[Archive] BFL trolling museum

[The-Real-Link]

  Ok so was thinking this over and wanted to spell it out in case I'm mistaken.  It would appear that once these BFL SC Jalapenos hit that they'll become pretty much a necessity to stay in the game, right?

  1.  Network right now is at ~12TH/sec.  
  2.  Average Joe can buy what was originally a pretty dedicated BTC setup of ~4GH/sec for barely $150.
  3.  If nearly everyone buys one unit, hashing power goes up by 10x so blocks are solved 10x faster.
  4.  Target Difficulty will still adjust after 2016 blocks but the first batch of blocks once ASIC is active should be solved very fast (comparatively).
  5.  Us still sitting on GPUs will have our shares and fractions of a pool % dwindled down to barely anything since we would then be making up even less of the total network hashing power.  We'd still submit the same shares as before of course but a block may contain 20 million shares instead of 2.
  6.  ASIC becomes self-aware and...  Ok kidding, haha.  ASIC becomes the main hashing method.

  Not saying I couldn't recoup the money invested as the machines are still plenty good for many things outside of Bitcoin.  At the same time though, it'd be nice to make sure I'm understanding it right to where hey, all of the time and effort spent in mining "the traditional way" will become pretty worthless in the coming months, block-halving notwithstanding.

  I'd imagine the difficulty may not jump as much because of many people leaving once these changes happen but still, a 10x jump in network hashing power is pretty big, let alone something like 20x.

[EskimoBob]

 Ok so was thinking this over and wanted to spell it out in case I'm mistaken.  It would appear that once these BFL SC Jalapenos hit that they'll become pretty much a necessity to stay in the game, right?

  1.  Network right now is at ~12TH/sec.  
  2.  Average Joe can buy what was originally a pretty dedicated BTC setup of ~4GH/sec for barely $150.
  3.  If nearly everyone buys one unit, hashing power goes up by 10x so blocks are solved 10x faster.
  4.  Target Difficulty will still adjust after 2016 blocks but the first batch of blocks once ASIC is active should be solved very fast (comparatively).
  5.  Us still sitting on GPUs will have our shares and fractions of a pool % dwindled down to barely anything since we would then be making up even less of the total network hashing power.  We'd still submit the same shares as before of course but a block may contain 20 million shares instead of 2.
  6.  ASIC becomes self-aware and...  Ok kidding, haha.  ASIC becomes the main hashing method.

  Not saying I couldn't recoup the money invested as the machines are still plenty good for many things outside of Bitcoin.  At the same time though, it'd be nice to make sure I'm understanding it right to where hey, all of the time and effort spent in mining "the traditional way" will become pretty worthless in the coming months, block-halving notwithstanding.

  I'd imagine the difficulty may not jump as much because of many people leaving once these changes happen but still, a 10x jump in network hashing power is pretty big, let alone something like 20x.

...and then, most of you will start mining Litecoins with your now useless GPU's? Damn...

[The-Real-Link]

 Ok so was thinking this over and wanted to spell it out in case I'm mistaken.  It would appear that once these BFL SC Jalapenos hit that they'll become pretty much a necessity to stay in the game, right?

  1.  Network right now is at ~12TH/sec.  
  2.  Average Joe can buy what was originally a pretty dedicated BTC setup of ~4GH/sec for barely $150.
  3.  If nearly everyone buys one unit, hashing power goes up by 10x so blocks are solved 10x faster.
  4.  Target Difficulty will still adjust after 2016 blocks but the first batch of blocks once ASIC is active should be solved very fast (comparatively).
  5.  Us still sitting on GPUs will have our shares and fractions of a pool % dwindled down to barely anything since we would then be making up even less of the total network hashing power.  We'd still submit the same shares as before of course but a block may contain 20 million shares instead of 2.
  6.  ASIC becomes self-aware and...  Ok kidding, haha.  ASIC becomes the main hashing method.

  Not saying I couldn't recoup the money invested as the machines are still plenty good for many things outside of Bitcoin.  At the same time though, it'd be nice to make sure I'm understanding it right to where hey, all of the time and effort spent in mining "the traditional way" will become pretty worthless in the coming months, block-halving notwithstanding.

