[Poll] Do you want Litecoin to stay GPU minable?

[tacotime]

With the possibility of Litecoin FPGAs coming out in the next few months, and then ASICs being a year or two off, do you feel like the chain should have future optimizations to prevent FPGA or ASIC mining?  Or would you be fine with ASICs/FPGAs coming out and subsequently purchasing them?

[mr_random]

With the possibility of Litecoin FPGAs coming out in the next few months, and then ASICs being a year or two off, do you feel like the chain should have future optimizations to prevent FPGA or ASIC mining?  Or would you be fine with ASICs/FPGAs coming out and subsequently purchasing them?

Litecoin FPGA's are a pipedream. The one person who is working on something has himself admitted it's proving a lot more difficult than he expected. I expect Litecoin FPGA's will give the user some advantage but nothing anywhere near as dramatic as the bitcoin ASICs.

[Nolo]

Not to mention as long as it has taken to develop BTC ASICs I would have to believe LTC ASICs would be more than 2 years away, as they would be even more difficult to produce.   

And as soon as that happens anyway, there will be a move to some other algorithm, just as the movement of large numbers of miners from BTC to LTC is happening. 

[crazy_rabbit]

With the possibility of Litecoin FPGAs coming out in the next few months, and then ASICs being a year or two off, do you feel like the chain should have future optimizations to prevent FPGA or ASIC mining?  Or would you be fine with ASICs/FPGAs coming out and subsequently purchasing them?

Litecoin FPGA's are a pipedream. The one person who is working on something has himself admitted it's proving a lot more difficult than he expected. I expect Litecoin FPGA's will give the user some advantage but nothing anywhere near as dramatic as the bitcoin ASICs.

Even Bitcoin FPGA's only advantage was Hash rate per watt of electricity. And I highly doubt most LTC miners are watching their electricity bill. The FPGA's weren't any more powerful then GPU's, just more efficient to run.

I'd be more worried about a rather small group of BTC GPU miners who leave BTC and try to 51% LTC. The GPU pools have enormous power compared to the LTC network currently, and they have a financial interest to attempt it. The LTC market is valuable and fairly deep. IF they were to 51% the chain, you would not know it in advance (as they work on their own chain in private) and then when the chain remerges they suddenly give themselves an enormous number of coins, sell them for BTC on the exchange, and walk away. You need ASIC's for protection. When bitcoin started there wasn't organised GPU computational power. Now there is- and lots of it. You can't avoid this fact. The faster LTC gains hash power the better. The fastest way is through ASICS.

[ehoffman]

Think about it.  In the main loop, there's 1024 memory fill rounds and 1024 memory lookup rounds (2048 rounds total).  For each round, some processing can be done in parallel but each round needs at least 50-something cycles.  So that's 100K+ cycles to get the result.  The max frequency of an FPGA may be 200MHz, so 200MHz/100K cycles is something like 2K to 3K Hahs/s.

Pipelining: All of this needs a 128K memory block for each try, so that's 128KB times 100K+ cycles.  So we're talking gigabytes of memory.  Access to memory from one round is something like 512 or 1024 bits wide (I don't remember exactly), so pipelining is pretty much impossible.

The only thing good at this is a general CPU running at much higher frequency, using SIMD instructions like MMX.  Or a GPU massive parallel and memory bandwidth.

I did think about doing FPGA (note that I did a successful SHA256 'Bitcoin' implementation), but I did the math and we're going nowhere with LTC scrypt algorithm

[debianlinux]

Ultimately, for the network to survive, there has to be mass adoption and near insurmountable hashpower to secure it. ASIC represents the coming of age for Bitcoin. Why would anyone who is not hoping for a pump and dump not wish the same for Litecoin?

[baggyp]

Ultimately, for the network to survive, there has to be mass adoption and near insurmountable hashpower to secure it. ASIC represents the coming of age for Bitcoin. Why would anyone who is not hoping for a pump and dump not wish the same for Litecoin?

I have been hearing this for months and months from 10's if not 100's of posters. I can see why ASIC assures securing of the network against 51% attacks without significant investment, but I don't know if I agree that ASIC will bring about positive change just yet. Pushing the GPU miners out of the game and over to another coin/algorithm is not only alienating a population who have been the sole reason that bitcoin has become so widely distributed and decentralized in the first place! (in my opinion only)

Until ASIC's are available in mass at a price affordable to the middle-class market, I believe we are going to see a transition of power into the hands of a small few.

 

[jubalix]

Ultimately, for the network to survive, there has to be mass adoption and near insurmountable hashpower to secure it. ASIC represents the coming of age for Bitcoin. Why would anyone who is not hoping for a pump and dump not wish the same for Litecoin?

for this reason

mass produced computers will advance faster than asics to ascis.

Eg the amount of R&D into mass produced computers is far more than asics can match, so the next generation of intel/amd whatever will be far faster than the last.

the next generation of asics not so much as opposed to the last generation.

Thus an algorithim hash that uses a normal computers will grow larger with each generation more than asic generations.

plus the potential of 100 millions as a userbase, not just 10, 000's asics.

LTC being resistant to ASIC maybe the feature that allows it to overtake BTC in the end, as more people want to be in, as a miner/transactor as well as a LTC user.

Further having millions in allows the network to be vastly more secure rather than a few easily identifiable ASIC farms

[tacotime]

Think about it.  In the main loop, there's 1024 memory fill rounds and 1024 memory lookup rounds (2048 rounds total).  For each round, some processing can be done in parallel but each round needs at least 50-something cycles.  So that's 100K+ cycles to get the result.  The max frequency of an FPGA may be 200MHz, so 200MHz/100K cycles is something like 2K to 3K Hahs/s.

Pipelining: All of this needs a 128K memory block for each try, so that's 128KB times 100K+ cycles.  So we're talking gigabytes of memory.  Access to memory from one round is something like 512 or 1024 bits wide (I don't remember exactly), so pipelining is pretty much impossible.

The only thing good at this is a general CPU running at much higher frequency, using SIMD instructions like MMX.  Or a GPU massive parallel and memory bandwidth.

I did think about doing FPGA (note that I did a successful SHA256 'Bitcoin' implementation), but I did the math and we're going nowhere with LTC scrypt algorithm

One of the FPGA developers sent me some details of his optimizations, but I don't know of their validity.  I'll forward them to you in a PM tomorrow.

[tacotime]

PM sent.  It's interesting to see that so many miners want to keep the chain GPU only.  I think it might be a good idea to have a technical consortium with LaSeek and Coblee to try to figure out ways that we can maintain this property if the network really finds it desirable, and possibly implement this at block halving time (~800k blocks into the chain).