Bitcoin Forum

The discussion on the Cyclone V launch got me thinking. Besides the fact that the Spartan LX150 is cheap on a per unit basis, is it the best option on a MH/$ basis? For example, could we be better served by a gigantic $1500-$2500 FPGA? I understand the first FPGAs to be used had to be cheap on a per unit basis as there was no established market. However, the BFL Single thread in the marketplace describing the number of orders just shows how much of a market we have now. This all stems from my interest in seeing something in the 1-5 GH/s range that bridges the gap between Icarus/X6500/Single and the BFL mini rig.

If it is a case that there are no economies of scale in purchasing bigger chips then that's a pretty quick and easy answer.

If I understood what you said correctly, I would LOVE that idea.  I don't get why there are 400mhash units, 800mhash units, and then right to 25Ghash then 50Ghash.

I want a good 5 Ghash unit for a few grand!!!

Typically price goes up per LUT so getting a bigger chip will tend to be more expensive per MH (because of lower yields). Getting a faster chip will also be more expensive because allowing for a higher clock speed is difficult. There are bigger chips and chips that could be clocked faster but on a per MH basis those chips would probably be significantly more expensive. One advantage of bigger chips would be that you could have only one board instead of several board so you save money by only having one copy of all the supporting circuitry instead of several. It would also be simpler to maintain but in the event of failure you lose the entirety of your mining operation instead of a single board.

I think that the biggest breakthrough in MH/$ will be through bulk and newer chips. There are HUGE bulk discounts available for FPGAs, but they require significantly more volume than the bitcoin fpga market currently has (especially when split between multiple venders). Newer chips (specifically the artix7) are supposed to be faster, lower power, and cheaper per LUT than the current Spartan 6 which will hopefully make fpga bitcoin mining cheaper per MH.

Bigger chips=bigger packages=bigger problems with PCB, cooling, powering=bigger cost=lower MH/$ ratio.
Also you will get faster price discounts from bulk orders when ordering small chips.

the Spartan 6 LX150 has a lot of logic gates at a reasonable price, which makes it very suitable for mining

There are other chips that are a lot beefier in terms on I/O which makes them better for traditional FPGA applications but they do not have a lot of logic gates and thus not all that great for mining.

For example you could buy mid grade Virtex-6 for $3k and would only get the hashing speed of about 2 x Spartan 6s

^ Exactly.