  I'd imagine the difficulty may not jump as much because of many people leaving once these changes happen but still, a 10x jump in network hashing power is pretty big, let alone something like 20x.

...and then, most of you will start mining Litecoins with your now useless GPU's? Damn...

  I suppose either that or dedicate the power toward a distributed scientific project.

[ice_chill]

Regarding the statement GPU/FPGA will soon be useless for Bitcoin mining, atleast we know that ASIC is the final say, once you invest into it, you can only expect future products to have x1.5/x2.5 speed increase and not x25 like we are seeing now

[Hawkix]

.. but you can EXPECT product costing half the price you paid. And you can expect huge volume of such products.

[boonies4u]

Yeah, at least you have the resale value of the GPUs. I'm pretty sure the ASICs will be useful for some time, so I don't think resale for them will be as big of a concern.

[Smoovious]

I really hope they code these asics to be p2pool-friendly this time...

-- Smoov

[kano]

At least 2 people have sent the list of requirements for ASIC that the BFL noobs kept ignoring on the FPGAs, that would allow them to work on P2Pool ...

[ice_chill]

At least 2 people have sent the list of requirements for ASIC that the BFL noobs kept ignoring on the FPGAs, that would allow them to work on P2Pool ...

P2Pool is not the final say in Bitcoin mining TBH

[btharper]

I really hope they code these asics to be p2pool-friendly this time...

-- Smoov

In theory the jalapenos solve a whole getwork in under 1.25 seconds, where the BitForce Single took almost 5 seconds to complete; which is what lead to getting really high stale rates (~40%+). The new one should have under/around 10-15% stale rate based the scan time.

Edit: should have read jalapeno instead of single

[Epoch]

I really hope they code these asics to be p2pool-friendly this time...

In theory the singles solve a whole getwork in under 1.25 seconds, where the BitForce Single took almost 5 seconds to complete; which is what lead to getting really high stale rates (~40%+). The new one should have under/around 10-15% stale rate based the scan time.

I'm not sure where you are getting 1.25 seconds:

Full nonce range: 2^32 ~= 4x1e9 hashes
BFL Single @ 800Mhps needs (4x1e9)/(800x1e6) = 5 seconds
BFL SC Single @ 40Ghps needs (4x1e9)/(40x1e9) = 0.1 seconds

The SC Single will work fine on P2Pool ... assuming P2Pool's 10-second blocks, an SC Single will on average lose 0.05 seconds of work (half of its 0.1s nonce rate) every 10 seconds, which represents 0.5% stales.

Sure, this can be improved. But even it is isn't, 0.5% is already very good.

[btharper]

I really hope they code these asics to be p2pool-friendly this time...

In theory the singles solve a whole getwork in under 1.25 seconds, where the BitForce Single took almost 5 seconds to complete; which is what lead to getting really high stale rates (~40%+). The new one should have under/around 10-15% stale rate based the scan time.

I'm not sure where you are getting 1.25 seconds:

Full nonce range: 2^32 ~= 4x1e9 hashes
BFL Single @ 800Mhps needs (4x1e9)/(800x1e6) = 5 seconds
BFL SC Single @ 40Ghps needs (4x1e9)/(40x1e9) = 0.1 seconds

The SC Single will work fine on P2Pool ... assuming P2Pool's 10-second blocks, an SC Single will on average lose 0.05 seconds of work (half of its 0.1s nonce rate) every 10 seconds, which represents 0.5% stales.

Sure, this can be improved. But even it is isn't, 0.5% is already very good.

Oops, should have said jalapeno, although there's a chance that it's using multiple getwork requests (like multi-threading) which would multiply the time by the number of concurrent instances. The SC singles are probably multiple chips and probably run multiple requests at a time if I had to guess.

[Epoch]

I really hope they code these asics to be p2pool-friendly this time...

In theory the singles solve a whole getwork in under 1.25 seconds, where the BitForce Single took almost 5 seconds to complete; which is what lead to getting really high stale rates (~40%+). The new one should have under/around 10-15% stale rate based the scan time.

