> We can see a correlation between GPU prices and the value of crypto currencies, but correlation does not mean causation. A causation is plausible in this case.
For me, that indicates that it's not a "general chip shortage" issue (i.e. not "ASICs are making the chip shortage worse" either), at which point https://news.ycombinator.com/item?id=29766829 makes it hard to see the correlation as caused by cryptocurrencies.
If you mention anything about the cloud, deep learning training on GPUs and the impact on the environment that has and the increasing costs of training and fine tuning your neural net, they won't care at all, they will still do it after all these years.
> If you mention anything about the cloud, deep learning training on GPUs and the impact on the environment that has and the increasing costs of training and fine tuning your neural net, they won't care at all, they will still do it after all these years.
There is a recent trend to discuss environmental impacts of training deep learning models in papers. See e.g. Latent Diffusion Models (Heidelberg), RETRO (Deepmind).
Further, I would posit that machine learning has a much clearer value proposition than cryptocurrencies which are a highly speculative.
Finally, your argument is an example of https://en.wikipedia.org/wiki/Whataboutism. Rather than arguing against facts presented you are changing the subject to say "look here though! this happens here and you're all hypocrites for not noticing it!" This makes your argument seem defensive, biased and unconvincing.
Except that I already agreed with the article before you commented [0], so it isn't Whataboutism. (You clearly knew due to your own reply to my comment). [1] Somehow, you thought I brought up the subject, or didn't I just simply agree with the parent comment too?
Seems like you rushed to comment towards the wrong person.
> ...than cryptocurrencies which are a highly speculative.
How can a regulated stablecoin like USDP or USDC, or even CBDCs sitting on a cryptocurrency technology like XRPL or the Stellar Network etc. be 'highly speculative'?
I hope you haven't shown your 'bias' and targeted me, instead of the parent comment.
> Except that I already agreed with the article before you commented [0], so it isn't Whataboutism.
Apologies but I really had a tough time parsing the last sentence in your initial comment for some reason. I see now that you agreed with the article. I'll concede that you were not technically using that strategy in its purest form. You were, however, deflecting to another issue.
> How can a regulated stablecoin like USDP or USDC, or even CBDCs sitting on a cryptocurrency technology like XRPL or the Stellar Network etc. be 'highly speculative'?
I said "cryptocurrencies" to refer to cryptocurrency broadly (so not just stablecoins). I had hoped that was obvious, apologies.
Machine learning has offered tangible benefits for natural language processing (GPT), semantic search (CLIP/faiss indices), digital art (GLIDE/DALL-E) and protein folding (AlphaFold).
Cryptocurrencies certainly promise to solve a lot of very severe issues in the world such as banking for the unbanked but it's less obvious when it accomplishes these goals. Could you list some tangible benefits that have been accomplished already in the crypto scene?
Your comment comes from a place of assuming the answer to the question, which is kind of twofold.
Does the profit from proof-of-work mining increase the acceptable GPU price for miners AND is there enough demand from these miners for GPUs due to the profit equation to significantly inflate those prices?
If you assume yes, you get yes as the answer without proving anything.
Assuming everyone commenting here has read the post, if it is 'cryptocurrencies''like' Bitcoin and Ethereum that use PoW? Then, that is an undeniable, and an undisputed; Yes.
However, if you are talking 'cryptocurrencies' that means you are talking about all of them. Not all cryptocurrencies are like Bitcoin, and Ethereum and not all of them use PoW. So the title should be:
'Is Bitcoin, and Ethereum to blame for High GPU Prices?'
or
'Are cryptocurrencies like Bitcoin, and Ethereum to blame for High GPU Prices?'
Rather than generalising for 'ALL' active cryptocurrencies in existence, which everyone knows isn't true but continue to attack anyway.
Are storage prices inflated at the moment? How is a physical commodity comparable to a ledger with hashes, ultimately assigned a value purely out of demand, despite being devoid of any actual purpose or usefulness?
The cryptocurrency is designed to demand storage for the sake of demanding storage (i.e. as proof of waste) rather than for any of the natural reasons for which people demand storage.
> Assuming everyone commenting here has read the post
Proof of stake is mentioned at the end of the article.
> Other mechanism for securing blockchains exist and successfully used (see Proof of Stake). Proof of Work creates external costs (environmental impact, higher electricity prices, GPU scarcity) to benefit a small group of miners.
It's a bit funny that the blogpost found a higher correlation between bitcoin and GPU prices than between ethereum and GPU prices.
You don't mine bitcoin with a GPU. Just because it correlates doesn't mean it's a causal effect.
Looking at shop prices is probably suboptimal compared to looking at auction prices, shop prices lag too much. This summer you could almost see demand shifting in real time based on the ethereum price. There are probably other confounding factors such as summer is vacation and warm weather time so demand from both gamers and miners is reduced and perhaps there was more supply as well?
Then the validity of the correlations is "verified" by checking if there exists a correlation between CPUs and cryptocurrency prices? What does this verify, it's just another correlation?
The argument is that bitcoin ASIC manufacturing takes capacity away that could be devoted to GPUs. I don't know enough about chip manufacturing to know if that's true.
Seems plausible, as does the fact that high end gpu crypto mining drives up low end gpu card prices as well. No one is using a 1030 to mine anything, but those prices are up too.
Bitcoin can't be economically mined by GPUs so it's not to blame at all in this crypto hype cycle or even the last one of 2017. The people on reddit you see with racks full of 3000 series cards are mining Ethereum.
This nuance isn’t helpful for this discussion I think. 1.4T of overall cryptocurrency 2.2T is POW (68%). BTC and ETH alone account for 57% of all cryptocurrency market share [1].
Proof of stake is 9% of the overall market. I’m not sure where the other 23% go but I don’t think any one idea has captured a more significant piece. It’s like arguing that saying the energy sector is responsible for CO2 emissions. Like sure, you could make the case that wind and solar are net negative but it’s not a helpful contribution to the discussion.
You have a point for the purpose of calculating the scale of the impact. But it is also true that a significant part of crypto does not drive demand for hardware or electric power.
There are proof of stake coins with multibillion dollar valuations that bring a novel combination of smart contracts and other "platform" capabilities and low cost. This is important because these technologies ensure that crypto can survive if the cost and latency of proof of work becomes a drag on both usefulness and valuation. Kind of like renewables have shown that coal can be phased out. In Germany, renewables are now 40% of electricity production, projected to become 60% by 2030. Trends matter.
>proof of stake coins with multibillion dollar valuations
I can mint a billion PoS shitcoins and sell myself one for ten dollars and suddenly, I have a valuation of ten billion. That doesn't make it less of a shitcoin. Every PoS coin is one such shitcoin.
There exists a (niche?) area of cryptocoins which are rewards for producing CPU / GPU intensive work - one example is GridCoin but many others are put there once you start searching. The result is there are work farms out there computing data arguably for good reasons (cancer research etc.) but in effect you have the same environmental cost as PoW chain computations as they chase the coin rewards.
I’m annoyed as the next guy that my deep-learning rigs are costing 5 figures, but “scumbag” might be a little strong for people who arbitrage NVIDIA’s repeated failures to effectively segment their GPU offering?
It is deeply weird to me that of all the people involved you place the failure on Nvidia here. Regardless, I'm ok with the term "scumbag" for anybody who tries to profit from a negative-sum system.
But they are. I ranted about this here a few weeks ago, but we're still getting intermittent rolling blackouts because of energy shortages caused by miners moving into the country (we have cheap electricity… for now at least).
I too would rather see cards going to people working on protein folding AI and similar.... but I must admit your use cases are likely the more dominant ones.
Alternate answer: No, the manifacturing cost of the gpu stays relatively the same, greedy sellers increase price because of the increased demand. It's not like they wouldn't make profit on regular prices.
I just don't get why people hide behind their finger. This is capitalism 101. You either like it and go by it or dislike it and try to change it. For some reason X the demand for product Y has increased which leads to price bump. If you are pro capitalist economy then you just accept it for what it is. If you don't like it then discuss alternate economies. The rules of the economy are the ones to blame, not the people who play by the rules.
It probably has something to do with the fact that crypto is perceived as having no social value, and therefore people are less inclined to accept negative consequences that may result from such an activity.
