Ok. This is what I was told by Stefan too. A lot depend on the external layout and components. So the efficiency numbers I quoted would make sense only for the ModMiner and X6500 as they both use pre-made modules.

From your link http://rijndael.ece.vt.edu/sha3/publications/DSD11SHA3.pdf I compute 400 Mhash/J for a 130nm ASIC. So probably ~1000 Mhash/J at 40nm. This is 50x better than a 45nm FPGA (Spartan6 = 20 Mhash/J). Edit: Actually 49 Mhash/J: 1000 (mJ) / 19.75 (mJ/Gbits) * 1e9 (bits/Gbits) / 1024 (bits per bitcoin hash) = 49.4 Mhash/J

But this still does not change my mind: the first mining ASICs will likely be manufactured on the 130nm node, so their 400 Mhash/J characteristic will make them a 20x efficiency increase over 45nm FPGAs. Not much different from past 10x technological leaps.

However I do agree that Mhash/dollar will be a more interesting metric to watch than Mhash/J. I wonder why you think ASIC will contribute a 1000x improvement in this area (going from $1 per Mh/s to $1 per Gh/s)?

I quoted efficiency when outputting 8A, for all converters, regardless of their max output current, so that we can compare them all fairly.

For one, devs don't care about all the mining business. I remember that Gavin, when questioned in an interview about Bitcoin last year, was unable to answer specifics about which GPU miner software to use, or even which GPUs to buy, because he was "uninterested and out of touch" with the GPU mining advancements.

You are delusional about the capability of ASICs

The efficiency increases from CPU, to GPU, to FPGAs, and to ASIC are all comparable. Roughly a 10x increase every time:
40nm CPUs do up to about 0.1 to 0.5 Mhash/Joule.
40nm GPUs about 1 to 2 Mhash/Joule.
40nm FPGAs about 20 Mhash/Joule.
40nm ASICs will do about 200 Mhash/Joule.

(All numbers above can vary with clock, voltage, etc, but roughly that's the baseline to have in mind.)

ASICs will only offer a 10x efficiency increase because a logic block like SHA-256 shrinks to about 1/10th the number of gates when ported from an FPGA to an ASIC. And the number of gates is the primary factor determining power consumption. As to the cost of ASICs, it won't be different from FPGAs. The Spartan-6 LX150 costs Xilinx maybe $10 to manufacture, but it sells for $100-150. Same thing for the future ASIC mining chips of the same die size, they will cost $10 and sell with a more or less similar markup. Bottom line, ASICs are not an "exponential leap in hash rate, power consumption AND eventually cost".

If the Bitcoin mining industry (vendors and miners) survived all previous efficiency increases, there is no reason it will not survive the next one. And the reason it survives the jumps is because 10x isn't that big of a deal. The electricity costs vary by even more worldwide ($0.02 to $0.40 per kWh), so mining with an ASIC with high electricity costs will be about as profitable as mining with FPGAs in an area with low costs.

Absolutely 100% agree with Vladimir's thoughts. ASIC mining is good for Bitcoin.

No. The traditional argument goes like this: Large miners have an incentive to protect their wealth (bitcoins or mining power). If they perform a majority attack, they will instantly destroy the trust in Bitcoin, and their own wealth.

I compiled a table comparing the voltage converters supplying VCCINT to the Spartan-6 LX150, across all board manufacturers. The table below lists the maximum current that can be provided to each FPGA, as well as the efficiency of the converter at 8A (for 12V input, and 1.2V output):

ngzhang's ICARUS	15A or 20A(*) per FPGA	Texas Instrument TPS40192 or TPS40193	unknown efficiency
Enterpoint Cairnsmore1	12A per FPGA	Micrel MIC26950	87% efficient (from datasheet)
btcfpga ModMiner	10A or 13.3A(**) per FPGA	Delta D12F200	84% efficient (from datasheet)
fizzisist's X6500	10A per FPGA	Murata OKR-T/10-W12-C	83% efficient (from datasheet)
Ztex 1.15x/1.15y	8A per FPGA	Alpha and Omega AOZ1025D	87% efficient (measured; datasheet says 75%)

(*) Some batches of ICARUS boards use the TPS40192 (rated 15A), others use the TPS40193 (rated 20A).
(**) The nice thing about the ModMiner is that you can decide to plug only 3 FPGAs in the mainboard, to give 13.3A to each FPGA, as they share a common 40A-rated voltage regulator.

eldentyrell: can you comment on how much current does your TML bitstream pull?

