All Eyes On Graphics Card Shortage, Few Answers Forthcoming

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.

InvalidError

Titan
Moderator

Nothing. The only thing you could remove from GPUs to make them unusable for mining is GPGPU support. If you removed that, this would ruin CUDA, OpenCL, DirectCompute, etc.

Ethereum GPU mining will die down once far more efficient ASIC miners become available. Nvidia and AMD don't want to massively increase die production because once that happens, GPU demand will die practically overnight and the used market will be flooded with used cards that will in turn render selling surplus dies and boards nearly impossible.

No one wants to write off millions of dollars worth of unsold GPU inventory when the Ethereum GPU hashing craze dies. Expect manufacturers to be very cautious with increasing production.
 

lp231

Splendid
Gpus that are mined with are put under constant extreme load in a short period of time, compared to a span of years under normal use. This is why 2nd hand mining cards are close to their functioning limit. Down on the microscopic levels, the soldering and such have already been worned out.
As for nzalog, either seem to be completely misinformed or is a "paid shill".
 

spagafus

Prominent
Jul 6, 2017
6
0
510


They don't wear out but they do become too expensive for the miners to run them. If anything a card that has survived useful life in a mining rig is more likely to last forever in a gaming rig.

Though moving parts like the fans may be near end of life.

 

iPanda

Reputable
Mar 25, 2015
87
0
4,640
I heard news that some manufacturers are worried about RMA abuse (like what happened with bitcoin crash). Basically if they are able to RMA card and then just resell that 'new' card at a premium. The manufacturers lose out on that end of the deal (not like miners have brand loyalty or anything). So I see that also compounding the issue. Hell, that would affect gamers who actually have to RMA a card and get an replacement as well (longer queue).
 

gnarly_90

Distinguished
Oct 13, 2009
2
0
18,510
This is ultimately good for gamers and the GPU industry. Once we have a solid baseline for the crypto demand, the GPU manufactures can justify the cost to produce more, higher yields will bring down average costs too. GPU companies having a new revenue path for the same cores we gamers need is also good. We want healthy competition and we want our GPU companies to succeed; not go under or be funding limited for R&D. While this may "suck" right now, it'll either pass, mining will become a passing fad, or the market will adjust, should mining become a new consistent and predictable revenue stream. Patience, this'll be a good thing.
 

none12345

Distinguished
Apr 27, 2013
431
2
18,785
As far as GPUs wearing out. These mining loads are not that bad on the hardware.

There are a couple different avenues for hardware death.

1) Heat cycles. Going from cold to hot to cold to hot to cold is very bad for electronics. Materials expand/contract when temperature changes. This puts mechanical stress on the solder joints and they will eventually fail. (ive had a gpu and mobo from ~2009 fail in the last couple years due to this problem)

2) Voltage. There is a sweet spot for any eletronics for voltage vs lifetime. The more voltage you try to stuff through something the shorter its life. (look up electromigration)

3) For fans mainly, the higher the rpm, and the more its spinning the faster you wear it out. Start/stop cycles are also not good for fans.

Mining wins on 2 of these, and loses on the 3rd. Miners run their cards 24/7 keeping them at a more or less constant temperature. They accrue very few heat cycles. They also generally run them undervolted to hit an efficiency sweat spot, less voltage is less electromigration(also less heat). Conversely gamers are far more likely to raise the voltage, increasing the likelyhood of electromigration. Mining loses on number 3. While there are less fan start/stop cycles, there is a far higher use of the fans lifetime. Tho you can mitigate this by the fact that you can replace fans fairly easily.

Id much rather have a used miner card then a used gamer card. Id likely have to replace fans at some point, but the chips should be in better shape.
 


It is a good question, though I think looking at this the other way around might make more sense. What could manufacturers change on mining-specific cards to encourage large-scale miners to buy them instead of gaming cards? By selling the miners cards that are more efficient for mining but unusable for gaming, they could avoid a situation where the market is eventually flooded with used gaming cards when a popular cryptocurrency goes bust.