Yes this is probably it. But if we walk down that route, what prevents anyone from having a say into what everybody else does with their products. Eg: If think there is no social value in crafting wooden cups, do I have a say in what you do with your purchased wood?
Cost is a major factor of price (so is profit margin). By stating that cost is relatively the same I highlight the stable factor in the price, at the same time highlighting the changed factor in the price (profit)
If a seller has limited capacity to produce, is it better for a motivated buyer to be able to buy one, albeit at a price higher than they’d like? Or is it better if the price is what all buyers would like, but many motivated buyers can’t get one at all?
Neither is perfectly “right”; it’s a direct trade off between “they’re available if you want one” and “they’re as inexpensive as you’ve come to expect from history”.
My personal answer on this question is the later. Price stays the same. Eventually the produced products will get into some people's hands. The only difference is how much of their money will end up in the seller's pocket. The motivation you have doesn't scale linearly with the amount of money you can spend. Eg you may have very poor people who would make great use of product X and are very motivated, but they can't grt their hands on it because it is too expensive for them. See the health/surgery industry as a great example.
There’s an additional wrinkle in that this choice tends to lead to intermediate sellers who take the latter choice and turn it into the former choice with additional friction and fraud risk (and no additional profit to the producer, which might give them incentive to scale up production to meet the overage of demand.)
High end GPUs are readily available on EBay and other secondhand markets and I’ve resorted to buying them there to get them in for ML training machines. (Same with consoles, concert and sporting event tickets, and other limited production, desirable items.)
If demand exceeds supply, but the manufacturer continues to sell at normal prices, a scalper market opens up.
Scalpers capture the difference between manufacturer sale price and market value. That's their business, so they have more time to dedicate to the search for product than the "poor people" who "should" have access to the product.
Surgery is not a good example because there is no reseller market there. Medication would be closer but of course med sales are highly regulated. I agree with the point you're getting at though -- access to essential health care should not be financially bifurcated.
No, my theory is that this the true core of the disscussion here should be how the economy works in general, rather than a specific case of supply and demand.
Don’t get me wrong, I love generality, but it’s clear that the trees don’t let you see the forest here.
Sellers were as greedy before the price rise, so greed is not the cause.
The difference here is that:
1) Supply difficulties mean there’s less in offer.
2) crypto miners have strong demand, since they can make money on those gpus.
I would argue that the market is working fairly well, since gpus are being allocated for their best use. Gaming is great, but people seem to really love their eth.
> greedy sellers increase price because of the increased demand
The way supply and demand works can be more subtle than that. Imagine two sellers selling the same widget, one has 50 units and is selling at $100, the other has 60 units and is selling at $120. As long as the demand is below 50 units, buyers will prefer the first seller, and the price will be $100. Once the demand gets above 50 units, the first seller has no more units, and the price is now $120 since buyers have no choice other than buy from the second seller. Notice that at no moment did any of the sellers increase their price. And the reason for the second seller starting with a higher price is most probably not greed, but higher costs (for instance, their store being in a place with a higher rent).
In addition, it looks like 9.25 million 1000 series cards (1060, 1080, etc.) sold in the first quarter in which they were released, which we can extrapolate out to about 30-40 million in the first year.
It would be interesting to see what percentage of manufacturing capacity NVIDIA is at, as 25%-50% would put ETH GPU mining at the majority of sales. Especially considering that supposedly higher volume cards such as the 3060 do a quarter of the hashrate of a 3080, which would be about 24 million cards. Of course, AMD is also part of the conversation as well, but I believe that they are quite a bit less volume than NVIDIA. Then of course, not all GPU miners are mining ETH necessarily and so on.
What if crypto currencies are to blame for low GPU prices?
In other words, what if the opportunity to make money via crypto currencies has accelerated the development of GPU capabilities and availability beyond what would have happened if crypto currencies never came into existence?
Do you have any evidence to support this hypothesis? As I see it, cryptomining is not what consumer GPU's are designed for, it's just something they happen to be good at
And what's to blame for Bitcoin's high price?
A broken fiat monetary system.
I can't believe some people still think that Bitcoin is more speculative than major tech stocks. Facebook, Twitter, Google, Uber, Snapchat... All propped up by money printing. I think if we ran a simulation using hard (non-fiat) money, the net losses of these big tech companies would be even more significant than Bitcoin's losses on electricity. They're only profitable because of the fiat monetary scheme. If you remove the scheme, I think it's likely that revenue would dry up and profits would quickly go negative. These companies are just not viable in a hard money system. More profitable competitors would quickly spring up and there would be constant churn.
The reason why Bitcoin is so successful and anti-fragile is that it's located everywhere. It always has access to the easiest money from all around the world. It has placed itself at the root of every scheme around the world.
It would matter if all of their customers' earnings were also denominated in BTC. It would dramatically affect their customers' spending habits. For example, while they might be willing to borrow fiat to pay for ads, they might not want to borrow BTC for this same purpose because they know that they may not be able to meet repayments (e.g. BTC's scarcity means that it may not be possible to acquire back that borrowed BTC in the future; cost of BTC is likely to be higher due to the deflationary dynamic). Hard money would affect behavior in profound ways.
One of the hallmarks of a tech company is that the costs associated with producing the product is relatively flat and the revenue from the product goes up linearly with usage. Most tech companies then spend their money on recruiting more engineers, to try to build more products that are cheap to maintain and generate smooth rising revenues. The increased fed spending changes the stock price, but it’s not changing the fundamentals of why selling network access software is so good.
My problem with it is not the content, it's the totalizing, evangelical approach. In tone it's about a half-step from the goggle-eyed people who want to talk about "international bankers" and how they control everything. I have never had a good conversation with anybody who starts out like that, because they are in the grip of a strongly self-reinforcing worldview.
People often say "follow the money, look at the incentives..." Yet those same people keep neglecting the system which produces the mother of all incentives. It's easy to see how heavily money-oriented the average person is nowadays. IMO, the way the economy works is that we're just like ants following a trail of breadcrumbs and the reserve banks decide which direction to toss all the new breadcrumbs...
This post is frankly delusional. You say a bunch of things but have zero explanation for WHY those things might be true. It’s clear at this point that Bitcoin has become an utter failure for almost all the things it was supposed to do, aside from making certain people rich due to speculation.
And what's to blame for the tulips high price? A broken fiat monetary system.
The reason why tulips are so successful and anti-fragile is that they can be grown everywhere.
By the way, if an activity like mining is only profitable because of money printing (depressing interest rates), given the world has been in zero to negative interest rates, this means the activity is not productive.
Some people have argued that tulip mania was driven by monetary policy. When discussing supply and demand, it's not only about supply and demand of the goods or services, it's also about supply and demand of the currency which those goods and services trade against. Changes in the supply of a currency can affect prices in unpredictable ways; inflation is not evenly distributed.
Actually what's to blame for Bitcoin's high price is the finite supply of Bitcoin, and the fact that there's wild speculation on its price (which incidentally is what makes it really horrible as a currency, but that's another topic).
And you're mixing up profitability with stock price speculation, which are two very different things - witness the Gamestop incident. Once the stock is sold to the public, its price fluctuations don't actually affect the bottom line of the company unless there's a buyback.
It's been years since the crypto boom - why hasn't production ramped up accordingly? Or are manufacturers waiting forever until it blows over? Because that may never happen.
I've read that takes 2 years to get a new semi-conductor fab up and running. While demand for GPUs has skyrocketed not only in crypto but in other areas.
Couple that with us going into 2 years of global lockdowns and subsequent supply disruptions have probably exacerbated the situation greatly.
It takes two years under normal circumstances; from what I understand backlogs for the lithography equipment are a huge bottleneck in the industry since there are only a few companies that make such specialized machines. They were already backordered for years pre-pandemic, with most of those getting pushed out because of supply constraint issues.
We’re never going to return to the level of industrial output pre-Covid. There are a dozen high-likelihood scenarios down the pipe in the 2020s (Russia, China/Taiwan, climate drought / famine, etc) that will continue holding global productivity down. We’re looking at global stagflation for most of the rest of our lives as a result, meaning crypto is a good hedge. NVidia just needs to start pricing its top-tier GPUs for crypto farms, then refurbishing the last generation for consumers when the crypto farms upgrade.
