Please provide data to substantiate this claim as it is completely contradictory to the last 50 years of manufacturing engineering knowledge of integrated circuits.
While the infant mortality rate dominates until year 10 of operation, it is accepted practice to run the device for at least the equivalent of 1 YEAR, to eliminate the majority of infant failures. So burn in is not days, unless you use a highly accelerated method such as temp chambers to simulate much longer operation as stated by @Kahooli.
While you can argue Innosilicon motives, you cannot argue that a pre-mined unit is less reliable than a new unit, as this is simply NOT true.
I don’t disagree with you at all. You’re totally right about the potential of failure of pre-mined versus non-pre-mined.
To clarify, I would rather take the chance (odds) of not having a bad unit, rather than having a working unit that was pre-mined. On the whole, what you’re saying is correct and overall pre-mining will decrease the number of RMAs of units shipped out - no disagreement there. I would rather do that pre-mining myself, get the rewards from it, and if something happens then it’s still within the warranty period. Pre-mining does decrease the lifespan of completely-working unit.
No, sorry that is actually NOT correct. Pre-mining does not decrease the lifespan of a working unit. Normal lifespan failures do not begin to appear until ~10 years of operation. Everything before that is infant failures. Therefore for a device that has an expected useful life of less than 10 years (as does a miner), not pre-mining will actually decrease the life expectancy, again NOT the other way around.
Most consumer electronics and ALL medical, military, and space electronics have extended burn in to eliminate infant mortality. Mission critical systems cycle temp, bias, humidity, and even shock to precipitate out infant failure before the active service life of the system.
So you expect Innosilicon to cover all of the infant mortality failure costs in their warranty??? I’m not sure how you could possibly run a successful modern electronics business that way.
Innosilicon is simply doing what every other electronic manufacturer in the world does. The only difference is they can actually make money doing it. So again, you can question their motives but it would be a poor business decision to implement accelerated stress screening on ASIC miners when a simple static screen/pre-mine represents a positive revenue stream AND reduces RMA’s at the same time.
Interesting to see that an attack on Zcash is only NiceHash-able at 6%, while an attack on Bcash is NiceHash-able at 12%, so there’s a lot less Equihash hashpower avaiblable for rent on NiceHash vs. SHA256. This could change as more ASICs come online.
It could, but so far it’s reverse. Bevor Asics the nicehash able hashrate for Zcash was 10%, now it’s down to 6%. So far it’s a security improvement. But i agree, someone must follow where it goes the next months, but i personally think the trend will continue and it will be in the 5% long time range.
Vitalik is proposing that if an independent observer of the network traffic (i.e. just the blockchain client a user is running, not a miner/validator) watches what’s happening in real time and pays attention to when messages appear, they can detect ‘foul play’ by miners performing a 51% attack and this can provide additional safety guarantees that can protect against such an attack.
Pretty sure this part/algo will be on the POS side after it’s mentioned that it won’t be done by miners, at least this is my logical conclusion. POW mining + POS staking and observing.
That’s the whole point, the verification would no longer be dependent on the miners but the nodes (independent observers) for consensus, thereby making a 51% attack by miners no longer possible.
First, your data is from general consumer electronics. You haven’t specified an optimal burn-in rate for ASICs, surely at some point they should stop pre-mining after that safety burn-in (a month? two? three?). Further, I don’t think ASICs are designed to last 10 years. If anyone has had a large farm with early Bitcoin miners that can shed light on the failure rate, I would be interested in that data, but it surely doesn’t match general consumer electronics or the environmental conditions that some of these miners are placed in (Vietnam, etc.).
My initial statement stands that I would rather take the chances getting a good unit (97-99% odds? Just a guess), lasting 2-5 years than with a burn-in/pre-mined unit lasting 1-4 years.
It’s not that I’m stubborn and won’t alter my opinion with facts, it’s that I think the data you are using are for consumer-grade electronics that doesn’t accurately translate into the flavor of operation that ASIC mining entails.
There’s nothing special about these asics. They’re built on relatively mature silicon processes and have the same failure modes as any other microprocessor built on that process.
The data I present is for ALL silicon based Integrated Circuits / components, and DIRECTLY applies to ASIC’s and ASIC miners. As @Kahooli stated “There is nothing special about these ASIC’s” and this is well known engineering data that has been around for half a century.
I don’t know how to help you understand integrated circuit design. You don’t design an ASIC with a lifetime in mind. Its never a concern since modern integrated circuits last for DECADES without any design considerations at all. The Weibull distribution for the VLSI IC process (from the 1970’s) has a 50% failure rate at 100 YEARS!
Does that lifecycle of average consumer electronics chart account for the fact that the ASIC will be ran at 100% speed/heatsoak or more for it’s entire operating time?
A better analogy would be a comparison of the average GPU lifespan vs GPUs that have been overclocked at 100%+. Even still, GPUs don’t suffer dramatic lifespan reductions during overclocked conditions as long as you keep them cool enough
The last factor to consider is the build quality of an ASIC vs a GPU. There aren’t any direct comparisons that I know of but I would be willing to bet that a mining ASIC is not as high “quality” as a consumer grade Nvidia GPU (heat dissapation studies, fan design, thermal compound quality, etc…).
Either way, I would agree with @root that getting a new consumer product is obviously more desirable than a used one. It’s like asking if a used car will fail sooner than a new one, the answer is not if you take care of it but you don’t get to enjoy that new car smell
MTTF inversely follows temperature, and yes it is log mean as long as the max temp never exceed silicon maximums. An RF Power FET for example can have MMTFs in the 1x10^6 hours below 170C
A little apples and oranges but still…
Don’t mistake my defense of stress testing as advocating for them actually using them to mine coin. They should test on test net if that’s really what they’re doing. In a clean, temperature controlled environment.
Do you think your intel processor hasn’t seen a single turn on before it’s delivered to you? Probably not. Would you be mad if part of their “qualification process” was a month or two in a datacenter? Heck yeah!
Not an expert here when it comes to lifecycle and lifespan, just putting in some thoughts, points and some experience.
is there a difference in these charts if an asic/gpu/fpga runs at 75%, 100%, 125% or even 150%?
does it matter/make a difference if an asic runs at 25C (Baikals for example), at 50c or even at 80c temperature?
while an Asic is designed for 24/7 work i doubt that gpu’s are designed for it. The industrial fans on the asics are a good indicator for that.
From my personal experience with running a bit below 100 asics and until some weeks 12x6 1080ti gpu rigs all time results:
Asics: 2x D3 boards stopped working and 1x Orange Pi on a Baikal too. L3, S9, B3, X3, (Z9m & E3 too new anyway) no problems.
GPUs: 5x Fan bearing problems, 4x total destroyed gpu’s (all on garantee) but took half a year to get them replaced, 2x not working properly gpu’s out of 3 on my home/work PC (mostly bearings too).
My conclusion is that gpu’s aren’t just designed for 24/7 work, hence i have more failures on them than with the asics.
