Topic: Official Open Source FPGA Bitcoin Miner (Last Update: April 14th, 2013) - page 34. (Read 432950 times)

6GH/s for $6-$8k? Even if it did 10GH/s no one would buy it. If that's the end game for FPGA then it doesn't look very good.

It's not just the fact you can use GPUs to reach 6GH/s cheaper, it's the fact the ATI cards have other uses. If you bought 6k worth of 6990s (plus parts), they'd get you 5-6GH/s. Lets say Bitcoin crashes and it's over. You could still go sell those 6k worth of 6990s on eBay and probably get 3k at worst with upside of getting 4-5K back. So your mining setup you're only really risking 50% of your momey at most, maybe closer to 30%.

If Bitcoin crashes and it's over, your FPGA rig is worth $0. Factoring in any power savings over how many months? Sure if you figure Bitcoin mining goes strong for 5 years it might make sense. The ATI gpus are more powerful, have other uses, there's a huge consumer market for used cards which means there's not a whole lot of risk. Computer hardware holds it's value. A one trick pony fpga card you'd be hard pressed to give away if you wanted (or forced) to stop mining.

Good Question, and you should keep in mind that a normal bitcoin hash needs 2 normal SHA256 Hashes, so the original Design of fpgaminer would mean 218 MHash for normal SHA256.

Look at the progress from the first public avaiable design @ 80 klut down to the 45??? klut requirement that was possible, i guess there could another step up  performance wise.

This data has been claimed as i asked for the data of the copacobana setup.

" .... this FPGA (Spartan 6 LX150) reach ~200 Mhash/s for SHA-256 ( per FPGA)"

I asume the key is a programm highly optimised for this particular FPGA

Is this for sha-256? I seriously doubt it.

The producers of the copacobana system claim their Spartan 6 Lx 150 to reach ~200 Mhash/s.
Therefore i believe this should be possible using a comperable cheap chip like the spartan 6 Series.

No offence but i still wait for somone other than artforz to verify this results.

While waiting for the miraclous ASIC chips i'd like to invite everybody interested to take part in my "Modular FPGA Miner Hardware Design Development" thread
http://forum.bitcoin.org/index.php?topic=22426.0 so we may provide an open source hardware plattform for the miners developed in this thread.

regards

Jens

I think that chip has already mean mentioned up thread and the projected hash rate for any of the 3E series was very low - 3-6 MH/s. I think it was reported as being capable of a depth of 3 or 4 and at 50MHz max could do < 5 MH/s. Unfortunately the Xilinx parts seem to route very poorly for this design.

Whats the rough LE count to gate count conversion? I just noticed this today and am hoping it would be enough to get one fully unrolled pipeline on it:

https://www.digilentinc.com/Products/Detail.cfm?NavPath=2,400,793&Prod=S3E1600

I'm pretty sceptical of ASIC production as well. There is some talk of it but with the amount of money needed to put in I'd have a hard time seeing a true business man willing to take the risk on it. It's still a pretty pie-in-the-sky idea business wise. There are just too many risks and generally business men are risk adverse - at least the successful ones are.

If they were produced it's quite possible they'd want to get them out and gone asap to recover the investment rather than get into the mining business and take even more market risk.

Buying GPUs is the least risky as there is still an active market for them - even though it could be depressed for some time if a crash happened.

If ever ASICs will be produced for mining, i see 2 szenarios:

* you can't buy them, because it gives some computing power that can't reached in other ways and it would be insane to sell the 'property' to competitiors.
* The chips will be sold to make profit from the chips, and then the run starts again, who can afford the most boards ful of minig ASICs.

But as bitcoin is still evolving, what if the price will drop due to lost reputation because of severe reasons, or there is an change  for the hashing method due to xome found vulnerabilities or there are some features added that would require a new computing scheme? A GPU can be reprogrammed, a FPGA can be reconfigured, a mining ASIC is a piece for the collectors shelf. Besides that, an GPU can still be used as a graphics card, can speed up computer animation rending and also for proper designed FPGA-Boards there are still scenarios to use them, and so you could sell them in case they have no use for you anymore.

ArtForz claims to get 190MH/s out of an XC6SLX150. While I haven't seen this with my own eyes I think he's trustworthy enough to rely on it.
He also claimed to be currently prototyping a 2U rackmount rig with 32 of them, reaching 6GH/s and using like 300 watts of power. He said that he's planning to sell machines like this, estimating a time to market of about 2 months and a cost of $6K-$8K.


Yeah, I think 240MH/s should be doable with these, possibly even a bit more.
What about trying to officially get permission to let them mine during the nights?

The FPGAs are not currently competitive and the only argument so far I've seen put forward is that they use less power. So when the difficulty level has doubled twice more they may play a part because electricity will be a much more costly factor making them attractive.

I think that at that time many miners will withdraw because the profits will be too measly for small operators to bother. This will have the effect of reducing difficulty again. I don't know where the stability point is but it's clear if ASICs ever get built and sold then they will most likely change that point such that neither FPGAs or GPUs are competitive. People will only play with them out of curiosity.