Maybe you've heard of global supply chain shortages, covering all manner of chips and other components? And a pandemic producing sudden changes in demand, as
Meanwhile, mining hardware demand depends on cryptocurrency prices, which go up and down like a yoyo. Only a food would invest billions in a new chip plant when demand might have collapsed by the time it's finished.
Cryptocurrency was and is volatile enough that the correct play is to not build out long-term capacity for what very well could be a flash-in-the-pan phenomenon.
Production of something that complex isn't trivial or quick to ramp up, nor is there an unlimited production ceiling. And that's a problem, because there is effectively an unlimited demand for cryptocurrency mining equipment compared to other GPU markets.
I might buy the latest GPU to play PC games at max detail. But there's no increase in value in me buying 20 GPUs so there's effectively a cap on how many GPUs the gaming market will demand. Same goes for other use-cases like scientific computing. In those cases ore GPUs do allow them to solve harder problems or finish calculations faster, but buying more GPUs is more expensive so eventually scientists hit the limits on their budgets.
Cryptocurrency mining is different though since it provides a direct path to turning more computing power into more income. As long as the marginal costs to mine on a GPU are below the expected return and the payback time for the fixed cost of buying the GPU in the first place isn't too long, it nearly always makes sense to buy more GPUs. To put it another way, if it makes economic sense for me to mine on 1 GPU, it makes sense for me to mine with 10, or 100, or 1000. In fact it probably makes more sense to mine with more GPUs since there are economies of scale involved. And since miners are more concerned with their marginal costs than the fixed costs, even huge increases in GPU prices don't alter their behavior much. As long as you're making money running a card, who cares if the payoff time for it increases a few months?
everyone is competing for the fastest TMSC silicon and it takes longer than the amount of time so far to decide to build new capacity and then build it and then make products.
Alternate opinion: proof of work mining provides a valuable service to society much moreso than videogames and the premium on GPUs reflects this.
This isn't mainstream opinion of course, however I predict it will be. Crypto mining is the true first electricity buyer of last resort. It's a productive activity which can use megawatts of power and be turned off at a moments notice in order for the grid to meet the demands of society (as what is happening in Texas right this very moment). What other MW scale industrial process can turn off at the flip of a switch? It's a defacto subsidy on green power projects.
The HN crowd hates to see it, but crypto mining is becoming integrated at the deepest levels with our electricity grid. It won't happen everywhere as some jurisdictions may ban this "unholy union" but the moment public opinion changes, this integration of grid and crypto is a slam dunk no brainer.
> Bitcoin miners rarely use GPUs anymore, but ASICs
Seeing the title and the conclusions, how can this be the case? Computationally they're not much behind ASICs per $, are much more broadly available to buy and are easier to sell when needed.
> Computationally they're not much behind ASICs per $ […]
Is this really true for Bitcoin? I’ve been out of the mining game for a long time, but ~5 years ago this certainly wasn’t true. Back then GPUs were at least an order of magnitude less efficient than ASICs.
BTC ASICS have much better performance per watt for mining Bitcoin. It’s almost impossible to be profitable with btc mining on GPUs. ETH and many other alts are mined on GPUs though.
It really doesn’t work that way. They don’t have moving parts (except the fan which is easily replaced); the question is electron migration and heat related problems. If they’re kept well-cooled they could have another ten years in them.
I'd be much more afraid of the GPU chip unsoldering itself from the mainboard (which is what happens usually if the card dies unexpectedly). If it wears out in 20 years instead of the expected 50… well, good riddance.
Actually, having uneven load is probably worse for GPUs than constant high load due to components contracting/expanding every time temperature changes (think rocking something from side to side to take it out).
For the same reason you might (needs detailed analysis) economically be better off overall by keeping the temperature in your house constant as opposed to saving energy with a schedule.
The article only attempts to show there's correlation, and ends up by saying "correlation doesn't imply causation, but maybe there's causation". So basically, we don't know.
I'm not blaming the author, because determining causation is hard, and checking there's correlation between two variables can still be interesting, but the article falls short of answering the question.
They provide decently good evidence of causation, because of the lack of correlation with CPUs. Correlation doesn't prove causation, but it is evidence for it.
one can imagine factors that would cause both CPU and GPU prices to increase together. for example general purpose computing demand or supply-side constraints might cause all chip prices to rise.
if those non-crypto factors were confounding (say also happened to correlate with crypto-prices maybe due to some third unknown factor) then we would see CPU and GPU prices rise in response to crypto prices. we do not see that.
> for example general purpose computing demand or supply-side constraints might cause all chip prices to rise.
This. Those who can't afford anymore powerful video cards now fall back to CPUs with decent integrated graphics, so that the higher than normal demand is therefore affecting their prices too.
They ran a pretty solid analysis of the subject. Not only do they look at correlation, but they regress out the trend line and show a correlation in the noise. Correlation of noise typically does imply a causal connection. They also look at correlation between GPUs that are used for mining and ones that are not and show that the correlation only holds for GPUs used in mining, so the case seems pretty tight. If you are just trying to poke holes in the analysis I would focus more on the possibility of there being other contributing factors that are more significant (i.g. supply chain issues). In any case the author was considerate enough to include the dataset and the python notebook so you can run subsequent analyses on your own.
Some quick things worth trying:
1. )Weight both the price of Cryptocurrencies and the price of GPUs for Market Size.
2.) Add in other explanatory variables to try to find the relative contributions of different sources.
I'm not trying to poke holes in the analysis, and as I said I'm not blaming the author for what they did. I'm just repeating what the author said:
> We can see a correlation between GPU prices and the value of crypto currencies, but correlation does not mean causation. A causation is plausible in this case.
And then they go on to repeat a very reasonable argument, which I happen to believe to be true, but the kind of argument that everyone was already making before looking at any data.
The fact that the correlations do not hold as much for CPUs is good, but still not enough to stablish causation. As you said, there might be some confounding factors that affect crypto and GPUs but not CPUs.
I work in deep learning and had to camp out in front of a best buy to get my hands on some 3090s. A good portion of the people in line with me were crypto miners who were waiting to get more GPUs to add to their collection because they're making money on each one and as long as it's profitable it makes perfect sense for them to buy up more.
It still doesn’t help that the Ethereum hashrate limiter is not imposed on the highest-end card (it’s FHR not LHR). The 3090 (aka 30-series Titan cards) are still really hard to find. The the 24GB of double-sided GDDR6x gets super hot though and all implementations of the card did not take that into account.
The 3080ti’s are easier to find as they’re LHR and price per hashrate ratio isn’t good. What I mean by easier is that bots have a tendency of going after cards with price/ratio.
It doesn’t help either that DDR5 RAM is also highly scalped. PC memory is in direct competition to GPU memory and vice versa.
You’re also contributing to global warming with mining and contribute to a scheme that allows criminals to transfer money. If your local hospital has to shut down because of ransomware you are making that possible.
If nvidia straight up just stopped selling any cards, declined all their client/customer base, and just put 100% of their global output to mining cryptocurrencies, I think it's a good chance nvidia can do a 50%+1 attack on both bitcoin and ethereum at the same time.
Once they successfully do the attack, they can take over the blockchains and allocate all cryptocurrencies to themselves, cash out, and it'd probably register as a good profit on their income statements.
Yeah, they could do that, but what most of these claims ("X could easily profit Y by doing 50%+1 attack") miss, is that once that attack has been made, who'd trade anything with the currency that was attacked? Lets say nvidia does "conquer" Bitcoin and have all the bitcoins, who'd in the right mind would continue to buy/sell Bitcoins after that? The market for Bitcoin would collapse faster than nvidia could excitedly scream "PROFITS!" to their shareholders, making the whole attack unnecessary in the first place.
No, the whole point of using a blockchain is to have a publicly verifiable ledger of all transactions. If anything goes wrong, there will be so many alarm bells ringing across the world that probably it'll be a hearable event for most people living in places that resembles modern civilization.
The idea that you could buy a real-world item worth half as much as the amount of ETH/BTC you would mine in an hour or two or 51% is simply unrealistic.
