Power regulator is the APTS050A0X3-SRPHZ module. And the junk around the module is not heat stress, but flux residue.

Most of the board is reflowed, but I decided to hand solder to power module later. This turned out to be quite a messy deal. The pins of the module are hard to reach underneath the PCB, and the solid copper pours around the module make it hard to get the temperature up.

We haven't actually tested that many components. We picked a certain type, tested it, saw it performed well, and stuck with it. In the bitcoin world, time = money Around each ASIC there's a ring of 3x5 small caps from AVX, part #04026D105KAT2A (0402 size, 6.3V, 1uF, X5R). In addition, there are groups of 8 caps, 47uF, 6.3V, X7R. On the S-HASH board they are 1206 size, but on the H-CARD they're 1210 because these are a lot easier to get, and there's plenty of room anyway.

The amount of caps is probably overkill, but you can never have too much overkill.

Ok. No problem. I don't have any samples anymore, unfortunately.

No, it's not the same board. Look at the pictures:


The right side of the board with the 16 ASICs is virtually identical, but the left side is different. On the H-CARD from bitfurystrikesback.com, you see that the left side is much simpler. It only has the power regulator for the chip core voltage and the edge connector.

On my S-HASH board, the left side contains a different power regulator (50A instead of 30A), and a CPU with Ethernet interface. Functionally, they are different too. The H-CARD on bitfurystrikesback.com plugs into an M-BOARD, and the M-BOARD hooks up to a Raspberry Pi computer board with the miner code.

My S-HASH board works stand-alone. You don't need any other boards, no Raspberry Pi. Just hook it up to 12V power supply, plug the ethernet cable in your internet router, configure it for a mining pool, and start making coins.

Of course, the S-HASH board will be more costly to produce than a bitfurystrikesback H-CARD, simply because it has more components and a more expensive power regulator. Of course, when you add the M-BOARD and the Raspberry PI, the S-HASH board will be cheaper to produce. But if you need more hashing power, and stick in additional H-CARDs, the bitfurystrikesback system will win again. So, it all depends on what people need.

The other difference is that the H-CARD runs at a lower core voltage, and will be more power efficient. So, for large applications where electricity is an important part of the cost, that may turn out to be a better choice. For small home miners, the cost of purchase will be more important, and power consumption/heat generation will be secondary. In that case, running the chips faster will get more coins for your coins.  

This line in spitest.c checks if the job is switched:

if (newbuf[16] != oldbuf[16]) { /* Job switched! Upload current vector and start processing new one! */

When a job switches, all nonces have been scanned (actually not all, as the cores only scan 756 out of every 1024 possibilities). While the cores work on the first job, you must prepare the next job, and send it to the chip. As soon as the job is finished, the chip will start on the next one, and you can read the results of the first. When the chip has produced some nonces they need to be decoded and checked in various ways (see spitest.c code).

Yes, that's the CPU model, but there are plenty of other models that could replace it without loss in functionality. Apart from ethernet, plenty of RAM and enough core power to do some SHA256 hashing the are not too many requirements.

The firmware is proprietary.

Edit: to give an example about the SPI interface, you can look at bitfury's example/test code here.

It's not a secret. Bitfury described it earlier in this thread. There's a SPI command that causes a chip to forward all SPI signals to/from the next chip. So, when you're talking to the first chip, send the command, and then you'll be talking to the next chip, send the command again, and you're talking to the 3rd chip. After you talked to the last chip, send another command to break the chain. This command will be broadcast through the entire chain, and cause each chip to stop forwarding. After this, you're talking to the first chip again and the process can be repeated.

No, not at this moment. I'll first see if there's anybody interested in a commercial collaboration.

Also, mounting QFN48 chips with thermal pad and a few hundred 0402 parts is not something that a lot of people can do reliably.

This Lineage Power module.

(Hmm... forum ate my reply. I'll edit this post and add it again).

Pretty much 40GH, yes. This is just a random sample of 16 chips, from what I suppose was the first batch of ASICs. There's some variation between them, and there's some fluctuations as I measure the hashing rate based on the count of valid nonces produced, and not the operating speed.

The voltage regulator on the H-CARD has a 30A capacity, which limits the overclocking range, and there are no simple replacements for the regulator.

For cooling I have a standard PC case fan set up above the board, blowing down on the surface. No heat sink, so it's not an optimal set up. The board has a couple of mounting holes, and a flat back, so the idea is to mount a heat sink on the back.

No plans yet, just having fun. I will not be doing any manufacturing or retail selling myself, but I'm open for collaborations with other companies.

First prototype of Bitfury S-HASH board is hashing:


Features:

• 16 Bitfury ASIC capacity
• Adjustable (through 0805 SMT resistor) voltage regulator between 0.7 and 0.9V
• Core voltage regulator has 50A capacity, so chips can be overclocked.
• On-board ARM Cortex M3 processor with standard RJ-45 100 Mbps Ethernet port.
• Built-in mining software can operate stand-alone. No PC or Raspberry PI needed, just an internet connection.
• TCP/IP stack with DHCP and DNS support. Just fill in pool server name, port number, username and password.
• Support for Stratum and backup mining pools.
• Built-in small webserver for chip status/speed reports.
• PCB temperature sensor, could be used for automatic shutdown when temperature gets too high.

If you don't have the budget for a large number of chips, overclocking is the best option, as it will get you 40GH/sec out of a card (probably more with better cooling), instead of 25GH for a standard H-CARD for the same 16 chips. At 40GH/sec, the card uses about 35 Watts, running off a standard 12V DC supply.

Looks good....

Yes, they plug in directly. And yes, if you buy the master board together with the h-cards, the distance between the cards is fixed at 20 mm. Also, the SPI bus goes in one card, through all the chips, and then out on the other side, into the next card. So, if you get a master board with only 8 h-cards, you must use slots 0-7.  You can't skip slots, because it would break the SPI chain. Of course, if you're not afraid of soldering, you could put some jumper wires across the unused connectors, or make some dummy PCB edge connectors that connect inputs to outputs and stick those in unused slots.

Here's a picture of the master board:

If you look at the board here: http://www.bitfurystrikesback.com/product/bitfury-55nm-asic-h-board/, you'll see a main area with the 16 ASICs, and a strip at the bottom with the edge connector and power regulator. On the other side of the board, there are a handful of small components underneath the power regulator. There are no components underneath the square 110x110 mm ASIC area.

The spacing between the boards depends mostly on your cooling arrangement. The closer they are, the harder it will be to get sufficient airflow. I think 20 mm is a good lower limit. The thickness of the components sets an absolute limit of about 10 mm (excluding heat sinks), but it will be challenging to get the heat out.

There is no temperature sensor on the boards.

The chips are quite small, and dissipate most of their heat through the big ground/thermal pad on the bottom, and plenty of vias spread that heat effectively through the PCB. If you want to attach a heat sink, it's best done at the back side of the PCB, which has been left free of components in the main ASIC area. Six mounting holes in the PCB could be used for screws, or you could use an adhesive type.

We're still doing full temperature tests. However, for a single hashing card, the dissipation is actually 15-20W, not 40W.

Master board is 80x368 mm. Hashing board is 110x139 mm. It's connected using SPI (on the GPIO connector). I don't know about the software.