Live data from Hacker News

One Bitcoin miner is buying 20,000 16nm wafers from TSMC per month

dvhardware.net

71–80 of 231 posts

Re: One Bitcoin miner is buying 20,000 16nm wafers from TSMC per month

#71
Anything that mentions bitcoin mining gets people worked up and negative for some reason. More demand for compute is a good thing long-term. More demand for video cards is great for all parallel computing long-term, hopefully someone besides NVIDIA seriously moves into the space.

Re: One Bitcoin miner is buying 20,000 16nm wafers from TSMC per month

#72
post #32

Who makes money in a gold rush? It's the guy selling shovels.

That was my thinking too. There are a lot of replies in this thread talking about Bitmain like they're the miner, when in reality they're the mining supply store selling miners the shovels, hard hats, and so on. Mining itself may be profitable, or it may not. But no matter what, the supplier should come out with a fairly healthy profit if they play their cards right.

Maybe. The lead time on this stuff is not insignificant. It's not that simple. Bitmain/Jihan understand that; they take a varied approach.

Re: One Bitcoin miner is buying 20,000 16nm wafers from TSMC per month

#73
post #50

Earlier quoted context omitted.

That's totally fair. Google's parts that I know about are security coprocessors (like a TPM) that wouldn't really need the newest process. They probably have NICs too that might make sense at that node though.

Their NICs probably use off the shelf ICs from Broadcom or similar. In other words, Google would only handle PCB layout and up. Besides, modern analog ICs are usually fabbed using custom processes and in-house lower node fabs. Designing high-speed RF ICs is a huge effort, and as far as I know, Google does not have such capability​.

IDK, they're still actively supporting the LLVM backend for Myricom's LANai chips, despite Myricom being sold off a few years ago.

And the NICs don't really need custom high speed RF, the foundry has a SERDES block that'll work just fine for connecting to a phy.

Re: One Bitcoin miner is buying 20,000 16nm wafers from TSMC per month

#74
post #50

Earlier quoted context omitted.

That's totally fair. Google's parts that I know about are security coprocessors (like a TPM) that wouldn't really need the newest process. They probably have NICs too that might make sense at that node though.

Their NICs probably use off the shelf ICs from Broadcom or similar. In other words, Google would only handle PCB layout and up. Besides, modern analog ICs are usually fabbed using custom processes and in-house lower node fabs. Designing high-speed RF ICs is a huge effort, and as far as I know, Google does not have such capability​.

Google has a custom NIC ASIC derived from the Myri-10G family. Google has also designed the TPUv1, TPUv2, Visual Core, and Titan chips (that we know of).

Re: One Bitcoin miner is buying 20,000 16nm wafers from TSMC per month

#75
post #13

Earlier quoted context omitted.

Nothing more useful for humanity, I think. It is literally mining electronic gold which facilitates transactions removing need for trusted third party. You may not ever pay for something "useful" in your life in bitcoin, like you've probably never paid with gold, which is still useful for governments and big investors as store of value.

The gold that is mined has useful industrial application. Not so sure about Cryptocurrency..

I think you are looking at it wrong; its like saying a bank teller or stockbroker or even a cash machine has no 'useful industrial application'. They are necessary for the currency to work, and they must be paid to do a job. I suspect paying a cryptocurrency miner their 'fee' is a lot cheaper (per transaction).

Re: One Bitcoin miner is buying 20,000 16nm wafers from TSMC per month

#76

As a cryptocurrency-confused individual, I read articles like this and wonder how I'm supposed to assign any meaning to the "value" being created here, and by blockchain PoW in general. So this company is buying a bunch of wafers... He's converting the value of the wafers into value on blockchain, I guess? Because I can't help but interpret these things as "some people are wasting a lot of electricity, and you should…

Perhaps it might help to consider it equivalent to one converting the value of picks and buckets into value in gold. Just as the picks and buckets are worth what they’re worth, the gold is worth what someone else is willing to pay.

Re: One Bitcoin miner is buying 20,000 16nm wafers from TSMC per month

#77

From the article- Bitmain, the largest Bitcoin miner on the planet, is now buying a whopping 20,000 16nm wafers a month from TSMC! That's double as much as in the previous quarter and a higher volume than what NVIDIA orders from TSMC. Related material- https://www.gmo.jp/en/news/article/?id=764 A competitor to Bitmain just finished their prototype 12nm Bitcoin mining chip and is moving forward to their target 7nm min…

How long do you think it takes them to amortize them?

Re: One Bitcoin miner is buying 20,000 16nm wafers from TSMC per month

#78
post #61

Earlier quoted context omitted.

Wouldn't the difficult per block rise, making it the same current usage as before?

Eventually. But newer hardware will tend to be the most efficient component of the hashrate and push out older hardware by generating more hashes per unit of electricity (hence the CPU->GPU->ASIC progression).

That doesn't mean anything.

Bitcoin network bandwidth remains the same, and energy usage has continued to increase in a race to the bottom in terms of work and waste for bitcoin mining - this is by design of how Satoshi's difficulty algorithm reduces efficiency with increase in hash rates.

Re: One Bitcoin miner is buying 20,000 16nm wafers from TSMC per month

#79
post #61

Earlier quoted context omitted.

The more efficient your hardware is the more profit you make per block (because you buy less electricity per block reward). If all hardware was equivalent and an unlimited amount were available efficiency wouldn't matter, but neither of those things are true.

Wouldn't the difficult per block rise, making it the same current usage as before?

The difficulty rises with the total hashrate across all miners, not on the individual hashrate/efficiency of any particular miner - if you can eke out a temporary efficiency advantage over everybody else, you can profit nicely while they're trying to catch up.

Re: One Bitcoin miner is buying 20,000 16nm wafers from TSMC per month

#80
post #74
post #50

Earlier quoted context omitted.

Their NICs probably use off the shelf ICs from Broadcom or similar. In other words, Google would only handle PCB layout and up. Besides, modern analog ICs are usually fabbed using custom processes and in-house lower node fabs. Designing high-speed RF ICs is a huge effort, and as far as I know, Google does not have such capability​.

Google has a custom NIC ASIC derived from the Myri-10G family. Google has also designed the TPUv1, TPUv2, Visual Core, and Titan chips (that we know of).

TPUs are digital ICs. I was talking about analog ICs.

That Myri IC seems interesting though.

Post reply on HN