For our mining rigs we have chosen to specify power supplies that are rated 20% higher than the highest wattage we expect even including significant overclocking. The GN chip uses 250W at nominal, and we have designed all the systems to handle upto 350W. When the mini motherboard is supplying 350W to the chip (this level of overclock should result in up to 540GH/s) the mini motherboard may consume up to 411W of 12v, due to losses in the VRM. Add 25W for fans, and you get to 436W per mini-motherboard. Multiply by 3 chips in the Sierra, and you get about 1300W. We've added a safety margin of 20% extra on top of this - so we needed power supplies that were rated to a minimum of 1560W @ 12v to meet these specifications, and maintain the headroom and margin we desired to ensure reliable, efficient long term operation.

There are very few manufacturers who make single 1600W supplies, and we didn't find a single unit from a high quality vendor. In addition this size of supply is more expensive than 2 smaller supplies. The Sea Sonic supplies are really well built, and even significantly overbuilt. In all tests they are capable of supplying power well above their ratings, and the output regulation is very accurate and smooth. They handle hot operating environments very well. Please go read the reviews of the supplies in the Sierra - the Sea Sonic X-850FM3.

http://hardocp.com/article/2013/04/04/seasonic_xseries_x850_power_supply_review/9#.UmDCKVCcc5g
http://www.kitguru.net/components/power-supplies/zardon/seasonic-x-series-850w-km3-power-supply-review/7/

Each GN chip is on it's own mini-motherboard, and each mini-motherboard has 2 separate 6 pin PCI-E power connectors. Each of the supplies feeds 3 of these connectors.

Datasheet says 38.35mm and the fets are 0.7mm nominal

Main aim was to get the down stream one a good distance above the water pump.

Contact sales@hashfast.com, they should be able to help you.

We are preparing a special promotion just for batch 1 baby jet purchasers. Coming soon.

Yes - chips will be available for developers. Please contact sales@hashfast.com for more info until we get them on the website.

Look forward to a special promotion just for those first batch customers. Launching soon.

The airflow is front to back - the push/pull fan configuration moves so much air that the temps in the box are kept reasonable. The coldest air hits the rad first, which is where it's needed to keep the chip temperature low. Also it's not possible to put the rad at the back of the box, since the PSU has to go along the side and reach the back - would mean a narrower rad, with lower capacity.

Chips and systems out at end of Feb 2014 - that means tape out sometime in December 2013, which means you've only just, or have not yet started physical design. Without being well into the physical design stage how do you know what your power or performance numbers are going to be with any confidence?

The package is a BGA, containing a multi chip module. There are 4 dies, each 9mm x 9mm, spaced out by 5mm. The core voltage and frequency vary according to the cooling available to the chip. The chip contains a temperature sensor on die, and increases or decreases the operating voltage and frequency to maintain a target operating temperature at the die. The allows the maximum possible performance to be achieved, given the cooling that is available. In a colder environment the chip will operate at a slightly higher voltage and frequency, and return a higher hash rate than in a warm environment. Simulation runs show that the best silicon will have a TDP of 250W when operating at the name plate (nominal) 400GH/s. Worst case silicon will consume a few % more power to reach this nominal 400GH/s. Note - simulation results can be out by +/- 20%, although they typically come in high (expect lower numbers in real silicon).

Overall, our chip most closely resembles a modern "hot CPU" design (Intel Sandy Bridge E, AMD FX-9, SPARC T4, etc.). The package is a BGA, 45mm x 45mm external dimensions. The total die area is approximately 324mm^2. The die is split in 4 - i.e. under the metal lid there are 4 dies in a square arrangement, each approx 9mm x 9mm with 5mm gap in between each one (for better heat dissipation and spreading).

One highly relevant feature regarding power use is that the GN chip incorporates on-die temperature sensors and a control system designed to adjust voltage and clock speed to the capacity of the cooling system.  Thus if the cooling system can dissipate a greater amount of heat, the software can "overclock" the chip to fit it's power usage to the heat dissipation capacity, and produce greater hashing capacity.

Similarly, the chip will "underclock" itself in response to external circumstances that reduce the available heat dissipation capacity (say, a very hot day, a failure in the cooling system, a blocked air vent...).  The overall design intent is for the chip to always operate at the maximum possible hashrate dictated by the circumstances.

