Pages:
Author

Topic: The Ethereum Paradox - page 30. (Read 99908 times)

sr. member
Activity: 420
Merit: 262
March 07, 2016, 07:59:33 PM
I suppose everyone understand this, but I will state it for the record.

With a $200,000+ monthly budget, Ethereum can afford fancy website design and paid promotion at sites such as Coindesk, etc..

We gave them $18 million to produce no solution to the fundamental challenges they never solved, and they spent our money on hype to fool other greater fools into buying at nosebleed, manipulated prices. We invested in screwing the newbies who come to crypto.

That is not a sustainable model for growing our ecosystem. We are poisoning the well from which we drink.

Sorta working for BTC, no?

Huh Huh Despite having become by now arguably controlled by the Chinese mining cartel, Bitcoin had some years of being mined from laptops decentralized and I can spend Bitcoin widely and use it to transfer funds internationally.

>Bitcoin had some years of being mined from laptops decentralized
Yeah, way back when you couldn't
>spend Bitcoin widely and use it to transfer funds internationally :-

Billed as noninflationary -> inflating @ 10 -12%/yr
Billed as virtually instant -> sometimes virtually instant
Billed as virtually free -> fees growing and are guesswork
Billed as money 2.0, the next world currency -> 3tps tops; not enough for a suburban shopping mall
Billed as p2p cash -> bank2bank sidechain settlement layer.

So yeah, with the only unique use cases being ransomware, child porn, online gambling & DNMs, I'd say fancy poker chips; speculative value only (regardless of teh backstory du jour).

And compared to what? Ethereum with 0 users versus Bitcoin's 1 million users.

Yes Bitcoin has stalled. We all know that. But the shit offered so far to replace it is useless/flawed lies and hype.

Bitcoin fulfilled a significant portion of its promise. Enough such that we now all have a crypto unit-of-account to trade from.

I of course agree that Satoshi hyped Bitcoin as a less debased gold, which is bullshit. But it worked! We now have the crypto unit-of-account to trade from.
newbie
Activity: 42
Merit: 0
March 07, 2016, 07:57:33 PM
I suppose everyone understand this, but I will state it for the record.

With a $200,000+ monthly budget, Ethereum can afford fancy website design and paid promotion at sites such as Coindesk, etc..

We gave them $18 million to produce no solution to the fundamental challenges they never solved, and they spent our money on hype to fool other greater fools into buying at nosebleed, manipulated prices. We invested in screwing the newbies who come to crypto.

That is not a sustainable model for growing our ecosystem. We are poisoning the well from which we drink.

Sorta working for BTC, no?

Huh Huh Despite having become by now arguably controlled by the Chinese mining cartel, Bitcoin had some years of being mined from laptops decentralized and I can spend Bitcoin widely and use it to transfer funds internationally.

>Bitcoin had some years of being mined from laptops decentralized
Yeah, way back when you couldn't
>spend Bitcoin widely and use it to transfer funds internationally.
Don't dwell on the past. It's over.

Billed as noninflationary -> inflating @ 10 -12%/yr
Billed as virtually instant -> sometimes virtually instant
Billed as virtually free -> fees growing and are guesswork
Billed as money 2.0, the next world currency -> 3tps tops; not enough for a suburban shopping mall
Billed as p2p cash -> bank2bank sidechain settlement layer.

So yeah, with the only unique use cases being ransomware, child porn, online gambling & DNMs, I'd say fancy poker chips; speculative value only (regardless of teh backstory du jour).
sr. member
Activity: 420
Merit: 262
March 07, 2016, 07:06:32 PM
I suppose everyone understand this, but I will state it for the record.

With a $200,000+ monthly budget, Ethereum can afford fancy website design and paid promotion at sites such as Coindesk, etc..

We gave them $18 million to produce no solution to the fundamental challenges they never solved, and they spent our money on hype to fool other greater fools into buying at nosebleed, manipulated prices. We invested in screwing the newbies who come to crypto.

That is not a sustainable model for growing our ecosystem. We are poisoning the well from which we drink.

Sorta working for BTC, no?

Huh Huh Despite having become by now arguably controlled by the Chinese mining cartel, Bitcoin had some years of being mined from laptops decentralized and I can spend Bitcoin widely and use it to transfer funds internationally.
sr. member
Activity: 420
Merit: 262
March 07, 2016, 06:58:31 PM
There are also asymmetric proof-of-work such as tromp's Cuckcoo and others which are based on NP hard problems.

Cuckoo Cycle is based on finding cycles in undirected graphs.
While the latter is obviously in P (and thus not NP hard, unless P=NP),
the full graph description is exponentially larger than the 128-bit siphash key that defines it.

