Does Google's new TPU chip suggest a new era of mining?

[Butts]

http://www.zdnet.com/article/tpu-is-15x-to-30x-faster-than-gpus-and-cpus-google-says/

I find this fascinating.
Moreover, the impact on the crypto-world could be next level.
I truly applaud their research into this new technology, however its got me asking questions;

• Google is in the pursuit of true AI, but what would stop them Google working on their own currency?
• Could TPU's have an ASIC-like effect on GPU mining?

Long time watcher, first time poster.
Peace.

[1715030444]

1715030444

[1715030444]

1715030444

[NiHaoMike]

They won't do anything for Bitcoin (ASICs already push the limits of what's possible) but could possibly enable a class of altcoins that mine well on them.

[QuintLeo]

These TPUs are a Machine Learning ASIC - not usable at all for mining on for ANY existing coin.

 I see ZERO impact of these on mining.

[c0inHnt3r]

This is a very exciting technology. Imagine a world of self-improving ASIC miners. A miner that finds new ways to optimize the way it works. It could potentially put massive bitcoin mining data centers out of work. I see the potential of it being like when the small USB block erupters came out. If you can do more with less space and power why wouldn't you.

[NotFuzzyWarm]

This is a very exciting technology. Imagine a world of self-improving ASIC miners. A miner that finds new ways to optimize the way it works. It could potentially put massive bitcoin mining data centers out of work. I see the potential of it being like when the small USB block erupters came out. If you can do more with less space and power why wouldn't you.

ROFL...
Riiigght... Oh the Wonder of bright-eyed youth, untainted with knowledge of how their technical toys actually work...

[Carlton Banks]

I hate to have to bring this up again.


But where is Google's revenue coming from to pay for all these cutting edge software and hardware projects?

It cannot possibly be from advertising revenue. Who's going to tell me they're the 1 person in the whole world that clicks Google Searh's sponsored links instead of the link they were actually searching for, or that they don't click Skip Ad instantly 99% of the time when using Youtube?

What advertiser in their right mind would pay Google any significant amount of money for such a service, and even if they do (which is incomprehensible), surely running the website with the largest amount of storage and the amongst the largest consumers of bandwidth on the internet eats a significant amount of any of this unlikely Youtube ad revenues?


Where could Google (and they're now Alphabet, Google is simply a division) possibly be getting all the money they need to pay top hardware and software engineers for long term self-driving car projects, multiple operating systems, buying and running robotics divisions, free cloud storage platforms, the list goes on and on, and the more one thinks about it, the more bizarre their revenue model appears.

Does Google Fiber and Chromebook sales produce enough revenue to pay for all this, even if we add that to the likely small revenues from advertising? I doubt that very much.


Where is the money coming from? That much is certain, Google are getting a huge amount of money from somewhere, and they're not using any of thei tech to generate sufficient profits to pay for all their product development and services.

[kano]

Advertising.

[NotFuzzyWarm]

Who's going to tell me they're the 1 person in the whole world that clicks Google Searh's sponsored links instead of the link they were actually searching for, or that they don't click Skip Ad instantly 99% of the time when using Youtube?

<Raises hand> If it is a product that I am looking for, yes, I often DO click sponsored links.
YouTube ads yes, mostly but not always skipped. Some can be even amusing.

[VentMine]

Online advertising is big bucks... for now, anyway.

Activate lightning network, and that revenue will vanish pretty quick when the middleman is cut out.

[Carlton Banks]

I don't find the advertising argument very convincing.


If you were the advertiser, be honest, how much would you pay for Google Ads? Considering how effective it is, I doubt it would be very much.

And think, Google have a huge amount of R&D, some of which is never commercialised, all of which requires top professionals in their field. It doesn't seem to add up. How can Google really be the biggest technology company right now, and probably the biggest ever, just from the advertising revenue that 80% or 90% or users never engage with?

[SaBit]

I don't find the advertising argument very convincing.


If you were the advertiser, be honest, how much would you pay for Google Ads? Considering how effective it is, I doubt it would be very much.

And think, Google have a huge amount of R&D, some of which is never commercialised, all of which requires top professionals in their field. It doesn't seem to add up. How can Google really be the biggest technology company right now, and probably the biggest ever, just from the advertising revenue that 80% or 90% or users never engage with?

