Topic: ZTEX USB-FPGA Modules 1.15x and 1.15y: 215 and 860 MH/s FPGA Boards - page 41. (Read 182450 times)

But p2pool might be an important step for bitcoin so that makes it a bit more interesting!

The stale rate (on the bitcoin chain) is a little bit less than 1%. For that reason I do not consider LP as super urgent.

That may be the reason for more rejects. 1% is not bad but with LP it will get even better !

Aren't invalid shares (e.g. shares that does not met the difficulty) detected by the p2pool software? (If not, even better. ;-))

But currently BTCMiner is not suitable for p2pool for another reason: it supports no long polling.

LP and another stale reduction algorithm (which also works if BTCMiner is only connected to a local bitcoin instance) will be implemented in the next release. Since BTCMiner can simultaneously run hundreds of FPGA boards this is not trivial (but not really difficult).

I think p2pool needs long polling, I asked SZ about a deadline and he said jan/feb.

Never tried it out. But i'm curious too.

Has anyone pointed their 1.15x(s) at P2Pool yet?  Should it work or is the current miner software incompatible with shares of higher than difficulty 1?

Multiple FPGA don't work together.  Multiple GPU don't work together.  Multiple ASIC wouldn't need to work together.

A SHA-256 hash is a trivial thing to calculate.  Takes less than a hundred millionth of a second in FPGA or GPU.  No reason to design chips to "work together" or co-ordinate work.  Simple give each chip something completely different to work on.

2 chips = work on 2x as many hashes at same time
4 chips = work on 4x as many hashes at the same time.
10,000 chips = work on 10,000x as many hashes at the same time.

Yes, from what I gather ASIC performance is not that much higher than FPGA per chip, it's power consumption that drops a little but mainly the price, so you could have lots of SHA256 chips cooperating. The tricky part is to build that cooperation hardware/software I guess.

Including one time costs tens of thousands of boards (with an efficient unrolled implementation) cost millions of $ and their total hash rate is similar to or higher than the current total hash rate.

Would Peter Thiel - or any one other stupendously rich libertarian?

As an example: I belong to the group of people whose interest in Bitcoin is not in making money, but changing society. I want thousands of people running "miners" at home, at a cost less than having a lightbulb switched on, to make sure the network is resilient and available for transactions. And they'll pay for the privilege of knowing that's what they do.

I still contend a ASIC SHA256 implementation would be the way to go.  Have a bunch of pairs of these, a few buffers, an FPGA controller and you're golden.

Hundreds of thousands not tens of thousands and commitments to produce tens of thousands of units.  And performance will be better than FPGA but not 10x better like what is possible in an true ASIC.

Would you invest 250K into Bitcoin right now? 

sASIC is usually referred to as an ASIC/FPGA hybrid.

http://www.design-reuse.com/articles/7179/hybrid-process-converts-fpgas-to-structured-asics.html

And a Bitcoin sASIC only takes some tens of thousands to kick off.

(I'm eerily reminded of a thread a few months back with someone who fiercely claimed that no one would be mad enough to make a Bitcoin FPGA board since the BoM would be too high)

I think my tin foil hat was just blown clean off!  

BS.

First of all there is no such thing as an ASIC hybrid.

Second while you can make ASICS uses 4 generation old processes (lots of low demand fab capacity at 130nm for example) to handle low bandwidth tasks like a voltage monitoring chip, or usb controller for a lower cost.  The issue is those processes are totally unsuited for high clock cycle, high utilization work like mining.

You aren't going to fab a current gen ASIC w/ less than a couple million (possibly ten million) in NRE costs and anything less is going to be utterly uncompetitive w/ FPGA and GPUs.  Worse no fab is going to give you access unless you commit (and show financial resources) for tens of thousands of units.  The reason is that retooling time is expensive and they don't want you business unless your product can keep the equipment occupied for a significant period of time.

So on top of the NRE costs you are looking at millions more in financial commitments.  The idea that someone is going to take that kind of risk with Bitcoin as small and unproven as it is today is just tin foil nonsense.

To compare it to the "cost" of some software work using OpenCL is even more silly.  Had Bitcoin died, or turned out to be illsuited for GPU acceleration the cost would be some hours spent coding and a couple graphics cards. 

Let me offer a different take on this than D&T's post.

WARNING:

"They" is the (sometimes) silent minority here with deep pockets. Some members will remember the transitory period last year (well, technically 2010) when CPU miners switched en-mass to GPU. Remember when Art Forz was mining with 25% of the network hashrate? He was into GPUs big time when most everyone else was still on CPUs.

That was a wakeup call to all the miners, who very quickly realized if they don't get onto GPUs, their profit would disappear. And it didn't take long for the network hashrate (and, subsequently, difficulty) to go up 20x.

We may be facing a similar situation in the near future when an ASIC is dropped on the scene. Imagine what a $100 1Ghash 10W board would do to the mining landscape. For the same $$$ as today's most efficient GPU (5870 or 5970), what if people could buy 5x or 10x the hashing power? And operate it with a fraction of the energy?

We already know what would happen. We've seen it before. Once the ASIC became available people would buy up huge ASIC mining farms for relatively little cost, driving the network hashrate (and difficulty) through the roof. 2x, 5x, 10x. GPU miners would very quickly find their profits dropping to zero and would face a decision: replace their GPUs with ASICs, or stop mining.

The only reason this scenario hasn't happened yet is the huge up-front cost associated with making an ASIC. But over the last couple of years, a few here in the silent minority (long-term and large-scale miners, early adopters, successful bitcoin speculators, large pool/exchange operators) have been profitable enough that this is something they could conceivably undertake. Developing an ASIC, or ASIC hybrid, may not take millions today; it may only be hundreds of thousands. ASICs are not something that can be developed quickly, but if it has already started 6 months ago then 2012 may hold a rude awakening for us.

"they" = nobody. 

There is nobody willing to put the millions necessary on something as risky as Bitcoin.

Maybe in 5-10 years if Bitcoin is 30x as popular, and on more firm legal ground you might see a venture capital firm with the kind of dollars necessary to back a play like that.

Who is "they"?

Opportunity cost.

In ~12 months the block reward will be halved.

Assuming constant relative difficulty and price (which is unlikely but guessing future difficulty/price is even more dubious)
1 GH purchased today would mine $2000 in gross revenue over the next year.