ETH mining at 500TH (a solid chunk over 51% just for ease) would net you 300 ETH an hour...that's 1.1 million USD, which means you would have to be able to purchase millions of dollars worth of real world items that would be released to you within the time-frame of your attack (in this case an hour) to make this even approaching "worth it" from a risk/reward viewpoint and even that calculation is ignoring just how profitable they'd be just mining normally for the foreseeable future.
No one is waiting for just an hour worth of confirmations for millions of dollars of real world goods exchanged for ETH.
no, contracts are immutable, I'm unaware of any base-layer chain attacks that can affect contracts negatively (outside of contract level logic that may be vulnerable itself)
A 51% attack can only reverse a transaction anyway, it can’t arbitrarily take money from someone you haven’t transacted with. But they could get short exposure to btc/eth in the market, profit when it went down from the ensuing panic, and then sell the graphics cards to retail (there are other complications that make this unrealistic, of course.)
Well, the history of actual 50% attacks on cryptocurrencies in the past few years seems to suggest that effect of a 50% attack on the value of the cryptocurrency is... pretty close to nil, actually. Admittedly, a successful attack on a major cryptocurrency like Bitcoin or Etherium is likely to have far different impact on cryptocurrency society than one on a minor cryptocurrency, but the ability of the community to twist every bad news into "this is good for Bitcoin" should not be underestimated.
I suppose any of the Faang companies could sacrifice a data center for a day or two to do the same. Just crash the currency and get out, to eliminate proof of work currencies.
NVIDIA would reduce demand, but for the FAANG companies it's a neutral proposition, aside from the cost of doing the work to crash the currency.
And there's plenty of reasons to want to do so: Proof-of-work currencies are what make ransomware payments possible, and there's a large daily CO2 cost for proof of work currencies. The question is simply when is the cost of the externalities greater than the cost of acting?
There is actually a good case to make here. Nvidia currently is in the tough spot of a) not having the capacity to saturate the market and b) not being able to expand the production to match.
B is because they don’t have enough confidence that the crypto people will be there as costumers in 5 years. And A is a problem, because their traditional user base (ie gamers) won’t match the spending, but will still get their fix (just not from Nvidia). And that breaks the strangle hold that they had a couple of years ago on that market.
So short term: having the high prices paid by crypto people is nice. But it destroys their relationship to their old customer base. Depending on where the crypto market will be in 5, 10, 15 years, this can be the 5x of their market cap or the 0.2x
They have done things like come up with a mining only SKU, apparently partially because they are annoyed that people aren't using their cards for their intended purpose. Probably, while a sale is a sale, they see the crypto market as flakier and would rather keep their gaming reputation intact.
Relying on single customer can have quite severe negative consequences (walmart crushing whole businesses is a good example). GPU shortage is destroying PC position as a viable gaming platform. Give it couple more years of GPU shortages and Nvidia will irrevocably lose whole market segment, leaving it with compute and shaky scam riddled crypto.
> One of the first ransomware attacks ever documented was the AIDS trojan (PC Cyborg Virus) that was released via floppy disk in 1989. Victims needed to send $189 to a P.O. box in Panama to restore access to their systems[0]...
Just guessing, but there's probably more risk to running a mining operation than miners realize, and Nvidia doesn't want that risk. There are also PR issues with energy use. There's enough money in selling pickaxes.
This is exactly it. NVIDIA are also not building more factories to meet the demand because they are (correctly) assessing that a bigger risk for them is demand collapse and a glut of used hardware.
> If nvidia straight up just stopped selling any cards, declined all their client/customer base, and just put 100% of their global output to mining cryptocurrencies, I think it's a good chance nvidia can do a 50%+1 attack on both bitcoin and ethereum at the same time.
No, not for Bitcoin. GPUs are no good at mining Bitcoin; you need an ASIC for that.
> Once they successfully do the attack, they can take over the blockchains and allocate all cryptocurrencies to themselves […]
No. Having a majority of mining power does not enable anyone to assign currency to themselves, except for the mining reward. That would require faking digital signatures.
Would this be legal? Obviously bitcoin has no "terms of use" that Nvidia (or anyone) is bound to, but the US has very broad "hacking" laws like CFAA.
I imagine that if Nvidia were to openly "steal" someone's bitcoin using a 51% attack then the victim could sue Nvidia in a US court and things could get interesting.
What laws would apply and what would the outcome of such a case be? And could the act also be considered criminal?
Stealing is stealing. 51% attack is just a method to rollback a transaction. No different than ordering stuff and doing a willfully fraudulent chargeback.
There is no reason enough GPUs can't do it. Yes it is more expensive then getting the same hash rate via ASICs but the profitability off a 51% attack is a lot higher and NVIDIA would get their own chips at a reduced rate.
For bitcoin at least, a top of the line ASIC miner can do ~1,000,000x more hashes per second at ~1/100th the power per hash compared to a top of the line GPU. The current total hash rate for the Bitcoin network is ~200EH/s. So a 51% attack would require ~2 trillion GPUs consuming/costing 100x as much power as the Bitction network does today.
You haven't been able to GPU mine bitcoin in like a decade, every miner uses dedicated ASICs specifically for mining btc. It's been impossible to compete without ASICs for a very long time.
Nvidia can design and manufacture asics too you know. Likely they have enough expertise to revolutionize the asic market as you know it, using the 3nm process at tsmc.
Since Bitmain asics are currently being made at around 16nm nodes, using 3nm process that Nvidia has access to - think about that for a bit.
Please bear in mind that 3nm is an "alpha" - a multiplier characterizing the process. If you took your time and find out difference between processes with different "alphas" you will find that transistor density is nowhere near what difference in "alphas" would suggest.
If I remember correctly, the difference in transistor density improves as 1/alpha whereas area of transistors, if everything gets shrunk with alpha decrease, should decrease as square of 1/alpha. I.e., if alpha differs two times, there should be 4 times more shrunk transistors whereas practice shows only 2 (and often less) more shrunk transistors.
This is because you cannot make transistors too small in every dimension, they will leak.
So the difference between 3nm and 16nm will be less than five times. Not 25 times as difference between these parameters would suggest.
The difference in getting masks and first prototype should be quite substantial in terms of upfront payments and time-to-production.
I remember that 25 square millimeters proof of concept in 180nm process needed $50K of money and half an year of delay. The 35nm (5 times less) prototype of the same area would cost about $500K and more than half an year delay. The difference between 16nm and 3nm is about the same, I can expect several millions of dollars of upfront cost of small chip and year long delay before prototype arrives for 3nm process.
These Bitmain guys should be assumed to be not stupid and they most probably have access to latest processes after second prototype. Yet they choose 16nm - we must ask ourselves "why?" and Nvidia should too.
As a matter of fact, Nvidia had its share of problems with tiny processes in past.
The part where you say "cash out" is hilarious. There is no cash in it. You only get to keep the bits. Any cash comes from buyers and if there is any attack all the buyers are scared away.
> Any cash comes from buyers and if there is any attack all the buyers are scared away.
Well, assuming any kind of rationality in the crypto market is always hazardous: meme coins (doge coin, shiba inu) are worth billions in market cap, so do hard forks of bitcoin, and Ethereum didn't not collapse after they decided to rollback a bunch of things they didn't like during the DAO “hack” while they are advertising Ethereum as immutable and tampering resistant.
You can never be sure how this market will react in any event (well, except Elon tweets I guess)
This is the terrifying thing about a bug in one of these systems or some other kind of attack. The whole network can suddenly drop to zero in the blink of an eye.
Hence the value of a stabler currency such as USD or GBP versus more volatile currencies. It is an implicit judgment on how much trust there is in the system that uses them.
As with any software, bugs can be fixed or forked out. The risk of it going to zero is smaller than most people think. Remember the inflation bugs Bitcoin had in it's early days? [1]
Sure, but basically every currency has some theoretical problem.
Usually it's just about the currency becoming worthless (usually only a problem in government controlled currencies that have loans in foreign currencies), or people just not accepting it anymore.
There's no inherent value in any currency, it's just a useful fiction because it's really hard to go get five goats to trade somebody for a laptop.