I'm not sure where you are getting 1.25 seconds:

Full nonce range: 2^32 ~= 4x1e9 hashes
BFL Single @ 800Mhps needs (4x1e9)/(800x1e6) = 5 seconds
BFL SC Single @ 40Ghps needs (4x1e9)/(40x1e9) = 0.1 seconds

The SC Single will work fine on P2Pool ... assuming P2Pool's 10-second blocks, an SC Single will on average lose 0.05 seconds of work (half of its 0.1s nonce rate) every 10 seconds, which represents 0.5% stales.

Sure, this can be improved. But even it is isn't, 0.5% is already very good.

Oops, should have said jalapeno, although there's a chance that it's using multiple getwork requests (like multi-threading) which would multiply the time by the number of concurrent instances. The SC singles are probably multiple chips and probably run multiple requests at a time if I had to guess.

Ah, yes, I see you've edited your post ... Jalapeno, not Single. The 3.5Ghps Jalapeno should see about 5.7% stales on P2Pool *if* it behaves as the current BFL Singles do (that is, if it computes a full 2^32 nonce range before returning anything). A 5.7% stale rate definitely has room for improvement.  

[runeks]

At least 2 people have sent the list of requirements for ASIC that the BFL noobs kept ignoring on the FPGAs, that would allow them to work on P2Pool ...

What is the issue with BFL's products and P2Pool, exactly? Sucky that they only work with pools (or solo)...

[Epoch]

At least 2 people have sent the list of requirements for ASIC that the BFL noobs kept ignoring on the FPGAs, that would allow them to work on P2Pool ...

What is the issue with BFL's products and P2Pool, exactly? Sucky that they only work with pools (or solo)...

Nothing is stopping anyone from using a Single on P2Pool; it will work.

But because of P2Pool's short blocks (10 seconds?), the fact that a Single spends 5 seconds hashing a full nonce range before returning anything means that a significant percentage of shares generated by the Single will be stale by the time it gets around to reporting them.

So even though you *can* use a Single on P2Pool, it would be foolish to do so if only from an income point of view.

[The-Real-Link]

  Thanks for the clarification there.

  I was joking to myself in thinking that hey, if you get a few Jalapens there wouldn't be any issue in just solo mining.  Then again two factors at work there:

  1.  Everyone will be getting that kind of horsepower
  2.  Difficulty should adjust fast so we'd be back to submitting shares, just on a greater hashing scale.

  Oh well.  Investment capital that's needed is far less than our giant GPU setups.  Faster payback and also a benenfit of cheaper devices to have more people mine so the idea itself spreads too.  Fine by me!

[Smoovious]

Anything stopping someone from creating a bitcoin-asic network of p2pool nodes, with a longer poll time, and adjusting the share chain to compensate, to make it more asic-friendly?

-- Smoov

[runeks]

At least 2 people have sent the list of requirements for ASIC that the BFL noobs kept ignoring on the FPGAs, that would allow them to work on P2Pool ...

What is the issue with BFL's products and P2Pool, exactly? Sucky that they only work with pools (or solo)...

Nothing is stopping anyone from using a Single on P2Pool; it will work.

But because of P2Pool's short blocks (10 seconds?), the fact that a Single spends 5 seconds hashing a full nonce range before returning anything means that a significant percentage of shares generated by the Single will be stale by the time it gets around to reporting them.

So even though you *can* use a Single on P2Pool, it would be foolish to do so if only from an income point of view.

So it's basically because you can't stop a Single, when you've first asked it to start looking for a nonce, and tell it to work on something new?

[kano]

Probably quicker to type again rather than find all the posts I've done of this ...

A single takes 5.2 seconds to hash
If you abort the hash it doesn't return any data.

A P2Pool LP is 10 seconds (on average)

Thus if you abort the work all the time, you will be throwing away a high % of your hashing time.
If you don't abort the work, a high % of your results will be stale.

[runeks]

A single takes 5.2 seconds to hash
If you abort the hash it doesn't return any data.

What data would you want returned? Someone else has found the nonce, unless you found it before them you might as well start with the new block, right?