Finally, one recent piece of news is that we have received results for the stage-III thermal test (full physical prototype) from our cooling system partner.  Stage III tests involve the creation of a full and complete prototype of the system, including the same case, fans and cooling system that will be used in the Baby Jet as shipped.   The only difference is that the chip is substituted by a variable-output heating element with the same form factor as chip. The test consists of running the Baby Jet as if in production, and increasing the wattage produced by the heating element while monitoring temperatures.

At the nominal operating point (400Gh/s) the best silicon will consume ~250W according to our simulations. There is some variation in silicon however, so some silicon will consume a few % more. This power level is at the chip only. The system has 2 power conversion stages between the wall plug and the chip - first an ATX power supply that outputs 12v. This supply is about 88% efficient. Then there is a second supply stage on the module board the chip is mounted on. This second supply stage converts the 12v down to approximately 0.7-0.8v that the chip runs at. The combination of the losses in both PSU stages and the additional consumption from the pump, fans, controller etc account for the difference between the 250W at the chip and about 350W at the wall.

The module board is ~4" wide and approx 10" long (may end up a little longer, up to 12" - we are still configuring the power connectors). Here is a draft layout - note this is not final and is subject to change. Dimensions in mm.

https://hashfast.com/wp-content/uploads/2013/09/draft-layout.gif

The tallest board components are two conventional air cooled heatsinks on the FETs that form part of the power supply. I don't have the height to hand - I will get back to you. Here is a rendering of a draft of the module boards, installed in a Sierra. Note - this design is not final, and is subject to change.

https://hashfast.com/wp-content/uploads/2013/09/ISO-RENDERING-1.jpg

Yes - the controller (which is a raspberry pi) should be able to control many boards. We have not analysed how many. Our chip has been designed with a particularly advanced internal interface, and supports variable difficulty levels and n-time rolling on chip. This massively reduces the traffic from the controller to the chip, keeping the CPU requirements on the controller very low. The interface supports daisy chaining up to 63 chips to a single serial port on the controller. The traffic levels and limits will depend on the difficulty levels the chip is configured to run at, and the use of n-time rolling.

Yes - we plan to add complete module boards to our website shortly. These can be directly controlled via either a serial port, or USB. We will be open sourcing the drivers in CGMiner.

There is a temperature sensor on the chip, and as long as the open source software has not been incorrectly modified, it will limit the temperature. In fact the software will automatically hash at the fastest possible speed, limited only by temperature. If you deliver better cooling, or colder air, it will go faster.

Because with open source software, and overclocking we don't know what people will try and push these things to do.

350W is a rounded up number for the whole system, including the power lost in the 2 stages of power supply, and fans etc. The chip itself draws 250W @ nominal.

Depends on how hard you cool and overclock your Sandy Bridge E. The company that is assembling our systems specializes in overclocking. They run Sandy Bridge Es overclocked to 350W (CPU power alone, not whole system), using the same cooling system we are using. We are also using a heatspreader.

Metal migration is a well understood phenomenon. We have followed all the fab's rulesets for electromigration so that the current levels we're going to see will not be a problem (even current distribution, and thicker metal layers). Currently in the simulator for EM our chip passes the test for a 5 year lifetime, but fails the 11 year test - and that is running somewhat overclocked, at about 540GH/s.

Hashfast's chip at nominal (400GH/s) is expected to consume about 250W of power, not 350W. i.e. 0.77W/mm^2. Overclocked Sandy Bridge E in shipping commercial products with the same cooling system runs 350W with a 425mm^2 die, 0.82W/mm^2.

Exactly that - they require a 'mock tapeout' before the real thing in order to check for simple mistakes. Things like that the metalization layers for the macros we have included (e.g. licensed temperature sensor) match up with the rest of the chip, and with what's on our order.

The Baby Jet will support standalone mining via an Ethernet connection, and upgrade to WiFi via a USB WiFi card.

The original design work on this chip started in 2011. The architecture work started then. The current version of the RTL code started in May 2013. The team have been working non stop 18 hour days, 7 days a week since then. We understand the importance of schedule. The chip design is almost complete. We're starting sales now because we're at the point where we the calendar of remaining items is fairly fixed, and we're confident we can deliver.

Simon

We are working on finalizing a guarantee program for purchasers that will provide some good coverage in case of late delivery, the text is being perfected now, and we will be publishing it in the next few days. It will be offered free to all existing purchasers.

Orders are taken in BTC, in the unlikely event we get to refunds they will be given in BTC.

Yes - production and fulfillment are outsourced - design partly in house and partly outsourced. These days there are many professional companies who specialize in these areas, ISO9001 certified, etc. No need to keep these in house.