cards are subjected to much less stress than cards that have been used for gaming, which will run hotter, at higher voltages, with much more thermal cycling, often overclocked. The lifetime of the fans is probably my only concern when it comes to a card that's been used for mining and is what I'd watch out for if buying used. Otherwise, I feel far more confident buying an ex-mining card than an ex-gaming card. Mining done properly seems like it should be far less stressful on a card than gaming, to me.
It would be interesting to see if any kind of concrete data has been collected on this actually - anyone know if any of the big mining farms have published any reliability data (akin to those data centre hard drive reliability reports that get released occasionally)?
Isn’t this similar to the “useful work”/“Mining Heater” situation discussed in the article?
Those people are not credible. Most of them are 1-MB-limited-Bitcoin maximalists who are desperate to harm the reputation and adoption of platforms that threaten Bitcoin's place as the top cryptocurrency in the public mind, like Bitcoin Cash and Ethereum.
Why would a GPU manufacturer prefer that an end user use their GPU for games rather than cryptocurrency mining?
You can see the effects of this: Nvidia is selling their GPUs at close to MSRP on their website with a set limit on the amount you can buy per customer. If they're making bank on cryptocurrency mining why would they hurt their sales by doing that? To me that move only makes sense if Nvidia wouldn't be making a lot of money otherwise and highly valued their current gaming customers.
It would be rather convenient for them if older miner GPUs were quickly made 'incompatible' with the needs of the consumer gamer / PC market.
They don’t prefer anything in particular, they’d just like to keep both markets. They can bifurcate the market and earn more from each demographic rather than let one dominate the other by inefficiently adapting gaming cards to homelab supercomputing.
In one scenario, they let things continue and the mining market reduces supply for the gaming market, which makes their overall market share more brittle. In the other scenario, they increase the volume of cards purchased by gamers, securing that market, while selling more optimized, higher cost mining cards to miners. This has the added benefit of product diversity.
Because one is a fad the other is their major customers (though the market for ML is growing and nVidia is investing there as well)
As for why they don't just make more cards: The volatility of cryptocurrencies makes manufacturers hesitant to ramp up production. They are afraid of being stuck with an oversupply of cards if the bottom falls out of the market.
Given that they use pretty different interfaces to the cards this seems pretty feasible.
So that they can price-discriminate by raising prices for mining and machine learning.
However, we see here that there are really social utility costs, beyond the environment, to locking up GPU hardware.
I suspect that the capital lockup costs are less damaging for the same dollar amount, because capital is more fungible, so the void can be filled by more sources of capital. In the case of GPUs, however, the void can only be filled after a time-lag in changing production focus.
Society happily survived without proof of work for millennia, but I think it’s important to remember that entertainment is the good part of life that proof of work is ultimately supposed to enable.
I mean this in the same sense the founding fathers did when the wrote stuff like:
>“I must study politics and war that my sons may have liberty to study mathematics and philosophy.”
That is, proof of work is the tradeoff that ultimately should enable people’s personal entertainment and happiness.
If more of the former means less of the latter, I think we are in a bad way.
The question is not whether society does fine without proof of work, but whether the PoW mechanism generates marginal trust at a lower cost than the existing social and institutional mechanisms. Lawyers, police, and the legal and state apparatus are expensive, centralized, and slow. Humans (especially educated, first world humans) are also energetically expensive to run!
Basically, you are only accounting for the cost of PoW, which are trivial to measure, but not carefully accounting for the benefits, which are difficult.
You can put an awful lot of lawyers, police, legal and state apparatus in 280'000 homes.
It is sufficient to house all state and full-time employees of the state of Illinois [0] entirely, which unlike bitcoin can do more than transfer wealth.
Illinois has a population of 12'800'000 people [2], bitcoin.info "claims" 23 Million wallet users, however that is a poor number since it only seems to count the number of wallet addresses ever used, the number of unique addresses used is closer to 500'000 and the number of daily users is 200'000 at the current date, which with some rough statistics yields a more realistic 3-8 million users.
In conclusion: Employing humans to do the entire state things allows to service 1.5 to 4 times as many people as bitcoin is currently doing and it also includes a much much broader range of services than bitcoin.
[0]: http://www.governing.com/gov-data/public-workforce-salaries/... [1]: https://www.thebalance.com/how-much-power-does-the-bitcoin-n... [2]: https://suburbanstats.org/population/how-many-people-live-in...
I agree that bitcoin's PoW mechanism as it is today has relatively poor bang for your buck, but there are many potential mechanisms for increasing efficiency by an order of magnitude or more (dPoS, state channels, sharding, etc.).
Ethereum has a lot more potential to not just be a distributed ledger, but a distributed secured computer. If all you are interested in is moving around value, something like Nano might achieve that while burning far less computation that the Bitcoin network would require per transactions.
We should keep in mind that when discussing Bitcoin we're comparing the first version of a nascent technology to mature financial institutions that have developed over hundreds of years.
Nowhere else is this kind of thing celebrated.
This is a weird critique. Blockchains and banks are very different in their use cases. It’s like you’re criticizing someone for choosing a sorting algorithm over a deduplication algoritihm - they do vaguely the same thing, but no one likes the sorting algorithm because it’s less efficient, they like it because it actually sorts.
If you are choosing a blockchain (algorithm A) over a bank (algorithm B), you are doing so because you want the things a blockchain offers that a bank doesn’t. You can’t do comparative algorithm analysis on algorithms that aren’t comparable.
Or, think of it this way: requiring trust is not a bug, it's a feature. Trust is the "clever hack" humans have that allows us to coordinate at scale without absurdly wasteful solutions like blockchain. From this POV, bitcoin as a part of global economy is making a step backwards.
As an analogy, if resources are spent on war, I think it's not so far fetched to say that there is an extraordinary amount of waste in terms of human talent, life, physical infrastructure, and diverted resources. Just because someone spends money doesn't mean it was well spent (unless you measure such things in money).
Its launch MSRP was $110. Today, they are either sold out, or available for $160 or more.
GTX 1060 6GB: launch MSRP was $250. Today it's either of stock or selling for $350.
It's no different for AMD and for higher end GPUs.
In both hashrate vs intial cost and hashrate vs power consumption.
Bitcoin is increasing the amount of nonsense and waste we can generate in ten minutes. And under standard economic guff like that expressed in the OP that's considered an increase in productivity.