Topic: Algorithmically placed FPGA miner: 255MH/s/chip, supports all known boards - page 43. (Read 119429 times)

So when are you planning to start offering the bitstream for Icarus boards? how much it will cost? and what are the procedures? We really have to send the boards to you? This actually would be imposible for me as am not in USA!

Indeed, a hypothetical 300MH/s bitstream would make your equipment 50% more valuable for bitcoin mining purposes (let's not kid ourselves here: the non-mining market for these stripped-down boards is nearly zero).  So a rational miner would be willing to pay 50% of the cost of their boards in order to get these results.  Rest assured that I have no intention of charging nearly this much, although it is an important data point in assessing the value of what I'm providing.


I think you are greatly underestimating the number of FPGAs out there that are mining.  The number of people involved in FPGA mining is small, but there are already several very large FPGA farms.  Also, I have pretty good info on Xilinx' pricing curve, which lets me put a lower bound on how many units the major vendors have sold.  Lastly, I've been watching AVNet's inventory data, and they are moving pretty huge quantities of LX150's compared to LX75's -- and I don't know of any other non-scientific-computing product that uses them.


Neither can I, which is why kickstartr is currently the most likely option.

But... all this is academic.  Until there is a larger performance boost and I have done a demo in front of independent witnesses there is neither enough data nor a real need for this discussion.


I'm not enthusiastic about the bitstream encryption route, but I do want to point out that this is just flat-out false.  ADDING a decryption key to a Spartan in no way prevents it from being used for other purposes -- you don't have to use it (in fact, most bitstreams don't!).  The encryption key is stored by writing to eFuses; Xilinx has multi-million dollar customers relying on those fuses.  They are no more likely to fail than the rest of the device (in which case you're screwed anyways).


I doubt it.  You really do have to start over from scratch to do an algorithmically mapped and placed design.  Sure, somebody might do that, but it most certainly won't arise as a derivative of an xst-and-xilinx-map design.  And even if they're twice as fast as me it will take them four months -- so if you haven't heard about somebody starting an open-source project like this today you can be sure there won't be any ready to use until at least 4Q2012.

I'm only on 5 Ztex 1.15x Boards ATM but i'd love to see a faster Bitsream for them. But since i'm only a small scale Miner i'm not shure how much i could invest into a new Bitstream...... Sending all my boards to the US or locking them to a single Bitstream is no option.

Maybe if there were a Community funded Project so that he can get some money for his efforts i could invest at least something into it

Question:

How safe is it to use that Bitstream? i really do not want my boards go up in flames while mining .....

I see Dr Tyrell's concerns about getting paid could result in this bitstream likely to be a big tease (and never released or sold).

If Dr Tyrell had a large capital sum to invest in his own bitcoin mining operation then I don't see why this thread exists at all, since he'd keep it to himself and his large array of LX150s. After all, with a lot of FPGA units and a 150% advantage over the open-source bitstream (which, on the Ztex units at least, is still competitive for those with high electricity costs), all you need is an assembly design and appropriate cooling - and you have a steady, respectable income stream with negligible running costs.

I would understand if there was a credible way of selling the bitstream to FPGA miners without any risk of piracy, too. I am an aspiring small-size miner (9 Ghash GPU capacity, being replaced in a few weeks by 5 Ghash of FPGAs, with the plan to decommission / sell all GPU rigs, and continue scaling up the FPGAs) - and FWIW I'm a happy Ztex customer. Hence I will soon have over 25 LX150 boards and it'd be trivial to use my business-case spreadsheet to work out how much I'd be prepared to pay for a bitstream that converts my 208 MH/s units into 300-odd MH/s units.

However, unless there are miners out there with tens of thousands of LX150-based boards already running, I can't see any single miner paying enough to 'compensate' for the number of hours development involved, assuming Dr Tyrell is looking at private-sector consultancy rates and how much he'd have been paid if he had developed this bitstream for a private client.

As it stands, I'd obviously benefit from a 300 MH/s bitstream for the LX150 so I'd be interested in 'buying' it - but since there is already a 208 MH/s bitstream for *free*, the value of the incremental speed hike (and concomitant increase in mining income) is necessarily limited. Even if I was prepared to pay a full year's worth of *all* mining income attributed to the increased FPGA hashing speed, it wouldn't be in 'professional consultancy' ballpark figures.


The question is whether Dr Tyrell wants to make some money from this, or not at all. Right now it's a bit of a tease, but because it shows that three instances per LX150 *is* possible, it's probable that the open-source effort will, eventually, squeeze enough hints from the thread to build their own version. Of course, this just slows down the open-source process by a few months, during which time Dr Tyrell will have had time to further optimise his design... but once there is an open-source bitstream running 3 instances (at around 300 MH/s), the window for Dr Tyrell to make money from his bitstream closes.

