Bitcoin Forum

...that's what I'm wondering for quite some time now (as the time FPGAs can be used for mining seems to be quite limited now with ASICs showing up).

What I usually read is something like "these boards cannot be used for anything else because they are stripped down to the minimum".
But ztex for example also mentions "Monte Carlo methods" and "Bioinformatic calculations".
I'm studying physics so I'm interested especially in MC, or quantum MC. I don't know much about FPGAs (yet!), but I would love to learn a HDL and write a QMC algorithm.

So... What dou you (FPGA-)guys think? Are the "bitcoin mining boards" really limited to bitcoin mining or can they be used for other stuff?

Nothing showed up atm :)


BTT:

Im new to FPGAS too and I'm also wondering what we can do if that "so called miracle overhyped" ASIC hits the ground.


For Ztex boards there's already a whole wiki (by Ztex) aviable.

 

Doorstop, bookend, paperweight, coffee table ornament... the options are limitless.

Sure, but please don't make this yet another BFL-topic ;) I think there is a very high probability that by the end of the year we will see the first ASIC miners, combine that with reward cut and using GPUs or FPGAs for mining will be a waste of resources or have a ROI of, what? 10 years?

If they can be used for other applications: Fine, I'll buy one or two, mine until the end of the year, get something like 30-50% ROI and then use them for other "fun" stuff, like QMC, if that's possible (and that's what I'm asking)

@ckolivas: Do you feel my question was stupid enough to answer like you did?

U could write your own password hasher (cracker for them).

Maybe someone invents BIONC for FPGAS?

It was a random troll and you should have recognised it as one.

On the other hand, there is a half-serious warning in it. I DO think they are realistically useless for anything else because the real market for such a board in the hand of a hobbyist is zero in my humble opinion. Just because there are other uses for these boards does not mean you're likely to find anything useful to do with them yourself.

Those skilled in FPGA can reprogram them to do a few other things, where you are asking it to do a very high frequency of relatively simple task.
While I've not done it myself, my father is in the telecommunication and general semi-conductor industry, thus worked with FPGA's.
He's used them on more than one occasion in his line of work, I don't understand all of what he does, but he's done a lot in the field of top tier communication hardware, servers, routers etc. Building a bitstream is my latest thing to attempt, and is still a totally new thing to me, but their is some potential to reprogram them, but realistically it would be hard for most to re-purpose them.

I've heard of FPGA's been reprogrammed to work along side relatively normal webservers, to aid in doing repeated tasks, searches and fetching. They've been doing it with GPU's for a little while now, so I don't see why not.

If however you don't have any means to really reprogram them yourself, then yeah it's just a bit of a paper-weight that will done nothing much else.

Even if he or you don't think so, I certainly feel that his answer is reasonable.

Simple example: the BFL device.
Yes they have said they will Open Source their device if BTC tanks.
But does that really mean you will be able to do anything with it?
How far will they go with the information they will provide - will it be enough? Will they do it at all? They are supposedly a company with 10 years of experience in FPGA/ASIC technology, so why would they be giving away this information?
If BTC tanks tomorrow, the BFL is a door stop until (unless) these issues are overcome and someone comes up with a use for it.
And just an FYI - I have 1 BFL (paid for with BTC and received 4.5 days after payment)

I'm not talking about BFL, should have made that clear. But what about all other FPGA hardware? ZTEX boards have a good documentation, for example.

I'm not talking about resale either. If FPGA mining is "over", *I* would like to do other stuff with that board.
My question was:
Am I limited by my knowledge about FPGA/HDL? Then there's a "simple" solution: I need to learn about these topics.
Am I limited by the hardware? I.e. how the FPGAs are connected, how much bandwidth is available to tx and rx data from/to the FPGAs etc... e.g. all bitcoin boards lacking external RAM. Is that a problem for only a small subset of problems or is that a problem for most algorithms except hashing?