One thing I can think of would be multi-gpu single-card designs. I suspect that packing multiple, undervolted GPU cores on a single card would require less power for a given amount of computational workload, and allow more GPUs to be installed in the same amount of space. There might also be less cost involved in building a card equipped with four GPUs than four cards equipped with one. So, by buying a card like that, the miners might not only pay less for the card initially, but also pay less in ongoing electrical costs.

These cards could also be made unusable for gaming to prevent them from potentially flooding the market down the line, while further optimizing their production costs, by omitting anything not required for mining. For mining, they might not require anything more than a PCIe x1 connection, so there should be no need to wire up traces for a whole x16 slot. They shouldn't need any sort of display outputs or associated hardware either. Plus, they would only require enough cooling to handle the card in its undervolted state, so the coolers wouldn't need to be overbuilt to handle gaming loads. The only issue I could potentially see with making the cards unsuitable for gaming is that some miners might not be willing to invest in cards that might not have much resale value once they are unprofitable for mining. On the other hand, if the reduced production costs were to make them notably less expensive to begin with, that might help make up for it.

There's also the issue of the supply of the GPU wafers themselves though. Nvidia and AMD might not want to make a huge production run of GPU wafers planned for cryptocurrency cards, when the popularity of those cryptocurrencies can vary wildly. Of course, those chips could always be put into gaming cards instead if that were to happen, and they wouldn't have to worry as much about those cards having to compete with used mining cards.


That's another thing that could be added to make gaming cards used for mining less attractive for resale. Lots of other computer hardware has usage tracking built-in, so why not graphics cards? Hard drives have SMART data that records these kinds of things, allowing a person to check how many hours the drive has been powered on, how much of that time the drive was sleeping, how many times it spins up, its maximum recorded operating temperatures, and so on. Some other hardware like certain monitors and motherboards track usage hours as well. It wouldn't be hard for graphics cards to record that kind of data, and make it easily viewable from the card's control panel. That way, someone could see if a card has been running at load for 5,000 hours, with an average operating temperature of 80C, for example, or if a prior owner had the card heavily overclocked or constantly thermal throttling. If this information were easily accessible, people might come to expect this data being listed when buying a used card.
 

InvalidError

Titan
Moderator

There is one major problem with that: the whole point why crypto-currency developers are trying so hard to make their algorithms inconvenient for ASICs is because it enables people who can afford to get ASICs made to hijack the block chain. Once that happens, the currency cannot be trusted anymore.

By making Ethereum heavily dependent on large amounts of RAM, it lowers the ceiling on how cost-effectively ASICs can scale as the cost and performance quickly end up dictated by memory instead of processing power.
 

103

Reputable
Jul 12, 2016
5
0
4,510
Of course computer hardware wears out, but how delicate are these graphics cards? After all, they are the smaller cousins of GPU's designed to run 24/7 in data centers and supercomputers. Suppose a graphics card has a life expectancy of fifteen years under "normal" use. If a miner takes ten years off its life in one year, it will still have five years of useful life and be outdated/worthless by the time it dies anyway.
 
Today I sold my two 970s in SLI for $480. Two months ago I would have only gotten $300 for them. Those of us who are on the fence upgrading and have recent generation GPUs for sale can take full advantaged of the overvalued used GPU market as well. It pushed me to jump for a 1080Ti upgrade for $260 net out of pocket (EVGA SC2 GAMING). My original plan was to wait until next year for the consumer Voltas to come out. I couldn't pass this GPU seller's market up.
 

Ncogneto

Distinguished
Dec 31, 2007
2,355
53
19,870



With all due respect, any manufacturer worth his/her salt is going to position themselves in such a way as to make maximum profits. To that end, a ramp up in production is all but inevitable. Now, this ramp up would most certainly be slow and gradual, unlike a knee jerk response that would lead them to a glut in overstock.

Furthermore, I would expect rather than modify their cards to make them less attractive to miners (and whoever came up with that idea is generally clueless), I would suspect they are paying very close attention and making small tweaks to their cards to make them more attractive to the mining crowd (when feasible).
 


