Topic: If ASICs don't make it, would you go for an FPGA model? (Read 2552 times)

For those people who still thinks ASIC will be made eventually. Making an ASIC is not simply as churning out an FPGA board or reprogramming an FPGA. Its lot more than that.

even companies like intel and AMD that churns lots of chips will keep on making various versions/spins due to the bugs they fund despite of intensive testing they do on there codes using simulation s/w and emulation h/w. they have years of experience under their belt to know all the issues at hand they will face while making an ASIC but still fail to deliver a product ready for mass production is first die revision called A0. Most of the chips will go base revision to B0 C0 to D0 may be even more. some problems can be solved by metal spins.

Base revision means full die change (even you need to add a single transistor in a 1billion transitor asic you need to do a full spin. When you do a base change die revision  will go from Ax to Bx to Cx and so on)
Metal revision means minor fixes (cost will depend on how many metal layers you need to change to fix the issue. When you do a metal change die revision will go from X1 to X2 to X3 and so on)
Let me give you an idea how much it will cost it 40nm from what i knew
At TSMC fab the full mask(base+metal) will cost you around 1mil USD. This is how much you pay for first mask (mask is like negative of a photo. YOu can print as many perfect pictures onces you get your negative developed properly)
If you have to do a metal spin for three metal you need to pay 50K/metal layer so for three metals you need to pay 150K. SO if you can resolve a bug by rerouting wires inside your ASIC you can go for a metal spin. Some times even a simple issue cannot be fixed by a metal spin and you need to go for a base spin)
A base spin you need to pay the 1mil USD again to get a new mask.

Few other things to know about 40nm fabrication technology
a) current fab out time means time to make your first chip from the day your deliver you final layout file to Fab 8-10weeks and add same for respins too
b) it costs a lot and need to pay thousands of dollar as license fee for using simulation and emulation tools


you may argue below

Q) they are going with a much lower fabricating technology. (65nm, 90nm,120nm--i dont know exact numbers beyond because i have worked only on 65nm, 40nm and now working on 28nm)
A) the ASIC they are making here needs to have raw hashing power that means most of the transistors here are switching a lot(changing from 1's to 0's and 0's to 1's). As the fabricating technology becomes smaller the switching current (dynamic current) becomes smaller so its advantageous only at lower tech. the higher the tech the more the price you need to pay for heatsink and cooling your chip

Post me more question and i will answer as far as my knowledge goes

Thanks
Pradeep
 

Nice sarcasm. I also think it's pretty whack that you're trying to stray people away from mining. Right now, with the current difficulty, a Modminer Quad will produce ~4BTC per month (http://www.alloscomp.com/bitcoin/calculator). According to blockexplorer, the next estimated difficulty isn't changing too much to affect results (http://blockexplorer.com/q/estimate). Assuming this stays the same, because ASICs aren't shipped, the price of BTC—by some—will be an estimated $30 by March 2013 (https://bitcointalksearch.org/topic/30-btc-by-march-136146). Finally, people are selling Modminer Quads for around $325, so a breakeven would be around three/fourth months. ASICs have been delayed for several months and still no proof is out there these products exist (or perform what they're advertising). So it's completely feasible for me to purchase a MMQ.

You know the network difficulty is self-regulating, right? It doesn't need any "help". If there are less devices hashing, the network self-regulates and makes it easier to find blocks. If more miners pop up, the network makes it harder to find blocks. The only reason to mine is to MAKE MONEY.

If you want to really help, do one of those protein folding programs that help cure cancer. That will make a real difference in the world.

I'm with you. I used to do distributed.net back when they were cracking RC5, then folding at home. Now I mine bitcoins. The other stuff I did for free for science. This I do to help the concept, the fact I can make some small amount of money is a plus. I even run a Tor exit node to provide another service I think is needed, aside from helping the Bitcoin network.

I'm not mining for profit. I like the whole idea, and I'll be helping out the network. TBH, I haven't seen any proof of ASICs coming out, so who knows - it may be more than a few months. And it may be profitable since the price of BTC can increase.

I don't know if you really understand this, but do you know that you "might" make about 7 bitcoins over the course of those couple of months. Totally not worth it unless you only pay about $120 for it. Just Sayin'

I'm in the process of buying a Modminer Quad, because
A) It fit's my price range
B) Super cool looking
C) It will still be profitable for a few more months
D) I have free electricity

Yeah, Avalaon at least having a pic of the board and soon to ship, I'll just wait it out until they take preorders, if ever.

Don't invest anything in FPGAs long term. Even if the four current ASIC developers all end up being vapourware, the market is so large and the price of Bitcoin is so high, mining ASICs will be developed and used within a year.

If all ASICs turn out to be vaporware, scam or whatever, yes I expect FPGAs to be the status quo for a long time.

Right now, FPGA miners are more efficient than GPU miners and use much less power, but only a few perform significantly better at hashing than the less expensive GPUs.  It comes down to how much you pay for electricity vs. how profitable you can mine with an FPGA.

I'm confident that ASICs will ship...eventually.  And once they come on line, then the FPGA miners will be no more profitable than the GPU miners.  I'm patient enough to wait it out.

Sir,

To reach same hash the FPGA solution is much more expensive than ASIC.

Would you go for an FPGA if still not enough asics meet the demand of people that want them?


Are you going for FPGAs because theyll be cheaper?


What can you get FPGAs?