[LTC] FPGA miner development - prototype december 2013

[funnow]

If I'm understanding you correctly, then we're looking at approx $1,300 for ~800Kh/s? Please correct me if I have that wrong. But at current difficulty and current LTC/$ (and we can only assume that LTC and scrypt in general difficulty is going to rise over time), it will take almost 2 years for this to achieve ROI. Perhaps LTC will increase dramatically, but perhaps not. Please tell me where I'm going wrong here?

Yes, the price is right, where did you get 800kh/s I don't know.

[nightengale]

If I'm understanding you correctly, then we're looking at approx $1,300 for ~800Kh/s? Please correct me if I have that wrong. But at current difficulty and current LTC/$ (and we can only assume that LTC and scrypt in general difficulty is going to rise over time), it will take almost 2 years for this to achieve ROI. Perhaps LTC will increase dramatically, but perhaps not. Please tell me where I'm going wrong here?

Yes, the price is right, where did you get 800kh/s I don't know.

See below... So you're saying 1 Mh/s?

Well if this is true we will have:
- 50Moperation/1000=50.000 hash/sec, now we should multiply it by 10 (10 cores inside the fpga) and we should have 50.000hash*10=100.000hash/sec, assuming that our assumption are still too optimistic we should have half, than 50.000hash/sec

When we put 16 fpgas the result should be 50KHash/sec will begin 50*16=800Khash/sec.
This is a really very worst estimation, the result is about 1Mhash/sec

[smoothie]

Just keep in mind that the price of LTC can change and then things become even less clearer for the future. I suspect a price rise in the next year but that's just my personal guess.

[nightengale]

I'm bullish on LTC price as well, the currency has a ton of room to grow I think.

But for me, when considering a mining investment, a possible future price increase doesn't factor into the equation very strongly. But that brings us back to the ROI argument that is all over the forum...   

[markm]

"chassis and controller 300usd"

Who uses a chassis? Or maybe first, what do you even mean by a chassis? A milk crate? a shoe box? If it is expensive many would likely prefer to get just the guts of the device I suspect.

What do you mean by a controller? A Raspberry pi like people use to control many many USB devices and/or FPGA boards? With maybe some extra RAM because cgminer mining scrypt needs more RAM than when it mines SHA256?

How many boards can a controller control? 127 per USB port?

-MarkM-

[funnow]

If I'm understanding you correctly, then we're looking at approx $1,300 for ~800Kh/s? Please correct me if I have that wrong. But at current difficulty and current LTC/$ (and we can only assume that LTC and scrypt in general difficulty is going to rise over time), it will take almost 2 years for this to achieve ROI. Perhaps LTC will increase dramatically, but perhaps not. Please tell me where I'm going wrong here?

Yes, the price is right, where did you get 800kh/s I don't know.

See below... So you're saying 1 Mh/s?

Well if this is true we will have:
- 50Moperation/1000=50.000 hash/sec, now we should multiply it by 10 (10 cores inside the fpga) and we should have 50.000hash*10=100.000hash/sec, assuming that our assumption are still too optimistic we should have half, than 50.000hash/sec

When we put 16 fpgas the result should be 50KHash/sec will begin 50*16=800Khash/sec.
This is a really very worst estimation, the result is about 1Mhash/sec

This is a really very worst estimation, the result is about 1Mhash/sec > you have to read all the part not only from: the result is about 1Mh/s

[funnow]

"chassis and controller 300usd"

Who uses a chassis? Or maybe first, what do you even mean by a chassis? A milk crate? a shoe box? If it is expensive many would likely prefer to get just the guts of the device I suspect.

What do you mean by a controller? A Raspberry pi like people use to control many many USB devices and/or FPGA boards? With maybe some extra RAM because cgminer mining scrypt needs more RAM than when it mines SHA256?

How many boards can a controller control? 127 per USB port?

-MarkM-

HI,
chassis is the container of fpga board or boards, power suply, processor....
assuming half of pc tower
the controller is a processor, Rapsberry could be one, but there are other available
the litecoin requires RAM,but do not think that is a lot (you can consider 150KB for each miner)
I think we  could have 4 fpga module, for an amount of 64 fpga
may be more, but at that time is too early

[markm]

Well then at $300 for something that miners could provide for themselves in the form of one or more milk-crates, a raspberry pi, an ATX power supply and maybe some extra RAM for the raspi, I suggest saving them not only that $300 but also gosh knows how much in shipping costs by leaving those things out...

-MarkM-

[Zalfrin]

It is resonable that we will have a result of 1000 operation for each hash

There's where you went wrong. You're off by a couple orders of magnitude.

A good whitepaper on the topic:
http://cdn.shopify.com/s/files/1/0241/1743/files/Alpha_Technology_-_Scrypt_Analysis_on_FPGA_proof_of_concept.pdf?709

[funnow]

It is resonable that we will have a result of 1000 operation for each hash

There's where you went wrong. You're off by a couple orders of magnitude.

A good whitepaper on the topic:
http://cdn.shopify.com/s/files/1/0241/1743/files/Alpha_Technology_-_Scrypt_Analysis_on_FPGA_proof_of_concept.pdf?709

Here is Znyq FPGA which has nothing in common with our FPGA, so you are not right.

[Zalfrin]

It is resonable that we will have a result of 1000 operation for each hash

There's where you went wrong. You're off by a couple orders of magnitude.

A good whitepaper on the topic:
http://cdn.shopify.com/s/files/1/0241/1743/files/Alpha_Technology_-_Scrypt_Analysis_on_FPGA_proof_of_concept.pdf?709

Here is Znyq FPGA which has nothing in common with our FPGA, so you are not right.

It's not a question of the FPGA used. It's the definition of the Scrypt algorithm.

[thcst8]

I'll take 4

[funnow]

I believe that a program running as sequential series of operations can not be comparared with fpga design, working in parallel the list of instructions included inside the doc can not be executed in series
I repeat, the fpga will work in different manner, it is not a processor, we can program to work in parallel the described asrchitecture is based on a xilinx fpga that is a soc system, and clearly the presence of processor inside force the architecture to work in this manner the soc are used to reduce the amount of hw developing a very comoact architecture
I am sorry compact architecture in that case instead of to use an fpga you can use a processor like arm or other working to 1GHz clock speed, in that the case the result could be 10 time faster