I'm not worried about what you can do with FPGAs (the fact that Bitcoin miner boards don't have RAM would be a problem for some kinds of work though), the problem is that the tools to build a bitstream seem quite costly and out of reach for any hobbyist.

gyverlb: the tools are not out of reach to the hobbyist. As a matter of fact, the first FPGA bitstreams (plural) for Bitcoin mining have all been developed by hobbyists on these very forums.

I plan on converting my BFL singles into ASICs

From what I read and could collect by looking for information, you need to pay several hundred (thousand?) dollars to get the Xilinx tools and the bitcoin miner bitstreams seems to need 10s of gigabytes or RAM to build. I don't know about others but my hobbies usually start smaller than that...

FPGAs are classical computers. They can be simplified to a Turing machine and thus cannot run quantum calculations.

Xilinx ISE has a free version available. And if you can afford a $400 FPGA board, you can surely spend $80 for 16GB RAM...

You are completely wrong.
1) I was not even talking about quantum computing. I was talking about quantum monte carlo, which is nothing more than solving some kind of equations.
2) You actually CAN run quantum algorithms on a "classical computer". You usually just don't want to do it because this requires an immense ammount of resources. E.g. you need to track the state of the system in every time step, which means you need TB of RAM even for small systems... If you run a quantum algorithm on a quantum computer that's not needed, because quantum physics already does that for you. But that's not even why quantum computers are faster than classical computers. They are because they can compute all solutions of a problem AT ONCE (ok, that's a bit simplified), u just have to pick the right solution.

FPGA's can do many things. They are programmable devices. You can for example made an "inteligent home" controler. Only obstacle in purely mining boards is lack of I/Os. But if you choose wisely there are boards with reasonable amount of I/Os. Then only you imagination will be limation factor.

FPGA are good for "number" crunching. and can be highly optimized to perform certain tasks.

mining boards lack IO that means that you can't make I/O heavy operations with it(ram, harddisk, network, ...).
mining boards are only good for computations, you could use it as a signing device for ecdsa or rsa, or make it a password cracker.

MiniRig as a password cracker would be quite good :) so long at it matches the performance of 50 x HD7970s as it does with mining.

I'm actually repeating the OP but ...

Yeah I see this often around here and other places on the net ... "Password Cracker"
But seriously why? What use is there to this except hacking into other stuff you aren't supposed to be hacking into?
... and by use I mean something that you would do night and day and pay $600 for a device to do it.
i.e. like BTC mining where there is a financial return.
Password Cracking seems like a complete and total waste of time IMO

yes, but....? he asked what it was useful for. i just gave examples.

Mining Litecoins? :)
* interlagos puts on anti-flame helmet and runs away *

Please read mrb's blog. He has an example of using password cracker to liberate the cell phones that are carrier-locked or SIM-locked. So it has both freedom angle and financial angle.

http://blog.zorinaq.com/

Hmm well when my daughter's iPhone was carrier locked I simply unlocked it visiting a web site ... took a few seconds.
So I guess we can find a 24/7 job for one FPGA device doing this for everyone on the planet.
What about the other thousands of FPGAs ...

Pen testing is big business, and cracking passwords in real time without having to spend a week on an expensive GPU cluster is attractive.

The free version (ISE Webpack tool if I'm not mistaken) doesn't seem to support the Spartan used in miner boards (at least not according to Wikipedia, and I'm not downloading a ~6GB tar.gz just to verify this), I don't have any computer available that would support 16GB RAM (one of my rig could, but it's kind of used for other things...). So the barrier to entry isn't that low : just to begin hacking I'd have to purchase a new computer and a development board with a supported FPGA instead of trying my hand on a Cairnsmore1 or ztex.

And the subject is about FPGA mining boards, not the FPGAs the free ISE supports : what good is the free ISE if it doesn't work for mining boards ?

It's kind of unrelated to the subject because its a matter of taste but I'm very reluctant to use a proprietary development platform too. I don't know of any free complete toolchain and I don't feel like beginning working on something like this with no free alternative.