I haven't see a report yet of a chip with better performance/cost then the CE115 chip and I consider that one too costly at $315 for 109 MH/s.

GPU est. 200 MH/s @ $115 & 100W. (ATI 5770)
FPGA est. 109 MH/s @ $315 & 5W. (CE115)

Very cool project.

I've read over the threads and I can see how it will get better and better. But from what I'm seeing I don't see how it can ever be cost effective. Sure the power to performance ratio is great, but the problem is power is cheap.

Can the fpga boards really come down in price to make sense? $400? $300? $200? Don't you have to rely on the people making the chips and those prices don't look cheap. Unless you're making your own chips, which we won't, it's always going to be pretty costly.

So the question is given the current board, the cost, which is unlikely to drop much can the software really be improved that much to see a 10 fold increase? Is there really a $400 board that can do 1Gh/s down the road? If so what are the road blocks and what has to happen to get there.

It just looks like there's a huge gpu chip maker pumping out boards that are quite good at hashing. With ever improving boards, with better performance. Is there really going to be a fpga board that blows the ati chips out of the water at a price that makes sense? Because from someone who has no experience in chip design/boards (I'm a programmer) it looks like a bicycle trying to catch a car and I'm having a hard time figuring out where the giant leap in cost/performance is going to come from. If you look at newMeat1s design, is it realistic to say "Great start. Now make it go 10x faster at the same price". It just seems like there are limits to the chips no matter how well the software end is, the chips are expensive, they're going to stay expensive, and everything else around the chip can be improved/optimized but it's never going to makeup for the core cost of the chip in terms of competing with ATIs gpus.

I'm sure there is an answer I just don't know it so I figured I'd ask. =)

Just letting everyone know about the open source hardware design that newMeat1 just released:

http://forum.bitcoin.org/index.php?topic=23049

Very cool!  Looks like he has put further development on hold, due to the heavy cost of building prototypes. It's a CE115 based design, and those chips aren't cheap! I wonder if the design is compatible with a smaller chip ...

Currently they aren't performing at all, so I cannot give you any accurate figures about their performance. You can read through this thread (I know, it's a bit lengthy) for morsels of information regarding the current state of Spartan-6 development. Click and Ctrl+F for "Spartan".

You may also want to check out this thread, where they are discussing the development of a modular FPGA mining board.

Please forgive me if this has already been mentioned. I'm curious about how the Spartan 6 FPGAs (particularly the XC6LX150) are performing vs. the Cyclone parts. I gather that it's harder to route the unrolled miner design in the Xilinx parts than in the Alteras, but what sort of final clock rates are achievable in the Spartans with adequate cooling?

The Spartan 6 FPGAs look compelling to me because of their relatively low cost, but it's not clear to me whether they compete against the Altera parts in terms of $/Mhash. If they are cost-competitive and achieve similar hash rates, then I'm tempted to try making low-cost boards optimized for low I/O pin count applications like this. I can envision a single-FPGA board with a USB-interfaced microcontroller to babysit it, as well as a larger form factor with an array of LX150 FPGAs (say, 16 of them) under the control of a supervisor (possibly an LX45T with an ethernet interface, running Linux on a Microblaze core, for a fully stand-alone mining appliance). I'd try to bring out as much of the I/O as I could get away with on a 4-6 layer board to make them more generically useful.

We use a lot of FPGA platforms at work for hardware emulation (mostly Virtex-5 XC5VLX320 parts). I have one sitting on my desk that's not doing anything while I'm not at work, but I don't think it'd be wise for me to run miners at work over the closely-watched corporate network. Based on the FPGA mining hardware comparison, I'm guessing that one of them might fit two unrolled miners for a total speed of around 240 Mhash/s?

Ah I remember the original post, I missed where you actually gave it a shot though. Thanks for keeping me straight

http://forum.bitcoin.org/index.php?topic=9047.msg213431#msg213431
and I confirm it a post or two below that.

Did you already read the first post of this thread? There it is but if largish Mhps numbers give you a thrill you could download quartus, load the design and testcompile it with different combinations of  processing options, regarding logic count, cacle time, power consumption and look what for results you get -> numver of logic cells, maximum cycle time, junction temperature with cooling options from 'no cooling' to 'large heat sink with Air flow' ...  i think the number of test setups 'mining' is bigger than the number of people writing in this thread. Really important to most people are new designs that need  lesser logic,  alleo for  higher clock rates and simetimes some tweeks in there is only a dozen of logis elements necessare to  decrement the LOOP-setting by one or th squeeze in a second hasher ...

Did I miss it in the thread where someone hit 109 Mhps?

It is recommended to request a new getwork at least every 60 seconds, so that you're working on one with a semi-current timestamp in the block. By buffering work I'm already exceeding that on slow FPGAs, this limit is just to prevent it from being much worse (at 3MH/s it would be like a 66 minute delay from the getwork request to share submission in the worst case, so you would usually only work on a single getwork before long polling kills it off).

Well i cant remember to have done something to the AMD section yet 

As this FPGA subjet in total is still experimental still one of the best  affordable board in comparsion to its Mhash ratio the DE-115 proposed by fpgaminer.

( or as now may be seen in the Mhash/$/€    )