Currencies are propped up by taxes. The IRS wants a significant percentage of the US GDP paid in USD, as long as the economy exists and the government doesn’t print truly vast amounts of money from thin air the USD can only move within a fairly narrow range.
That’s less significant than you might think as US GDP is almost 1/4 of global GDP and foreign countries don’t keep that many USD on hand. Aka even in global terms US GDP is vast.
Also, Bitcoin has significantly more vulnerabilities than just a 51% attack or a crash as nodes need to be connected to the internet and therefore can at least in theory be hacked. What exactly happens after that point is anyones guess, but it could easily be worse than a traditional 51% attack.
> That’s less significant than you might think as US GDP is almost 1/4 of global GDP and foreign countries don’t keep that many USD on hand. Aka even in global terms US GDP is vast.
Sure, but the value of the dollar (like everything else) is in how easy it is to get (supply). The US GDP is not a measure of the supply of dollars on the market, it is a measure of how often those dollars are exchanged.
> Bitcoin has significantly more vulnerabilities than just a 51% attack or a crash as nodes need to be connected to the internet and therefore can at least in theory be hacked.
That's fair, if we're including implementation vulnerabilities, it does expand the potential problems quite a bit. The post I was responding to was specifically about a 50 + 1 attack. Though if we include these, we should probably also expand physical currencies' problems to include counterfeiting in that case.
> Sure, but the value of the dollar (like everything else) is in how easy it is to get (supply). The US GDP is not a measure of the supply of dollars on the market, it is a measure of how often those dollars are exchanged.
The US GDP directly translates into the amount of taxes paid. Suppose a 50% drop in the value of USD happens. If nothing else happens taxes increase by 100%, and people need to acquire more USD which drives the value of the currency up.
You might wonder about economic hardship, but historically that causes deflation. +/- 2% per year without printing money is easy, but there is a reason you don’t see huge swings without the government printing money.
> counterfeiting
An issue for any physical currency but electronic currency actually operates independently. Trying to counterfeit large quantities of cash is therefore surprisingly difficult even for nation stale level actors. A great fake bill can get passed around, 10 billion in fake bills which is still a trivial amount would trip up various monitoring systems.
> If nothing else happens taxes increase by 100%, and people need to acquire more USD which drives the value of the currency up.
Oh definitely. I did not mean to imply that taxation in USD does not have a modulating effect (and if I came across that way, I apologize). I'm just trying to say that you can have some pretty wide swings even with that.
> but there is a reason you don’t see huge swings without the government printing money.
You don't generally see it, no. Mostly because other countries use the dollar as a store of wealth, and they don't want to tank the value that they've squirreled away. I'm bringing it up as a hypothetical with the same likelihood as a 50 + 1 attack on bitcoin.
China, for example, seems to hold ~2 trillion in USD (~60% of 3.5% trillion total reserves).[0] I assume most of this is held as digital assets, as there's only ~2 trillion physical dollars in circulation.[1] Dumping that much on the market would cause some pretty wacky inflation (as well as destroy China's stability). In a similar vein a 50 + 1 attack on bitcoin would allow an actor to give themselves all of the bitcoins, but would also destroy the value of those bitcoins because nobody is going to trust the network after that.
Both of these attacks are really only useful if you value hurting other actors more than you value your own resources/your own sense of self-preservation.
>The US GDP directly translates into the amount of taxes paid. Suppose a 50% drop in the value of USD happens. If nothing else happens taxes increase by 100%, and people need to acquire more USD which drives the value of the currency up.
?
GDP is produced and denominated in USD, domestic taxes are paid in USD. How does a drop in the relative value of USD compared to a foreign currency influence domestically paid taxes in any way?
> Any cash comes from buyers and if there is any attack all the buyers are scared away.
I used to assume that, but there are coins which have been attacked successfully before and people still use them anyway. I don't attempt to explain the logic there :P
Nah nah. What you (and all the other comments miss), is if you were going to do that, you'd just reduce the sale of cards tremendously, while hoarding your own production output to take over. You wouldn't cut out all sales and hoard all of them.
See, if they just stopped selling them, people would ask questions. If, however, you just reduced the supply by like 80%, said it "was due to all those crypto miners hoarding them", and then proceeded to use that hoarded 80% to do a 50%+1 attack, you'd be able to -quietly- take over, and turn around and sell those coins to the still preserved market.
applies tinfoil hat In fact, who is to say that isn't exactly what they're doing, and hence the existing shortage?
>Once they successfully do the attack, they can take over the blockchains and allocate all cryptocurrencies to themselves, cash out, and it'd probably register as a good profit on their income statements.
So they're going to manage to not sell any cards (without disclosing that as a public company), set up the infrastructure and do a 51% attack on bitcoin in front of everyone, then "cash out" as if there would be people lined up waiting to buy bitcoin when this has just happened in front of their eyes?
I hadn't considered this before, but that's a really compelling reason that a 51% attack would not be worth anyone's while. Seems like a successful 51% attack would just wreck bitcoin, and there wouldn't be any way to cash out.
Although AMD may deny it, I suspected this was a reason for their current success, not what they claim publicly...but this study seems to find no correlation (with ether and bitcoin prices)
Not sure what you are trying to say here but that list is weird. Why is 5700xt that great but 5700 not even listed? Not much difference between the two
ETH. I bought a new computer in March of 2021 for $1,900 including tax. The computer happened to have an RTX 3070 card in it.
I have mined and sold $2,800 of ETH in the 9 months I’ve owned it. The extra power has cost me a couple hundred dollars.
I’ve made anywhere from $3 to $75 per day during that time (thank you Shiba Inu launch date for the highest numbers).
I’m continuing to make about $3 per day. You can look up current profitability for any graphics card at whattomine.com.
Bottom line is I could have added infinite graphics cards to this setup and the profit would’ve gone up linearly.
Anyone that doesn’t think GPUs are high due to ETH mining has no idea what they’re talking about. The analysis from OP seemed ridiculous to me. I didn’t even see LHR mentioned in the article which are GPU manufacturers attempting to make GPUs unprofitable for miners so they can only be used for gaming. (Hint: didn’t work very well).
Bitcoin is the biggest crypto currency and uses proof of work. As mentioned in the conclusion, ASICs use the same resources as GPUs. This can have an effect on GPU prices. Additionally bitcoin is still the "gold standard" and most crypto currencies follow the price development of bitcoin.
NVidia has a far greater complexity to handle. GPUs do a lot of things and are more generalized with each iteration. In comparison ASICs are super simple chips.
If GPUs would still use the fixed rendering pipeline they started with ( only push your triangles and texture to the GPU and you get your image) they could also handle their business with a few hundred employees.
Just shrink your chips to the latest fab size, add more computing cores, maybe improve some algorithm if a better solution is found.
But that's not the question. It asks whether they're to blame for high GPU prices, which presumes a certain threshold before the criteria is met. If there were a few hundred GPUs bought by miners, and that demand caused GPU prices to climb $10 over MSRP, I doubt that would qualify as them being responsible for "high" prices, even though they did raise it.
One of the craziest videos I saw last year was a mob of people rushing a Micro Center to get their hands on a 3080 TI: https://youtu.be/UNcx9JdC9VM?t=239
I guess we'll find out. It's increasingly likely the merge will happen his year. When ETH moves to PoS, mines will likely move to other currencies, or sell their hardware, that said the return for alternative crypto's still on PoW is unlikely to match that of ETH. I would be very surprised if demand from miners is not reduced.
Right, because GPU's are expensive and scarce because of crypto, and all other inflation, demand and supply chain issues are a completely unrelated topic that just happens to be coinciding... /s
GPU's went up as crypto gained popularity, crypto has gained popularity as governments kept printing more money handouts. We now have high inflation, high demand, chip shortages, supply chain shortages and high crypto popularity because of out of control fiat printing.
This also coincides with the governments sudden desire to regulate the industry despite it existing for nearly a decade now. Its a leak in the their system. You can't print money and hand it out where you want if someone has a hedge against your money printers.