Are you kidding. That's one of the funniest things I have ever read. You should consider a career in standup.

https://www.statista.com/statistics/266249/advertising-revenue-of-google/

[Carlton Banks]

That's just numbers on a page, no proof at all. Where's the proof?


I ask again, if you were any business, large or small, how much would you pay for Google advertising? My answer would be "close to zero", because it's just not effective advertising at all when the user can so easily ignore it.

Are huge corporations really so uninterested or ignorant as to how ineffective web advertising is? Really? C'mon, you people are just believing anything that's served up to you in a bar chart. I suppose you believe the accounts "information" that your governments publish for how tax revenue is spent, too?

[SaBit]

Do you have any real world experience in advertising?

The organisation I work for is a medium sized retail chain in a relatively small country (ranked 41 by gdp) , and around 20% of our media spend is with Google.  Our annual Google spend is close to a million dollars,  so those numbers are not just numbers on a page.

They are very realistic numbers!

[NotFuzzyWarm]

Then throw in the fact that Google also hosts the networks for many businesses and cities. That alone is one helluva chunk of change.

Of course the doubting Thomas here could just look up Googles public P&E statements that are filed with the SEC. Or are those also just numbers on a webpage?

For that matter, where is your *proof* that none of this is true? Refusing to believe is one thing, attempting to argue your point against others without providing at least some sort of evidence to support your view is just pointless rhetoric.

[QuintLeo]

That's just numbers on a page, no proof at all. Where's the proof?

 SEC filings for Google are the proof - as I recall they do break down income by "source" in at least fairly wide catagories, one of which is "advertising" which is an amazingly high number and has been so for many years.

[Carlton Banks]

That's just numbers on a page, no proof at all. Where's the proof?

 SEC filings for Google are the proof - as I recall they do break down income by "source" in at least fairly wide catagories, one of which is "advertising" which is an amazingly high number and has been so for many years.

No, that's also just numbers on a page. Since when did the SEC and Google develop so trustworthy a reputation that we can believe anything written in a report they write, that contains no actual evidence?

Face it, believing the contents of SEC revenue reports is exactly that: belief.


I'm not saying I know the reports are false. I'm saying that no-one (except SEC and Google themselves) know what the truth is.


Part of my appreciation for Bitcoin (and I'd like to think everyone who appreciates Bitcoin understands the same thing) is that the standard of proof for what happens on the network is unimpeachable. I have high standards in general, and Google and the SEC no longer represent practice of high standards with their pieces of paper that basically amount to nothing more than a chain of rich privileged technology tsars and accountants doing pinky swears.

I don't trust a chain of bible swearing, I trust a chain of SHA-2 hashed datasets, performed by anyone with the resources to do it.

[NotFuzzyWarm]

Perhaps THIS will let this thread finally die http://spectrum.ieee.org/tech-talk/computing/hardware/google-details-tensor-chip-powers
In short summary - it would highly sucketh for mining.

[adamvp]

Interesting article.. but what do you think about new cryptocurrency algorithm using this TPU?

[NotFuzzyWarm]

Interesting article.. but what do you think about new cryptocurrency algorithm using this TPU?

sigh...  
What part of "it would highly sucketh for mining" is not clear?

It is USELESS for modern crypto... In the old days before modern computers yes it could have been used to spot patterns to aid in decrypting messages using the same key or pattern of key changes over and over but these days -- forget it.

Deep learning, Neural networks and data cache control logic (what Google originally used it for) are 'fuzzy logic' systems. They react to trends and do not process nor produce precise hard data, it only gives general pointers/guidance to a desired result. They cannot produce hard data because by design they constantly adapt how they react to inputs...

[Andre_Goldman]

These TPUs are a Machine Learning ASIC - not usable at all for mining on for ANY existing coin.

 I see ZERO impact of these on mining.

Indeed, TPUs are optimized to solve other statistical problems (p-np?) I would not bet too much on SAT Solving;

http://jheusser.github.io/2013/02/03/satcoin.html

According to a paper published on ISCA;

Many architects believe that major improvements in cost-energy-performance must now come from domain-specific
hardware. This paper evaluates a custom ASIC—called a ​Tensor Pro​cessing Unit (TPU)— deployed in datacenters
since 2015 that accelerates the inference phase of neural networks (NN). The heart of the TPU is a 65,536 8-bit MAC
matrix multiply unit that offers a peak throughput of 92 TeraOps/second (TOPS) and a large (28 MiB)
software-managed on-chip memory.