WebPack ISE do not support LX150 but you can download 30day tial that supports all parts. And according to Xillinx documents synthesis and PAR requires little above 3GB of RAM for LX150. Higher end Virtexes requires more.

i guess they can be sold solely for the FPGAs if some company (or individual) can reap 'em off the board and use them elsewhere ;)

4GB may not be enough, see :

https://bitcointalk.org/index.php?topic=44891.msg619585;topicseen#msg619585

I think I saw another post saying pretty much the same thing but can't find it again.

If your build takes 24hours, you'd certainly want to benefit from several processor cores and more memory to test several designs. I'm the kind of guy who prefers interpreted language because testing the code is faster... Waiting 24 hours is kind of frustrating, so I'd definitely want to test several designs in parallel if I can.

You can do it yourself. With a liitle skills and cheap equipment (200$).

I know kano is trolling here but it is worth further comment.

The grey market is has literally truckloads of cell phones and cellular modems that can be purchased for just slightly above scrap value. They are always not-so-recent generation and frequently have branding and logos that are obsolete.

This type of business requires factory-style work arrangements, not a customer accessible web site. The unlocking needs to be done before the end-user sale.

And I see 3700MB peak for 64 bit sytem. So 4GB should be good. Propably in amateur designs you will never see that amount occupied. Besides it's always a virtual memory ;) Fast SSD reqiured though. ISE do not support pararelism so more cores don't help. But virtual machines and ISE on each one is of course possible...

Sorry but what does system memory have to do with FPGA reuse ?

I mean for $100 you can have 16GB RAM, or for $240 you can have 32GB RAM in an almost budget motherboard with budget CPU.

if you want to use your fpga for other things, you must make your own code and compile it.
to compile fpga code, alot of memory is required. so, if you want to use your fpga for other things you must have alot of ram.

4GB wont be good enough. You will get into a swapping storm with almost no CPU use and 100% disk wait. I made that mistake by installing ISE on a 4GB laptop that cannot be upgraded beyond that.

8GB laptop works just fine with total memory usage in the 5-6GB range and at least one core is 100% occupied most of the time.

If system at startup consumes 1GB then of course at some point it will cause HDD to heart attack ;) I have old laptop with 4GB and 32 bit XP (It sees only 3.75GB) And I never saw a "swapping storm" as you called.

There are reuse scenarios that I did't address : you can install another bitstream prepared by someone with the tools and know-how or you can sell the board.

Here's the scenario. You paid $600 for a board, connected it to one of your system to mine. Bitcoin mining isn't profitable and you want to hack your own boards as a hobby (or for profit, who knows ?).

The question is how can you use your $600 board ?

Here are the question you have to ask yourself : do you want to invest in a ISE license to use ISE more than 30 days ?
Here are the ones you *may* have to ask yourself :
- do you need 16GB RAM or more to do what you want with the board and ISE... apparently serious hackers do, but will you, do you know anyone that did hack the boards with ISE running on a 4GB laptop and wasn't frustrated ?
- if positive, do you already have a system lying around that support such an amount (I don't if I don't count the mining rig in the basement which would probably be sold long before I ask myself this kind of question) ? Are you prepared to upgrade your existing desktop if you have one or build one if you use a thin laptop like me (thin laptops don't support 16+ GB RAM...) ?

In the end, using your board for playing around could be quite costly and given that they don't have any RAM, you won't even be able to tinker with any algorithm needing any meaningful amount of local memory to work.

All these unknowns, costs and limitations make me believe that it would probably be frustrating to try hacking these boards as a hobbyist (not that I wouldn't like to try myself, I'd love to see a free toolchain that doesn't need humongous amount of RAM).

What's the point of your post? Install the ISE and implement one of the open source LX150 designs.

For the people who actually like to read with understanding I have one more recommendation: have a dedicated ISE machine. The full implementation runs for LX150 take in the order of one to two days. It is worthwhile to not use your normal workstation for such lenghty jobs in case you'll discover you need to run some maintenance tasks that may require reboot or other systemwide changes.

Well, at least that's no problem... I have multiple systems with i5 or i7 CPUs and 4-8 GB RAM (can be upgraded to 32 GB and no, those are not my mining rigs). Also loads of SSDs.