While I agree with you that out of control fiat printing is a problem, I think you're overstating things. You can't ignore the role of the pandemic in the supply chain issues, and you can't ignore that many turn to crypto only to gain more fiat. I say this as someone who has been in crypto since 2011, and I've seen what happens during the bear markets. Many who aren't "true believers" or whose bags are too heavy abandon it, even though the money printers haven't changed during those times.
PC Hardware can come from an Era with so many manufacturers and true-actual fierce competition, resulting in complete commoditisation of components. However, now in 2022, there are only 2 CPU vendors,2 GPU vendors and even with storage the number of manufacturers is small. Only TSMC can deliver 5nm and Mr Cook has all that capacity on lock (AMD and Intel will get a look only when he is done with it)
Mining like with politics is just some easy target to blame, the problem is Monopolies and duopolies.
Events have catalysed a market rife with monopolies and duopolies: these rule the supply chain, they have seen the true prices the market will accept, how can we expect prices to go back to "normal"?
I walked into a computer store last night. There were shelves full of CPUs and SSDs and HDDs, and buying several of them at once would have no one batting an eye. Yet there was a policy: only 1 graphics card per customer per month, and that applies solely to graphics cards (and prebuilt computers containing graphics cards).
In other words, there is visibly a shortage of graphics cards that is not affecting any other components. That strongly suggests that there is something driving demand for graphics cards that is not driving demand for other key components of computers, which are available in sufficient supply. And the list of things that extra demand could be coming from is rather short indeed.
Yeah, I couldn't find ECC memory when I was building my Threadripper workstation mid 2020. I ended up with consumer-grade memory which was ridiculously cheap, however. (I think I got 128GB for less than $400.)
351 comments
[ 5.3 ms ] story [ 144 ms ] thread> We can see a correlation between GPU prices and the value of crypto currencies, but correlation does not mean causation. A causation is plausible in this case.
https://www.pcgamer.com/why-you-still-cant-buy-a-graphics-ca...
Don’t look for a zebra when all you have is a horse.
As you know it is still a valid point.
There is a recent trend to discuss environmental impacts of training deep learning models in papers. See e.g. Latent Diffusion Models (Heidelberg), RETRO (Deepmind).
Further, I would posit that machine learning has a much clearer value proposition than cryptocurrencies which are a highly speculative.
Finally, your argument is an example of https://en.wikipedia.org/wiki/Whataboutism. Rather than arguing against facts presented you are changing the subject to say "look here though! this happens here and you're all hypocrites for not noticing it!" This makes your argument seem defensive, biased and unconvincing.
Seems like you rushed to comment towards the wrong person.
> ...than cryptocurrencies which are a highly speculative.
How can a regulated stablecoin like USDP or USDC, or even CBDCs sitting on a cryptocurrency technology like XRPL or the Stellar Network etc. be 'highly speculative'?
I hope you haven't shown your 'bias' and targeted me, instead of the parent comment.
[0] https://news.ycombinator.com/item?id=29759242
[1] https://news.ycombinator.com/item?id=29759347
Apologies but I really had a tough time parsing the last sentence in your initial comment for some reason. I see now that you agreed with the article. I'll concede that you were not technically using that strategy in its purest form. You were, however, deflecting to another issue.
> How can a regulated stablecoin like USDP or USDC, or even CBDCs sitting on a cryptocurrency technology like XRPL or the Stellar Network etc. be 'highly speculative'?
I said "cryptocurrencies" to refer to cryptocurrency broadly (so not just stablecoins). I had hoped that was obvious, apologies.
Machine learning has offered tangible benefits for natural language processing (GPT), semantic search (CLIP/faiss indices), digital art (GLIDE/DALL-E) and protein folding (AlphaFold).
Cryptocurrencies certainly promise to solve a lot of very severe issues in the world such as banking for the unbanked but it's less obvious when it accomplishes these goals. Could you list some tangible benefits that have been accomplished already in the crypto scene?
The partnership between Moneygram and Stellar. [0] Using the USDC stablecoin on the stellar network for near-instant settlements and transfers.
Stellar pilot into CBDCs in Ukraine, [1] and Ripple used for CBDCs for Bhutan. [2]
[0] https://ir.moneygram.com/news-releases/news-release-details/...
[1] https://www.coindesk.com/policy/2021/12/14/ukraine-commercia...
[2] https://ripple.com/insights/bhutan-advances-financial-inclus...
Does the profit from proof-of-work mining increase the acceptable GPU price for miners AND is there enough demand from these miners for GPUs due to the profit equation to significantly inflate those prices?
If you assume yes, you get yes as the answer without proving anything.
However, if you are talking 'cryptocurrencies' that means you are talking about all of them. Not all cryptocurrencies are like Bitcoin, and Ethereum and not all of them use PoW. So the title should be:
or Rather than generalising for 'ALL' active cryptocurrencies in existence, which everyone knows isn't true but continue to attack anyway.I have no idea what the point of your second question is.
Thus artificial.
Proof of stake is mentioned at the end of the article.
> Other mechanism for securing blockchains exist and successfully used (see Proof of Stake). Proof of Work creates external costs (environmental impact, higher electricity prices, GPU scarcity) to benefit a small group of miners.
For people interested, look up the "nothing at stake" issue.
Then the validity of the correlations is "verified" by checking if there exists a correlation between CPUs and cryptocurrency prices? What does this verify, it's just another correlation?
Seems plausible, as does the fact that high end gpu crypto mining drives up low end gpu card prices as well. No one is using a 1030 to mine anything, but those prices are up too.
Proof of stake is 9% of the overall market. I’m not sure where the other 23% go but I don’t think any one idea has captured a more significant piece. It’s like arguing that saying the energy sector is responsible for CO2 emissions. Like sure, you could make the case that wind and solar are net negative but it’s not a helpful contribution to the discussion.
[1] https://coincodex.com/cryptocurrencies/sector/proof-of-work/
There are proof of stake coins with multibillion dollar valuations that bring a novel combination of smart contracts and other "platform" capabilities and low cost. This is important because these technologies ensure that crypto can survive if the cost and latency of proof of work becomes a drag on both usefulness and valuation. Kind of like renewables have shown that coal can be phased out. In Germany, renewables are now 40% of electricity production, projected to become 60% by 2030. Trends matter.
I can mint a billion PoS shitcoins and sell myself one for ten dollars and suddenly, I have a valuation of ten billion. That doesn't make it less of a shitcoin. Every PoS coin is one such shitcoin.
But why you’re trying to lay your municipality’s busted, corrupt-ass utility’s issues at their feet is perplexing.
Ethereum miners at their worst are the maggots crawling around in your community’s failure to form a voting bloc to reform utilities.
At best they provide an arbitrary demand for electricity that makes solar panels obvious.
So I missed the part where you’re a newcomer troll.
Blame whomever you wish.
https://www.theverge.com/2018/3/16/17128678/plattsburgh-new-...
https://www.govtech.com/policy/washington-utility-will-uphol...
If we were to discuss other aspects of mining then yes, i would point to other greedy participants in this chain
Neither is perfectly “right”; it’s a direct trade off between “they’re available if you want one” and “they’re as inexpensive as you’ve come to expect from history”.
High end GPUs are readily available on EBay and other secondhand markets and I’ve resorted to buying them there to get them in for ML training machines. (Same with consoles, concert and sporting event tickets, and other limited production, desirable items.)
If demand exceeds supply, but the manufacturer continues to sell at normal prices, a scalper market opens up.
Scalpers capture the difference between manufacturer sale price and market value. That's their business, so they have more time to dedicate to the search for product than the "poor people" who "should" have access to the product.
Surgery is not a good example because there is no reseller market there. Medication would be closer but of course med sales are highly regulated. I agree with the point you're getting at though -- access to essential health care should not be financially bifurcated.
Don’t get me wrong, I love generality, but it’s clear that the trees don’t let you see the forest here.
Sellers were as greedy before the price rise, so greed is not the cause.
The difference here is that: 1) Supply difficulties mean there’s less in offer. 2) crypto miners have strong demand, since they can make money on those gpus.
I would argue that the market is working fairly well, since gpus are being allocated for their best use. Gaming is great, but people seem to really love their eth.