I love the simplicity of the design! There are so few components on the PCBs, compared to other boards (Ztex, Enterpoint, Icarus, etc). You seem heavily inspired from the X6500 design, whose design is also extremely simple.

Correct. This is why ponzi schemes work: people don't care where the coins come from.

Those who care don't invest and don't become victims of such schemes

No, but 830 + shipping, yes.

I received notification that 2 of my 10 singles ordered on March 27 were shipped on June 9, and that the 8 others should ship on June 11.

Dust is probably the root cause of your fan failures (are you mining at home?). My miners run in a datacenter with filtered air. When I have compared my low fan failure rate to other 5970 miners, dust has always turned out to be the main difference between my environment and theirs. Your fans would have likely failed just as spectacularly even when running at lower speed...

Also, most of my fan failures affect disproportionately 5970s, not 6990s, even when taking into account the fact that my 5970s are generally older than 6990s (I have 5970s whose fan failed despite having been in operation for less time than most of my 6990s). I would venture that the fan on 5970s is of lower quality than 6990s.

On high-density 5970s miners, at 25C ambient, the GPU temp diff between the fan at 50% and 100% is a full 10C or more. Not negligible!

- BFL said a mini rig card mines at 1.5Gh/s, therefore it will have 17 cards for 25Gh/s total (even though mini rig prototype pictures show it may have more cards, they don't show all angles).
- Each card should weight at most 1kg (heatsinks...) -- this is an overestimation, for reference an HD 6990 weights 1.16kg.
- The 1400W ATX PSU will weight about 2-3kg.
- The case and fans will weight about 10-15kg.

Therefore I estimate a mini rig will weight 30kg +/- 5kg, still kind of transportable by an adult in average shape

I heard painting strips make the mini rigs faster.

Seriously, I don't think any miner dedicated enough to spend $15k on mining hw cares about scratches on cases caused by manual assembly. Especially when there are many more legitimate concerns to have (underestimated delivery times, suboptimal cooling design, etc).

As a farm operator who is mining on dozens of 5970s and 6990, continuously, with the fan speed manually forced at 100%, and with my oldest cards mining for as long as 18 months, let me tell you this claim is rubbish! (My fan bearing mechanical failure rate is 6%, which is low and expected.)

The claim that "running fans harder than 80% kills them" is spread by paranoid teenager geeks who covet their expensive gaming hardware without an understanding of the electronic and mechanical engineering work to control failure rates.

When a fan datasheet says it is designed for a lifetime of X years of continuous operation, it assumes a 100% duty cycle. In other words, it assumes the fan running at full speed. Fan manufacturers don't build fans and say "don't run it at full speed, or else..."

For example, if the datasheet says 2 years, running the fan continuously means it should last at least 2 years; running it every other day means you can prolong its lifetime to 4 years; etc. At 50% duty cycle ("50% fan speed"), the PWM voltage alternates between 0 and the nominal input voltage, the fan should last somewhere between 2 and 4 years (but much closer to 2 years because the bearings are still moving when the voltage is 0, just not quite as fast as full speed.) At 80% duty cycle, the fan should last between 2 and 2.5 years (but closer to 2 years).

To significantly improve the operating lifetime of a fan, you would have to run it at such a low duty cycle that it would be a bad idea for high-end graphics cards, because the cooling would be so much reduced that the high temperature would start seriously reducing the operating lifetime of all the other components on the PCB (capacitors, VRMs, GPU ASICs, etc).

Bottom line, fans are designed to run at 100% duty cycle. And there is no significant difference between 80 and 100%. Some people choose to run at 80% because it does not significantly decrease cooling (for the same reason that it does not significantly increase lifetime), but because the fan consumes 20% less power (and the fan is just a tiny fraction of the overall power consumed by the card).

(I choose to run at 100% because I pay a fixed price per electrical circuit, and the tiny extra power consumption costs me nothing.)

Unfortunately I don't do loans.

Sorry, no trading. I don't have time to resell the 7970s. I could sell you a smaller batch maybe? 5 6990s for 365 BTC.

fizzisist, the downtime is not such a big deal, but my time is very precious. Why spend 10 minutes disassembling and re-assembling a stack to swap 1 board, when you could do it in 30sec if the board was not in a stack? If you have a small work bench, I can understand why you like the space-saving advantages of stacks of boards. But I don't lack space. I lack time. I work very hard to automate everything and optimize my time as much as I can. I run a farm approaching 80 FPGAs. The difference between a task that takes 30 sec per FPGA, and 10 min per FPGA, is 13 hours of work. So every tiny bit I do to simplify the maintenance of my farm saves me tons of time.

I will post pictures of how I arrange my own boards soon...