The way I see it now, at least for computations/number crunching, the missing RAM could be critical? And there is probably no way working around missing RAM, is it?

My next planned build is going to have 64GB of RAM to start with, to be upgraded in the future, and will have a virtualization platform of some kind so that I can avoid having multiple machines. The plan is to run the high-cpu and high-memory stuff in a VM with 40GB of RAM and several processor cores, and use the remainder for my daily desktop needs.

I tried to work like that for a while. I've found out that it isn't a workable solution for me and many other developers:

1) need to reboot the host too frequently for various reasons like hung USB driver/device, video driver up-/down-grades, global resource reconfiguration, etc.

2) accurate timekeeping problems in the guests;

3) network virtualization problems/bugs;

4) software performance debugging/troubleshooting became next to impossible.

5) things like plain-old midrange-desktop CQ6600 with 6 SATA drives in matrix RAID outperforms almost-top-of-the line Dell+Vmware server with 12 SAS + 2 SSD drives by more than an order of magnitude. I (and my co-workers) just don't have time to troubleshoot that.

The current software quality of the virtualization solutions for x86 is likely even going backwards. It is completely unlike that of IBM where many low-level hardware tests are virtualizable.

But I honestly bid you good luck, no irony. My current experience is that virtualization in the office setting is good only for the QA departments.

Well, I've had good luck with Citrix XenServer, and Intel has a hypervisor that I've been wanting to try. I've only needed to reboot the hosts that I run for upgrades, which really have been optional and not needed ("But we have to run the latest stuff for security! blah blah blah....")

So I guess I'll find out how well I can pull it off. If all else fails, I'll have some hardware that someone would probably want to buy. Parts list: http://secure.newegg.com/WishList/PublicWishDetail.aspx?WishListNumber=19183865
Box wasn't intended to run ISE, but I might install it and compile a few designs just for fun.

FPGAs way less powerful than the Spartan6 (which all the mining boards except BFL use) have been used for password recovery on .zip files, wifi password cracking, Bluetooth cracking, etc.  Code here:

http://openciphers.sourceforge.net/oc/index.php

Service that charges for this:

https://www.cloudcracker.com/

...an enterprising miner could launch their own.

What evidence do you have they are using FPGAs ?

Most services that are doing this are using Nvidia GPUs simply because the FPGAs / AMD GPUs driver support is not mature enough and "misses" passwords.

CUDA is best for these applications.

Moxie, the guy behind cloudcracker.com, uses Amazon EC2 GPU instances, which are very expensive, which explains why his prices are so high. A competitor using FPGAs could significantly undercut him.

Ignoring FPGAs, AMD GPUs are probably the best for this type of application, for the same reason that they are faster per $, compared to Nvidia, for ALU-bound workloads like pw cracking or Bitcoin mining.

A very popular GPU pw cracking tool is oclhashcat, which has excellent support for AMD GPUs. bulanula doesn't know what he is talking about  8)

I don't know that cloudcrack uses FPGAs, but there were a couple guys at Defcon last year doing WPA cracking on Spartan3-based boards plugged into their laptops.  I'll try to dig up a link; I'm almost positive the bitstreams were based on the OpenCores sources I linked.

My point is that you *could* run a cracking service using FPGA hardware repurposed from mining, and people *are* willing to pay for such a service.

Lets create out own FPGA reuse suite, and once BFL are no longer producing FPGA, we sell them on for even more than they cost us :)

Don't tell me what to do with my FPGA after BFL no longer produce them !

(pretty fucking pathetic....mostly on your part, as seen here first: https://bitcointalk.org/index.php?topic=88903.msg979901#msg979901 )

Please share link.

I think this is beyond the skill level of most miners.

Compiling FPGA bitstream is serious business ... takes 48 hours and needs huge resources.

I don't have the skill but with GPU it would be EASY. Not so easy with FPGA.

I am not FPGA expert but neither are the rest of the miners I would say ...