The way supply and demand works can be more subtle than that. Imagine two sellers selling the same widget, one has 50 units and is selling at $100, the other has 60 units and is selling at $120. As long as the demand is below 50 units, buyers will prefer the first seller, and the price will be $100. Once the demand gets above 50 units, the first seller has no more units, and the price is now $120 since buyers have no choice other than buy from the second seller. Notice that at no moment did any of the sellers increase their price. And the reason for the second seller starting with a higher price is most probably not greed, but higher costs (for instance, their store being in a place with a higher rent).
It would be interesting to see what percentage of manufacturing capacity NVIDIA is at, as 25%-50% would put ETH GPU mining at the majority of sales. Especially considering that supposedly higher volume cards such as the 3060 do a quarter of the hashrate of a 3080, which would be about 24 million cards. Of course, AMD is also part of the conversation as well, but I believe that they are quite a bit less volume than NVIDIA. Then of course, not all GPU miners are mining ETH necessarily and so on.
In other words, what if the opportunity to make money via crypto currencies has accelerated the development of GPU capabilities and availability beyond what would have happened if crypto currencies never came into existence?
I can't believe some people still think that Bitcoin is more speculative than major tech stocks. Facebook, Twitter, Google, Uber, Snapchat... All propped up by money printing. I think if we ran a simulation using hard (non-fiat) money, the net losses of these big tech companies would be even more significant than Bitcoin's losses on electricity. They're only profitable because of the fiat monetary scheme. If you remove the scheme, I think it's likely that revenue would dry up and profits would quickly go negative. These companies are just not viable in a hard money system. More profitable competitors would quickly spring up and there would be constant churn.
The reason why Bitcoin is so successful and anti-fragile is that it's located everywhere. It always has access to the easiest money from all around the world. It has placed itself at the root of every scheme around the world.
One of the hallmarks of a tech company is that the costs associated with producing the product is relatively flat and the revenue from the product goes up linearly with usage. Most tech companies then spend their money on recruiting more engineers, to try to build more products that are cheap to maintain and generate smooth rising revenues. The increased fed spending changes the stock price, but it’s not changing the fundamentals of why selling network access software is so good.
Oh wait
The reason why tulips are so successful and anti-fragile is that they can be grown everywhere.
By the way, if an activity like mining is only profitable because of money printing (depressing interest rates), given the world has been in zero to negative interest rates, this means the activity is not productive.
And you're mixing up profitability with stock price speculation, which are two very different things - witness the Gamestop incident. Once the stock is sold to the public, its price fluctuations don't actually affect the bottom line of the company unless there's a buyback.
Couple that with us going into 2 years of global lockdowns and subsequent supply disruptions have probably exacerbated the situation greatly.
We’re never going to return to the level of industrial output pre-Covid. There are a dozen high-likelihood scenarios down the pipe in the 2020s (Russia, China/Taiwan, climate drought / famine, etc) that will continue holding global productivity down. We’re looking at global stagflation for most of the rest of our lives as a result, meaning crypto is a good hedge. NVidia just needs to start pricing its top-tier GPUs for crypto farms, then refurbishing the last generation for consumers when the crypto farms upgrade.
Meanwhile, mining hardware demand depends on cryptocurrency prices, which go up and down like a yoyo. Only a food would invest billions in a new chip plant when demand might have collapsed by the time it's finished.
I might buy the latest GPU to play PC games at max detail. But there's no increase in value in me buying 20 GPUs so there's effectively a cap on how many GPUs the gaming market will demand. Same goes for other use-cases like scientific computing. In those cases ore GPUs do allow them to solve harder problems or finish calculations faster, but buying more GPUs is more expensive so eventually scientists hit the limits on their budgets.
Cryptocurrency mining is different though since it provides a direct path to turning more computing power into more income. As long as the marginal costs to mine on a GPU are below the expected return and the payback time for the fixed cost of buying the GPU in the first place isn't too long, it nearly always makes sense to buy more GPUs. To put it another way, if it makes economic sense for me to mine on 1 GPU, it makes sense for me to mine with 10, or 100, or 1000. In fact it probably makes more sense to mine with more GPUs since there are economies of scale involved. And since miners are more concerned with their marginal costs than the fixed costs, even huge increases in GPU prices don't alter their behavior much. As long as you're making money running a card, who cares if the payoff time for it increases a few months?
This isn't mainstream opinion of course, however I predict it will be. Crypto mining is the true first electricity buyer of last resort. It's a productive activity which can use megawatts of power and be turned off at a moments notice in order for the grid to meet the demands of society (as what is happening in Texas right this very moment). What other MW scale industrial process can turn off at the flip of a switch? It's a defacto subsidy on green power projects.
The HN crowd hates to see it, but crypto mining is becoming integrated at the deepest levels with our electricity grid. It won't happen everywhere as some jurisdictions may ban this "unholy union" but the moment public opinion changes, this integration of grid and crypto is a slam dunk no brainer.
It’s a profitable activity. There’s a difference there.
Seeing the title and the conclusions, how can this be the case? Computationally they're not much behind ASICs per $, are much more broadly available to buy and are easier to sell when needed.
Is this really true for Bitcoin? I’ve been out of the mining game for a long time, but ~5 years ago this certainly wasn’t true. Back then GPUs were at least an order of magnitude less efficient than ASICs.
For the same reason you might (needs detailed analysis) economically be better off overall by keeping the temperature in your house constant as opposed to saving energy with a schedule.
I'm not blaming the author, because determining causation is hard, and checking there's correlation between two variables can still be interesting, but the article falls short of answering the question.
if those non-crypto factors were confounding (say also happened to correlate with crypto-prices maybe due to some third unknown factor) then we would see CPU and GPU prices rise in response to crypto prices. we do not see that.
This. Those who can't afford anymore powerful video cards now fall back to CPUs with decent integrated graphics, so that the higher than normal demand is therefore affecting their prices too.
> We can see a correlation between GPU prices and the value of crypto currencies, but correlation does not mean causation. A causation is plausible in this case.
And then they go on to repeat a very reasonable argument, which I happen to believe to be true, but the kind of argument that everyone was already making before looking at any data.
The fact that the correlations do not hold as much for CPUs is good, but still not enough to stablish causation. As you said, there might be some confounding factors that affect crypto and GPUs but not CPUs.
The 3080ti’s are easier to find as they’re LHR and price per hashrate ratio isn’t good. What I mean by easier is that bots have a tendency of going after cards with price/ratio.
It doesn’t help either that DDR5 RAM is also highly scalped. PC memory is in direct competition to GPU memory and vice versa.
For real though, cut that out.
Once they successfully do the attack, they can take over the blockchains and allocate all cryptocurrencies to themselves, cash out, and it'd probably register as a good profit on their income statements.
ETH mining at 500TH (a solid chunk over 51% just for ease) would net you 300 ETH an hour...that's 1.1 million USD, which means you would have to be able to purchase millions of dollars worth of real world items that would be released to you within the time-frame of your attack (in this case an hour) to make this even approaching "worth it" from a risk/reward viewpoint and even that calculation is ignoring just how profitable they'd be just mining normally for the foreseeable future.
No one is waiting for just an hour worth of confirmations for millions of dollars of real world goods exchanged for ETH.
And there's plenty of reasons to want to do so: Proof-of-work currencies are what make ransomware payments possible, and there's a large daily CO2 cost for proof of work currencies. The question is simply when is the cost of the externalities greater than the cost of acting?
So short term: having the high prices paid by crypto people is nice. But it destroys their relationship to their old customer base. Depending on where the crypto market will be in 5, 10, 15 years, this can be the 5x of their market cap or the 0.2x
[0]: https://www.crowdstrike.com/cybersecurity-101/ransomware/his...
Assuming this was even possible, your plan has a fatal flaw.
Who the heck is going to buy the BTC back?
Why doesn't Nvidia use the cards themselves to mine? Not for the attack but for the profit?
Each card is obviously cheaper to produce than they are selling it for. Why not maximize all the profit?
No, not for Bitcoin. GPUs are no good at mining Bitcoin; you need an ASIC for that.
> Once they successfully do the attack, they can take over the blockchains and allocate all cryptocurrencies to themselves […]
No. Having a majority of mining power does not enable anyone to assign currency to themselves, except for the mining reward. That would require faking digital signatures.
