Perhaps. But if FPGAs do the job cheaply (2k per board as he calculates it), then there is not much incentive for ASIC manufacturers unless they can offer much better price for performance. If the FPGA costs 2k (and is sold 3k on retail) and the ASIC goes for 1gh at 10k retail price, it cant be sold. People would rather buy multiple FPGAs. It'll have no serious cost/hash advantage over the FPGA and FPGAs are reprogrammable/reusable as far as I know, which allow them to be resold when they are obsolete as x11 miners (something that can't happen with ASICs, limiting their life).


I have no idea how they perform really, but I'll say this which is somewhat "weird": Three days ago I saw a dream that FPGAs could go 52x over GPUs on X11 (that would indicate ~150Mh). I do not know if its prophetic or subconscious junk, but it was too specific to ignore - which usually indicates that it's not subconscious junk. Last time I saw something about cryptocurrencies, it was about Litecoin going from 16$ to 25$ (and it happened within the next 2 days).

+


He's talking about FPGA with chips ("6 Spartan 6 LX150") that are on the market, not ASICs. ASICs are a different issue.

I've been following the discussion on the x11 thread ( https://bitcointalk.org/index.php?topic=556277.180 ) and there were two interesting highlights - at least interesting to me:

1. A guy estimates that it'd take a few months and a 50k cost to roll out ~300MH FPGAs for x11, with 2k per board cost: https://bitcointalk.org/index.php?topic=556277.msg6077250#msg6077250

If this happens, diff will make block reward go to 5 pretty quickly (and reducing the number of coins perhaps?)

2. Scrypt-N is already ASICable with an ASIC miner that seems to be able to handle N-factors up to 35 years ahead:


Also from their faq:

Min = 5

As it appears the prime selling point of these are scrypt, but they just happen to also mine scrypt-n with a large number of N. So it won't be money lost really because they'd still perform their job as scrypt miners.

http://blissdevices.com/tech-specs/

"Scryptr also provides a high level of flexibility to support mining of newer coins that are “fairer” to miners such as Vertcoin by allowing configurable Scrypt parameters. Also, the chip’s PCIe interface enables faster and greater data transfer speeds, if and when new algorithms require more network data."

http://www.reddit.com/r/scryptmining/comments/20x6r0/bliss_devices_announces_new_prices_on_scrypt/cg8xcd6

"Our chip supports configurable N parameters which allows it to mine N-factor scrypt up to N=262144, which is still about 35 years away."

...so when can we expect the change of PoW?

Not if ASICs came online gradually instead of "spiking".

Darkcoin (which first used x11) didn't do any form of self-promotion or self-marketing. It's one of the most under-promoted coins out there. Darkcoin didn't even use x11 as it's main selling point but rather DarkSend. How can x11 be a marketing ploy when there was no marketing for x11?

Because the farm expanded in stages and the cards were bought in batches, thus hurting homogeneity (different models).


Yep the electricity in particular.


He goes "anonymous".


I haven't asked permission to disclose, and I think I wouldn't be given permission even if I asked.

That particular one has a mix of cards and is actually a nightmare in terms of homogeneity due to the various models, pcs, etc. Generally mines doge & lite. He's usually top spot or second spot in any pool he mines at any given moment.

All KGW coins that have made no effort to patch the problem are vulnerable.

There's margin for improvement. My 6850 overclocked @950 pulls 1065kh while in scrypt it does ~280. I find that experimental intensity works better than traditional intensity, usually with a value of 400 or 800.

I happen to know one gpu farm that pulls ~1 ghs in scrypt. So, that'd be ~3gh in x11.

The cost is not in the many-million dollar range, the gain could be potentially far bigger and an ASIC farm doesn't really need to outhash the network, just go for a 20-30% of it and benefit accordingly.

If scrypt-n coins are spewing, say, 1000 btcs per day in mining output and someone gains 20-30% of them while masquarading as GPU farms that switched from scrypt, it's millions of USD per month.

If we are contemplating chances of cracking a certain hash, then shouldn't we be worried more about single-hash coins rather than 6 or 11 hash coins - which would require the cracking of multiple hashes in a given timeframe?