While I agree with most of what you are saying here there is a catch to Option 1: Yes this will give the company's the greatest returns in there investments, leaving the market stagnant and allowing the the cards to fetch a premium ( The Fast and the Furious, a week before Race Wars reference) this forces the consumers to start looking into other alternatives for there parts leaving the market ripe for whom ever can get the chips to market faster. This also has a side effect of loosing the consumers confidence in them opening the market up for others to try to enter. The basics of supply and demand. Right now there is a high demand from the base consumers ( "base" being the normal consumers that purchase these components when they get a new machine whether it is a company or privet ) because there are no GPU's to be had a the normal price because of the Cryptocurrancy. This can also fall into price manipulation( market manipulation) if these company's, AMD, Nvidia, are not careful. If they intentionally hold back already manufactured chips this would be Market Manipulation: Market manipulation is a deliberate attempt to interfere with the free and fair operation of the market and create artificial, false or misleading appearances with respect to the price of, or market for, a security, commodity or currency. Lets look at AMD's RX 580 as an example. XFX had there's reviewed here on Tom's Hardware and at the time of the writing it was going for $279 USD and now it is $600+ USD over double the price.

So I ask you, do you really think Option 1 is the least risky? Because if the price jump I'm sure a few eyebrows have been raised in the FTC.
 

InvalidError

Titan
Moderator

That would be dangerous as that would also introduce the possibility of false positive that would kill your PC for no reason. Also, deliberately crashing someone's PC would likely be considered a computer crime and make the manufacturers' liable for losses.
 

Nintendork

Distinguished
Dec 22, 2008
464
0
18,780
The problem is that mining "friendly" gpu's have pretty much no resale value for a gamer on the 2nd hand market since they're so limited in video outputs, quality of the PCB or ventilation grills (miners use them without a case so it's no an issue).
 
NZalog,
Lucky you...
I had a motherboard die, a graphics card die, 3xmonitors, WD media player, M-Audio speakers, Auzentech sound card (replacement has lasted years), Corsair TX750 PSU, EVGA GS750 (brand new lasted two hours), 2x XBOX 360 USB receiver for PC (from Microsoft), 3xHDD's (internal), 4xHDD's (USB... though it was the AC adaptor and SATA->USB electronics that were dead so I recovered)

That's all for one guy over a ten-year period and I know there was a few more things. On top of that, I always do a lot of RESEARCH to look at quality. Considering my sampling over many different manufacturers I think it's a big PROBLEM.
 


[EDITED]
I can understand the other points of view here as to why the opposition of mining and/or specialized mining cards (and actually, some specialized cards have been, or are in the process of being, released.

But why down-vote instead of just correct what you feel is wrong with this guy's post? ESPECIALLY by the drive-by down voters?
.
 

InvalidError

Titan
Moderator

That's assuming the failures were from internal causes. Bad AC power can kill PSUs prematurely and bad PSUs can kill other components prematurely.

Since I've started preemptively inspecting and replacing generic electrolytic capacitors in devices I want to continue using well beyond the warranty a decade ago, I have practically eliminated failures.
 

nzalog

Respectable
BANNED
Jan 2, 2017
541
0
2,160


Well first off HDDs should not ever be something you expect to last forever, as with many things with lots of moving parts. If you can get at least 4 years out of them that should be considered a success. They also tend to last longer when you don't let them spin down and keep them somewhat cool.

Also things that fail 2 hours in are faulty you can't count that either.

One habit I've always had, is always checking on temperatures of my electronics and correcting high temp issues. You can't expect electronics that are overheating to last... Invest in a ir thermometer gun and that should help you figure out components that are just running too hot.

Most hardware that has failed for me, I could somehow trace back to something I might have done wrong. Like a PSU where I made custom cabling for (don't this it was close enough to stock, worked for a long time then started to act up), or a processor that I might have over-volted too much due to loadline calibration being too aggressive while overclocking.

Lastly, for the most part I kind of gave up on overclocking (except in really rare cases where is obvious there is a ton of headroom, but I can not give you a good recent example of this).


 

valeman2012

Distinguished
Apr 10, 2012
1,272
11
19,315

All i saying AMD or Nvidia should prevent mining from non-specialized mining graphics card.

Let the Gamers buy the Gaming Graphics Cards (10 Series)
and
Let the Miners buy the Specialized Ming Graphics Card (and the Mining GPU will be the one having the supply issues)
 
Status
Not open for further replies.