I imagine that if Nvidia were to openly "steal" someone's bitcoin using a 51% attack then the victim could sue Nvidia in a US court and things could get interesting.
What laws would apply and what would the outcome of such a case be? And could the act also be considered criminal?
What if you are a drug dealer or dictator?
Nvidia wouldn't do this because it might hurt GPU sales.
"A GPU attack on BTC" Jésus help us
Since Bitmain asics are currently being made at around 16nm nodes, using 3nm process that Nvidia has access to - think about that for a bit.
If I remember correctly, the difference in transistor density improves as 1/alpha whereas area of transistors, if everything gets shrunk with alpha decrease, should decrease as square of 1/alpha. I.e., if alpha differs two times, there should be 4 times more shrunk transistors whereas practice shows only 2 (and often less) more shrunk transistors.
This is because you cannot make transistors too small in every dimension, they will leak.
So the difference between 3nm and 16nm will be less than five times. Not 25 times as difference between these parameters would suggest.
The difference in getting masks and first prototype should be quite substantial in terms of upfront payments and time-to-production.
I remember that 25 square millimeters proof of concept in 180nm process needed $50K of money and half an year of delay. The 35nm (5 times less) prototype of the same area would cost about $500K and more than half an year delay. The difference between 16nm and 3nm is about the same, I can expect several millions of dollars of upfront cost of small chip and year long delay before prototype arrives for 3nm process.
These Bitmain guys should be assumed to be not stupid and they most probably have access to latest processes after second prototype. Yet they choose 16nm - we must ask ourselves "why?" and Nvidia should too.
As a matter of fact, Nvidia had its share of problems with tiny processes in past.
"Clue" and "holding coin" have not so much overlap.
Well, assuming any kind of rationality in the crypto market is always hazardous: meme coins (doge coin, shiba inu) are worth billions in market cap, so do hard forks of bitcoin, and Ethereum didn't not collapse after they decided to rollback a bunch of things they didn't like during the DAO “hack” while they are advertising Ethereum as immutable and tampering resistant.
You can never be sure how this market will react in any event (well, except Elon tweets I guess)
1: https://medium.datadriveninvestor.com/itcoin-unhackable-it-h...
Usually it's just about the currency becoming worthless (usually only a problem in government controlled currencies that have loans in foreign currencies), or people just not accepting it anymore.
There's no inherent value in any currency, it's just a useful fiction because it's really hard to go get five goats to trade somebody for a laptop.
Given the number of countries using the dollar as their reserve currency, value could collapse quite a ways without the US doing anything.
It's not likely, but neither is a 50 + 1 attack on bitcoin given the amount of computing power currently tied up there.
Also, Bitcoin has significantly more vulnerabilities than just a 51% attack or a crash as nodes need to be connected to the internet and therefore can at least in theory be hacked. What exactly happens after that point is anyones guess, but it could easily be worse than a traditional 51% attack.
Sure, but the value of the dollar (like everything else) is in how easy it is to get (supply). The US GDP is not a measure of the supply of dollars on the market, it is a measure of how often those dollars are exchanged.
> Bitcoin has significantly more vulnerabilities than just a 51% attack or a crash as nodes need to be connected to the internet and therefore can at least in theory be hacked.
That's fair, if we're including implementation vulnerabilities, it does expand the potential problems quite a bit. The post I was responding to was specifically about a 50 + 1 attack. Though if we include these, we should probably also expand physical currencies' problems to include counterfeiting in that case.
The US GDP directly translates into the amount of taxes paid. Suppose a 50% drop in the value of USD happens. If nothing else happens taxes increase by 100%, and people need to acquire more USD which drives the value of the currency up.
You might wonder about economic hardship, but historically that causes deflation. +/- 2% per year without printing money is easy, but there is a reason you don’t see huge swings without the government printing money.
> counterfeiting
An issue for any physical currency but electronic currency actually operates independently. Trying to counterfeit large quantities of cash is therefore surprisingly difficult even for nation stale level actors. A great fake bill can get passed around, 10 billion in fake bills which is still a trivial amount would trip up various monitoring systems.
Oh definitely. I did not mean to imply that taxation in USD does not have a modulating effect (and if I came across that way, I apologize). I'm just trying to say that you can have some pretty wide swings even with that.
> but there is a reason you don’t see huge swings without the government printing money.
You don't generally see it, no. Mostly because other countries use the dollar as a store of wealth, and they don't want to tank the value that they've squirreled away. I'm bringing it up as a hypothetical with the same likelihood as a 50 + 1 attack on bitcoin.
China, for example, seems to hold ~2 trillion in USD (~60% of 3.5% trillion total reserves).[0] I assume most of this is held as digital assets, as there's only ~2 trillion physical dollars in circulation.[1] Dumping that much on the market would cause some pretty wacky inflation (as well as destroy China's stability). In a similar vein a 50 + 1 attack on bitcoin would allow an actor to give themselves all of the bitcoins, but would also destroy the value of those bitcoins because nobody is going to trust the network after that.
Both of these attacks are really only useful if you value hurting other actors more than you value your own resources/your own sense of self-preservation.
[0]: https://en.wikipedia.org/wiki/Foreign-exchange_reserves_of_C... [1]: https://www.uscurrency.gov/life-cycle/data/circulation
? GDP is produced and denominated in USD, domestic taxes are paid in USD. How does a drop in the relative value of USD compared to a foreign currency influence domestically paid taxes in any way?
I used to assume that, but there are coins which have been attacked successfully before and people still use them anyway. I don't attempt to explain the logic there :P
See, if they just stopped selling them, people would ask questions. If, however, you just reduced the supply by like 80%, said it "was due to all those crypto miners hoarding them", and then proceeded to use that hoarded 80% to do a 50%+1 attack, you'd be able to -quietly- take over, and turn around and sell those coins to the still preserved market.
applies tinfoil hat In fact, who is to say that isn't exactly what they're doing, and hence the existing shortage?
I go back in time with a super computer and mine all the coins the first year and lock them up in my wallet. I then return to the present day.
Are they worth anything?
So they're going to manage to not sell any cards (without disclosing that as a public company), set up the infrastructure and do a 51% attack on bitcoin in front of everyone, then "cash out" as if there would be people lined up waiting to buy bitcoin when this has just happened in front of their eyes?
Unlikely.
https://www.windowscentral.com/best-gpus-crypto-mining
That’s at least six months off and has already been delayed repeatedly.
I have mined and sold $2,800 of ETH in the 9 months I’ve owned it. The extra power has cost me a couple hundred dollars.
I’ve made anywhere from $3 to $75 per day during that time (thank you Shiba Inu launch date for the highest numbers).
I’m continuing to make about $3 per day. You can look up current profitability for any graphics card at whattomine.com.
Bottom line is I could have added infinite graphics cards to this setup and the profit would’ve gone up linearly.
Anyone that doesn’t think GPUs are high due to ETH mining has no idea what they’re talking about. The analysis from OP seemed ridiculous to me. I didn’t even see LHR mentioned in the article which are GPU manufacturers attempting to make GPUs unprofitable for miners so they can only be used for gaming. (Hint: didn’t work very well).
If GPUs would still use the fixed rendering pipeline they started with ( only push your triangles and texture to the GPU and you get your image) they could also handle their business with a few hundred employees.
Just shrink your chips to the latest fab size, add more computing cores, maybe improve some algorithm if a better solution is found.
Its like the Oak Island treasure. If they ever actually dig something up, it'll kill the tourist revenue.
This also coincides with the governments sudden desire to regulate the industry despite it existing for nearly a decade now. Its a leak in the their system. You can't print money and hand it out where you want if someone has a hedge against your money printers.
Mining like with politics is just some easy target to blame, the problem is Monopolies and duopolies.
Events have catalysed a market rife with monopolies and duopolies: these rule the supply chain, they have seen the true prices the market will accept, how can we expect prices to go back to "normal"?
In other words, there is visibly a shortage of graphics cards that is not affecting any other components. That strongly suggests that there is something driving demand for graphics cards that is not driving demand for other key components of computers, which are available in sufficient supply. And the list of things that extra demand could be coming from is rather short indeed.