I like the answer too but the question I have is what is this gap? Why shouldn't the output of one cycle go straight to the next cycle of hashing as input with zero waiting? Does it go to main RAM first, and hence the bottleneck between cpu ram / gpu ram? If so could it all be done on the GPU instead of cpu-gpu-cpu-gpu (just speculating).

It's possible to have 100% GPU load and 1000kh in one instance, and 100% GPU load and 2000kh in another, more optimized one. The difference is that the code will be executed in less GPU cycles not that the GPU will be less loaded with work.

For example the first incarnation of the darkcoin miner gave me ~600-700kh. The second (sph-sgminer 4.1) gave me 900+. It was always 100% load, it's just that the optimized version did it better with the available GPU cycles.

1) The algorithm is probably less ram intensive and it probably has plenty of room for optimization. That's understandable because the GPU miner was launched as a necessity in a short timetable, after DRK community gave a bounty to have a GPU miner developed quickly to counteract the possible existence of secret GPU miners.

Personally I tried doing #pragma unroll and #pragma unroll 1 (disable unrolling) tests to check the performance on my 5830 on the 15 sets of "for i =" of darkcoin.cl.

I could get around 5% more hashrate by #pragma unroll 1 on the fourth "for i" (947kh vs 939) and a #pragma unroll 1 with the 8th "for i" (979 vs 939kh). Interestingly if I #pragma unroll 1 on both the fourth and the 8th, speed goes lower than 939, so I kept the 8th.

Thing is, the unrollings don't always work the same on series 5xxx and, say, series 7 cards. I've also gained a couple percent on scrypt too: https://bitcointalk.org/index.php?topic=528062.msg5863333#msg5863333 by playing with unrollings but it doesn't translate to other cards well, so I wouldn't say scrypt was intentionally unoptimized.

2) So far as everyone is using the same program and only a few people (<1%) use something more optimized, everyone is better off since everyone is burning less electricity. Given the extremely low profitability right now of most altcoins, electric cost is a problem.

3) Having >1 hashes is better for security in the long term. If say a hash is broken / proven unreliable, you have another 10 to go on. If however your ONLY hash is broken = game over for a coin.

Spotted a nice, classy gold/black logo: https://i.imgur.com/nLGuUAV.jpg

Litecoin/Vertcoin (Scrypt and Scrypt-N) have made two choices that will boomerang in the long-term.

1) Declaring asic immunity (which is false - one only needs to see vertcoin's title that says "no more asics") and using it as a selling point. Litecoin discovered their claim was bogus when it was GPU mineable. They are true when they say that if a hash can be GPU-mined, it can be ASICed. Same is true for scrypt, scryptN, x11 etc.

2) Using much RAM as a cost-deterrent. This will only make ASICs much more expensive than normal ASICs and if it works as they say it will, then it will have the unintended consequence that scrypt and particularly scrypt-N ASICs will be out of reach for most people => worse decentralization in terms of hashing power.

Theoretically x11 hashes are easier to implement on ASIC (as single hashes - as a chained hash we'll see how it goes). But x11 asics will eventually be cheaper for the masses (better asic decentralization) as they do not carry the ram cost that will make scryptN asics much more expensive.

In any case, when a new mining equipment comes out that is orders of magnitude faster, it should be available to everyone for fairness reasons. Coins that will try to remain at the GPU-only phase will always have the threat of ASIC on the horizon on what would happen if an ASIC came along. It will be a dejavu situation with early February over here and cpu mining vs fears of gpu miners being available in the wild. And if a coin like Vert states it will change PoW when ASICs are out, then they've opened their cards and ASIC manufacturers will outplay them: They'll make the ASICs and use it to mine scryptN networks while these networks are oblivious to ASICs existence. I did an estimation that if a manufacturer makes a few ASICs that capture 30% of an enlarged scryptN market in the next year that produces something like 1000 BTC per day in altcoins, then it's 300 BTC per day in profit / 9000 BTC per month. Even if BTC is at 500$ it'll be 4.5mn USD per month - and this power will masquarade as GPU power. If BTC is at 1000, we are talking about 9mn per month.

Btw, I laugh when I see gpu miners talk about their "concerns" on decentralization. We all know that miners go to large pools irregardless of the hashpower concentration as they don't care that much. In order to avoid variance, they can go to pools with 20-30-50% of the hashpower and then complain that ASICs will be bad for centralization and bad for the network.