But while the "is there a cycle for this key" version is in NP\P (again assuming NP != P),
it almost certainly is not NP hard (that is, you cannot reduce all other NP problems to it).

It was 3 hours past my sleeping time and so I just wrote NP as a placeholder, because honestly I really couldn't reload the complexity classes in my head last night.

My point is that the Cuckoo Cycle is verifiable in polynomial time (i.e. fast) but apparently a brute force search for a solution is slower, which is afaik is what makes it an asymmetric (fast to validate, slow to compute) proof-of-work. Thus is it EXPTIME and not P? (just asking and haven't really looked at the Cuckcoo Cycle algorithm lately and I have forgotten it since I last did)

As for the major unsolved problem of mathematics and computer science of whether P = NP, I think perhaps I outlined conceptually how to prove they are not equal. Intending to come back to that in the future when I have free time.
legendary
Activity: 2142
Merit: 1010
Newbie
March 07, 2016, 02:37:35 PM
My abacus also doesn't have a memory bottleneck.
That doesn't make it superior to memory-hard tasks:-(

Good analogy, looks like I was wrong all the way.
legendary
Activity: 990
Merit: 1108
March 07, 2016, 02:11:14 PM
ASICs can also be used once a good configuration is found on a FPGA. ANN don't have memory bottleneck, this is why they are superior to memory-hard tasks.

My abacus also doesn't have a memory bottleneck.
That doesn't make it superior to memory-hard tasks:-(
legendary
Activity: 2142
Merit: 1010
Newbie
March 07, 2016, 01:59:16 PM
Why would you use a neural network to solve a problem which doesn't require learning? Surely you should be more concerned with ASICs?

ASICs can also be used once a good configuration is found on a FPGA. ANN don't have memory bottleneck, this is why they are superior to memory-hard tasks.
legendary
Activity: 1008
Merit: 1007
March 07, 2016, 01:51:38 PM
Certainly, no reasonable sized ANN can compute a memory-hard function.

This requires to be proven. If an ideal ANN can solve Halting problem, then a real-world ANN still ought to do amazing things.

Why would you use a neural network to solve a problem which doesn't require learning? Surely you should be more concerned with ASICs?
newbie
Activity: 42
Merit: 0
March 07, 2016, 01:43:31 PM
I suppose everyone understand this, but I will state it for the record.

With a $200,000+ monthly budget, Ethereum can afford fancy website design and paid promotion at sites such as Coindesk, etc..

We gave them $18 million to produce no solution to the fundamental challenges they never solved, and they spent our money on hype to fool other greater fools into buying at nosebleed, manipulated prices. We invested in screwing the newbies who come to crypto.

That is not a sustainable model for growing our ecosystem. We are poisoning the well from which we drink.

Sorta working for BTC, no?
BRB, off to the ponzi sub of this forum, teach intrepid investors about unsustainable business models.
legendary
Activity: 2142
Merit: 1010
Newbie
March 07, 2016, 12:22:21 PM
Certainly, no reasonable sized ANN can compute a memory-hard function.

This requires to be proven. If an ideal ANN can solve Halting problem, then a real-world ANN still ought to do amazing things.
legendary
Activity: 990
Merit: 1108
March 07, 2016, 12:05:40 PM
There are also asymmetric proof-of-work such as tromp's Cuckcoo and others which are based on NP hard problems.

Cuckoo Cycle is based on finding cycles in undirected graphs.
While the latter is obviously in P (and thus not NP hard, unless P=NP),
the full graph description is exponentially larger than the 128-bit siphash key that defines it.

But while the "is there a cycle for this key" version is in NP\P (again assuming NP != P),
it almost certainly is not NP hard (that is, you cannot reduce all other NP problems to it).

I really fail to see how artificial neural networks (ANN) are applicable to proof-of-work.
Certainly, no reasonable sized ANN can compute a memory-hard function.
sr. member
Activity: 420
Merit: 262
March 07, 2016, 10:53:52 AM
I suppose everyone understand this, but I will state it for the record.

With a $200,000+ monthly budget, Ethereum can afford fancy website design and paid promotion at sites such as Coindesk, etc..

We gave them $18 million to produce no solution to the fundamental challenges they never solved, and they spent our money on hype to fool other greater fools into buying at nosebleed, manipulated prices. We invested in screwing the newbies who come to crypto.