There's no way that cautious FPGA miners will send their hundreds-of-boards investment to someone for encrypted bitstreams to be loaded out of their sight. If a unit fails, and needs re-programming, then it'd need to be sent back to Dr Tyrell and we're not all in the USA. If the only way to prevent 'theft' of the bitstream would be to lock the FPGA so it can't be used for other purposes (or reprogrammed) then, again, no miner would agree to this (since standard FPGA boards still have value if Bitcoin crashes, and can be resold to their original market - whereas hardcoded Bitcoin-specific devices cannot).


Hence I'm a bit confused by this thread. I'm willing to pay for a 300 MH/s bitstream, but the price can't be much more than the actual difference in income between a *free* 208 MH/s bitstream and Dr Tyrell's. If the assumption is that the first sale of unencrypted code will result in piracy, then Dr Tyrell needs ONE big customer... and how many FPGA miners have scaled up to thousands of units yet? This big customer also has to be found *before* the open-source effort finds out how to replicate the design.

So - is there a 'group buy' process in place yet? I'm happy to club together with other multi-FPGA miners to get a *significantly* faster bitstream... but if the open-source guys figure out the 'tricks' then I'll just use that... certainly this thread has given the chances of Dr Tyrell's work actually being 'independently discovered' a big boost, due to the information he's supplying.

I totally agree with you catfish! You can count me in such club! But before all Dr Tyrell should inform us whether he is willing to sell his bitstream to us or not!

And in case he is not! We may then fund the open-source project to speed up its development!

Ooh new speed update, what do we have to do to encourage you to make this available to all to use?

Even if only binary blobs initially.

X6500's would be awesome with this

Sorry for falling off the radar there.  Real life, quality time with git bisect, and some voltage drop issues on my own boards conspired to slow things down the last week or so.

Yes.

The BFL single definitely isn't a Spartan 6.

BTW, I will offer a 10BTC bounty to anybody who posts the JTAG IDCODE readout from the BFL single -- merely to satisfy my curiosity.  There was a JTAG header on the last PCB I saw them post.

Interesting use case; so essentially no added latency using them this way?

Nah; I use the DSP48s as big fat FIFOs; they have lots of registers inside and if you configure them right everything's a no-op.

This isn't my "secret sauce", but it is unique to my design.  When I run out of SRL16s in the places where I need them, I use the DSP48's as 32-bit-wide 16-bit-wide, 6-bit-deep FIFOs.  Useful trick.

Very interesting thread but I can actually chime in on this conversation about using a phase-change system to remove heat. If you're talking a traditional single-stage gas system like in your refrigerator then forget about it. The piping required plus the MINUSCULE load for each cold-head would make this cost prohibitive. Your best bet would be to repurpose a mini-fridge, use a proper condenser and throw a TXV for refrigerant metering and use something like r134a or n-butane/iso-butane and aim for evaporator temperatures in the 0-20 Celsius range. Then stick with your dinky little heatsinks and fans and not worry about having to mill expensive evaporators for such a small heatload.

Mini-fridges or even something like a deep-chest freezer would be the perfect insulated box to work with. The issue is such systems are designed to remove the heat from a load that doesn't generate additional heat. Without modification you will kill a freezer/fridge. That's where the replacement condenser and TXV come in to place. Make the compressor happy and you'll have shockingly low compressor loads and could very well run these FPGAs at astonishing speeds.

That all being said, compressed gasses are fun but can easily explode in your face with dire consequences if you aren't careful. Plenty of forums out there for amateur refrigeration. Take a gander at some of the things people have made and consider the tool costs. My own set of tools and gasses would buy a number of FPGAs and likely make me more money in the process

Any good solution for that will be some kind of DIY, otherwise it will be either unsuitable or expensive.

For phase-change I would use two contours - one with normal liquid passing through many special waterblocks and second with freon for cooling the first contour (if we need lower-than-environment temps, or course).

What about just using a BFL single instead of wasting half the electricity on cooling? 

How would a phase-change heat pump work for cooling a chip? Is there an off-the-shelf device? Or would it be DIY?

For the peltier I realize power is needed and heat is also produced but if your goal is to cool the chip it would work for that and peltier makers target this application of their products.  e.g. micropelt

Perhaps a better way to efficiency is to create your own cheaper power and don't create as much heat by running miners at lower freq.

Peltiers are cool and handy, but for big mining operations phase-change heat pumps will be FAR more efficient.
Remember that Peltier modules consume a lot of current and act almost as 200% efficiency heaters on the other side :)

Yeah cause those are free and need no power to run!

Maybe a peltier device would suffice.

Hahaha - funny that you mention it.
You guys have found that out the hard way, haven't you?
(By the way, I have a total of 12 BFL singles on order, so I'm not anti-BFL at all.)

But you are correct and you raise a valid point.

At 200 MH/s, Dr. Tyrell's design dissipates about 8 W, and so it's fair to assume that it dissipates 12 W at 300 MHz, which is
probably stretching the boundaries of a tiny 20mm x 20mm plastic chip like that. I mean, you can mount a big cooler on it,
but there is a thermal resistance from the FPGA die to the cooler.
Maybe these devices have to be run inside a freezer to successfully achieve a consistent hash rate of 300 MH/s and beyond.
Time will tell.