That is not a sustainable model for growing our ecosystem. We are poisoning the well from which we drink.
sr. member
Activity: 420
Merit: 262
March 07, 2016, 10:42:00 AM
The algorithm for the "memory hard" hash function isn't going to allow you to implement it with a non-sequential architecture. The parallelism that is attained is to due to reducing memory latency by coalescing memory accesses, or by trading memory accesses for increased computation to maximize the consumption of available computational resources (e.g. on the GPU).

1. (Ideal) ANNs can do super-Turing computations.
2. ANNs don't have the memory bottleneck.
3. ANNs can be simulated on FPGAs.

Ergo: FPGAs can do hashing of such functions efficiently.

Or am I wrong?

I am too sleepy to learn enough about super-Turing computation and alternatives to von Neumann to attempt an answer right now.

The memory hard hash functions are designed to sufficiently randomized such that there is no alternative to reading the memory locations sequentially.

There are also asymmetric proof-of-work such as tromp's Cuckcoo and others which are based on NP hard problems. These may be susceptible to other algorithms on other architectures. This may be a very deep topic, certainly more than I can muster right now sleepy.

What you say tromp?
legendary
Activity: 2142
Merit: 1010
Newbie
March 07, 2016, 10:35:17 AM
The algorithm for the "memory hard" hash function isn't going to allow you to implement it with a non-sequential architecture. The parallelism that is attained is to due to reducing memory latency by coalescing memory accesses, or by trading memory accesses for increased computation to maximize the consumption of available computational resources (e.g. on the GPU).

1. (Ideal) ANNs can do super-Turing computations.
2. ANNs don't have the memory bottleneck.
3. ANNs can be simulated on FPGAs.

Ergo: FPGAs can do hashing of such functions efficiently.

Or am I wrong?
sr. member
Activity: 420
Merit: 262
March 07, 2016, 10:24:06 AM
The FPGA directly implements the memory bound PoW, just as an ASIC would,
except it doesn't need the ultimate speed/efficiency of an ASIC, since DRAM latency is
going to be the bottleneck.

I'm talking about architectures without von Neumann bottleneck. Artificial Neural Networks is an example of such architecture.

The algorithm for the "memory hard" hash function isn't going to allow you to implement it with a non-sequential architecture. The parallelism that is attained is to due to reducing memory latency by coalescing memory accesses, or by trading memory accesses for increased computation to maximize the consumption of available computational resources (e.g. on the GPU).
legendary
Activity: 2142
Merit: 1010
Newbie
March 07, 2016, 10:19:48 AM
The FPGA directly implements the memory bound PoW, just as an ASIC would,
except it doesn't need the ultimate speed/efficiency of an ASIC, since DRAM latency is
going to be the bottleneck.

I'm talking about architectures without von Neumann bottleneck. Artificial Neural Networks is an example of such architecture.
sr. member
Activity: 420
Merit: 262
March 07, 2016, 10:16:01 AM
Readers aren't going to understand what we are referring to with this technobabble about latency bound hash algorithms.

Suffice it to say that afaik, Monero has the most CPU friendly hash function deployed. I think it is possible to do better though. Ethash unless they changed it significantly since before Ethereum was conceived is more GPU friendly than CPU friendly. So I presume CPU friendly would be more decentralized, given all other factors being equal.

A hash function being CPU-only would still not be sufficient to prevent the mining from centralizing, due to the other economic factors of economies-of-scale (e.g. more hashrate has lower validation and propagation costs).
legendary
Activity: 990
Merit: 1108
March 07, 2016, 10:10:54 AM
On the contrary.

In a world where everybody has an FPGA, DRAM becomes the ideal mining ASIC,
with the FPGA making sure that DRAM access is saturated and thus the DRAM is the
actual bottleneck.

You seem to talk about replicating von Neumann architecture on those FPGAs, right?

The FPGA directly implements the memory bound PoW, just as an ASIC would,
except it doesn't need the ultimate speed/efficiency of an ASIC, since DRAM latency is
going to be the bottleneck.
legendary
Activity: 2142
Merit: 1010
Newbie
March 07, 2016, 10:07:25 AM
On the contrary.

In a world where everybody has an FPGA, DRAM becomes the ideal mining ASIC,
with the FPGA making sure that DRAM access is saturated and thus the DRAM is the
actual bottleneck.

You seem to talk about replicating von Neumann architecture on those FPGAs, right?
legendary
Activity: 990
Merit: 1108
March 07, 2016, 10:05:06 AM
Memory bounding doesn't look as a good feature in a world full of FPGAs.

On the contrary.

In a world where everybody has an FPGA, DRAM becomes the ideal mining ASIC,
with the FPGA making sure that DRAM access is saturated and thus the DRAM is the
actual bottleneck.
Pages:
Jump to: