I Know we want to keep the parts count low, but I like indicators.
A green LED to show power and a blue LED to show operation, controlled by the PIC? I don't know what the resource map of the PIC looks like. Maybe there isn't a port available.
Do we need a momentary switch for resetting the board?
Topic: Klondike - 16 chip ASIC Open Source Board - Preliminary - page 170. (Read 435369 times)
That be sweet. 40 x 40 configuration in a rack mount. 512 chips 144+ GH/s. Using what? 2 x 1000 W PSU to be safe?
http://www.aliexpress.com/item/Huntkey-baisheng-4u-industrial-computer-case-s400-4u-rack-server-computer-case/735776385.html
2 layers and I am good. Just need to find a good home for the remaining 28 chips.
http://www.aliexpress.com/item/Huntkey-baisheng-4u-industrial-computer-case-s400-4u-rack-server-computer-case/735776385.html
2 layers and I am good. Just need to find a good home for the remaining 28 chips.
What are your thoughts regarding powering the bricks? One big PSU or individual ones for each klego (pain to power up...)?
I just wanted to post a better diagram of what I had in mind with Klegos.
This was very simple and cheap to implement at the board level. 5 thru-holes at front, left and right ready for joining to each new board, in an unlimited array.
I'll have a small thumb size repeater that can relay the I2C data to extend how many boards can be stacked and tiled. (This is 3D, since a 3 pin hdr stick can be placed at the corner to raise a floor).
This diagram shows the path I2C data takes from master to all it's slaves in the array. When the signal isn't strong enough we simply slip a repeater stick onto the joiner pins. I love this because it gets rid of having to make little jumper data cables to connect all these boards together.
If you want to see this in detail the original SVG file has been put up on github.
256 chips in 40cm x 40cm.
(oh, I missed two jumpers in the diagram)
This was very simple and cheap to implement at the board level. 5 thru-holes at front, left and right ready for joining to each new board, in an unlimited array.
I'll have a small thumb size repeater that can relay the I2C data to extend how many boards can be stacked and tiled. (This is 3D, since a 3 pin hdr stick can be placed at the corner to raise a floor).
This diagram shows the path I2C data takes from master to all it's slaves in the array. When the signal isn't strong enough we simply slip a repeater stick onto the joiner pins. I love this because it gets rid of having to make little jumper data cables to connect all these boards together.
If you want to see this in detail the original SVG file has been put up on github.
256 chips in 40cm x 40cm.
(oh, I missed two jumpers in the diagram)
There could be a number of reasons that they choose small boards. I think they made a reasonable choice.
A smaller board is certainly easier to debug and troubleshoot and costs are less. If there is an issue with some components, the whole board shuts down and taking down 10 chips instead of 80 is better. Also easier for repairs.
There are also possible issues with a single large board with warping and flexing causing solder and component cracking with temperature cycles that are less likely with a smaller board.
The cost of populating a large board is probably also more expensive. Bigger equipment needed to handle the bigger footprint and the placement and verification of the placing of over a thousand components compared to 150 - 200 or so (10 chip board).
A smaller board is certainly easier to debug and troubleshoot and costs are less. If there is an issue with some components, the whole board shuts down and taking down 10 chips instead of 80 is better. Also easier for repairs.
There are also possible issues with a single large board with warping and flexing causing solder and component cracking with temperature cycles that are less likely with a smaller board.
The cost of populating a large board is probably also more expensive. Bigger equipment needed to handle the bigger footprint and the placement and verification of the placing of over a thousand components compared to 150 - 200 or so (10 chip board).
do you think there was a specific reason the layout of the avalon boards is what it is? ie groups of 10 in larger modules. I would have to wonder about why they didnt go for one big board instead of groups of small boards. Not having the data sheet and limited information on how these chips work, I think it would be pretty perspicuous to think large chip boards are going to be designed and working before the smaller chip count boards.
Im not trying to say one is better then the other, as I honestly have no clue what the "best" design may be. I do hope that there are multiple board designs with different chip counts so a greater number of people can get involved.
Im not trying to say one is better then the other, as I honestly have no clue what the "best" design may be. I do hope that there are multiple board designs with different chip counts so a greater number of people can get involved.
Time constraints... pricing as well and that rush to market might be the reason as well as limited chip supply or even the idea how can we get 10 more GH/s out of a machine compared to the competition. Their design parameters were likely based on those sort of issues I think. Could be wrong obviously but given the price point and total hash power... they went that route.
I think BKKcoin has a different objective? Can you talk about your thinking about the Klondike so far BKKcoin? Your plan basically pack as many chips as you can into a small space and make them cheap as possible right? Max gh/s per square cm? If he guesses right and the chips do not have to be arranged as maximum of 10 per board then I think the design of the avalon had different goals than BKKcoins idea.
Great info guys... thanks for that... I am going to work on those ideas you posted and come up with some possible configs. Might just use my third floor at school and turn it into a server area with small AC unit and server rack and compare that to data center pricing. Might be able to get something customized here in Indonesia or used.
do you think there was a specific reason the layout of the avalon boards is what it is? ie groups of 10 in larger modules. I would have to wonder about why they didnt go for one big board instead of groups of small boards. Not having the data sheet and limited information on how these chips work, I think it would be pretty perspicuous to think large chip boards are going to be designed and working before the smaller chip count boards.
Im not trying to say one is better then the other, as I honestly have no clue what the "best" design may be. I do hope that there are multiple board designs with different chip counts so a greater number of people can get involved.
Im not trying to say one is better then the other, as I honestly have no clue what the "best" design may be. I do hope that there are multiple board designs with different chip counts so a greater number of people can get involved.
You could have card slots 20cm high with guides and they slide in. As long as you get good air flow, and solid mounting almost anything should work. Mine will just be sitting on a wire rack here at home.
This is what I had in mind. But you could also strip everything out of a 4U cabinet except fans and stack them 2 high and as many wide as can fit.
Looks more like Klondike Sudoku.
What about fitting them into a server rack? Thoughts?
rack dimensions is 48.26 cm x 44.958 cm . You could have 4 * 20x20cm boards per layer.
If you have to use a case (hosting in a datacenter) i guess a 1U case should be sufficient for 1 layer(by removing fans from heatsink and placing them in front/back of the case). Otherwise just stack em up using spacers.
1U would be great but 4U might be cheaper as you can use tome "regular" heavy duty 120/140mm fans and regular ATX PSU.
I just bought some Techolo TS-S200 4U 1mm steel (equiv. to Rosewill RSV-R4000 or sth, I think). Plenty of space for Klondike boards (mounted GPU-like and possibly in "column" too, i.e. more than one per row), 2x 120mm fans and another 5 optional.
I'm also thinking PCI-E connector/riser to 1/ get the power directly from the mobo and 2/ attach the boards "neatly" without needing custom rigs. If the PCB is less than 10 USD, spending even 20 USD won't make much of a difference when you add the ASICs.
Welcome to Klondilopolos or is that a Klondominium.
Call it KLego for Miners.
+1
Klego Brix to build a Klondominium.
Maybe leave it for v2.0? We want v1.0 out the door!
Looks more like Klondike Sudoku.
What about fitting them into a server rack? Thoughts?
rack dimensions is 48.26 cm x 44.958 cm . You could have 4 * 20x20cm boards per layer.
If you have to use a case (hosting in a datacenter) i guess a 1U case should be sufficient for 1 layer(by removing fans from heatsink and placing them in front/back of the case). Otherwise just stack em up using spacers.
Looks more like Klondike Sudoku.
What about fitting them into a server rack? Thoughts?
I would just use the corner holes with spacers to bolt them to a case like a motherboard. I could see an upper and a lower plane, heat sinks facing each other, with a row of fans at the end blowing air thru the tunnel between them. But there's lots of ways to mount them depending on what you have to work with. What about fitting them into a server rack? Thoughts?
You could have card slots 20cm high with guides and they slide in. As long as you get good air flow, and solid mounting almost anything should work. Mine will just be sitting on a wire rack here at home.
Looks more like Klondike Sudoku.
What about fitting them into a server rack? Thoughts?
Do you need to populate all 64 chips on the board?
can you run a 64 board with only 40 chips?
You can have as few as you like and add them later.can you run a 64 board with only 40 chips?
I'm not sure yet if it will auto-detect or if you'll have to configure it. Also, it's not definite until we have full chip docs from Avalon. There could be some internal chip constraints I'm not aware of.
Do you need to populate all 64 chips on the board?
can you run a 64 board with only 40 chips?
can you run a 64 board with only 40 chips?
if someone is interested in a distribution business, you can find people like http://www.screamingcircuits.com and http://www.sunstone.com to do your pick and place and pcb fab, respectively. With doing 100+ boards, assembly and PCB should come out to be around $50-60/board +parts for the klondike 64 (just guessing some of the specs from bkk's posts). Add in $600 worth of asics and you can make a 16 GH/s miner for ~$700. put some markup on that, might be an interesting venture.
Wow. $100 each, Qty 50, to just assemble the K16, not including parts and board making. I'd suggest checking what sensei will be able to offer. There are several assemblers hovering me with questions and if they get something going where I ship them boards/parts and they build then I'd hope it comes in well under this.
That would be awesome. I bet a populated 64 chip board for ~10btc would be very popular and would have about 5 btc margin per board.
Probably another salesman of that company .. notice his post count only 77 and join in May 2011 must be sleeping all this while but is now awake
Are the 4 boards in the K64 already daisy chained together?
The K16 design will have the I2C pads in an L shape at two corners of the board. This allows them to be panelled in any size with simple pin hdrs/jumpers. The 64 design would just be 4 of these pre-arranged similarly, and would also allow panelling them in any size. The 64 chip board only needs 1 usb connector installed per board, but could conceivably have 2 for redundancy. Power connectors still need to be installed. On the K64 that means 4 connectors and a splitter cable.
Together with the corner holes for stacking I can imagine some pretty nice cityscapes made up like this and all using NO cables:
Welcome to Klondilopolos or is that a Klondominium.
Call it KLego for Miners.
if someone is interested in a distribution business, you can find people like http://www.screamingcircuits.com and http://www.sunstone.com to do your pick and place and pcb fab, respectively. With doing 100+ boards, assembly and PCB should come out to be around $50-60/board +parts for the klondike 64 (just guessing some of the specs from bkk's posts). Add in $600 worth of asics and you can make a 16 GH/s miner for ~$700. put some markup on that, might be an interesting venture.
Wow. $100 each, Qty 50, to just assemble the K16, not including parts and board making. I'd suggest checking what sensei will be able to offer. There are several assemblers hovering me with questions and if they get something going where I ship them boards/parts and they build then I'd hope it comes in well under this.
That would be awesome. I bet a populated 64 chip board for ~10btc would be very popular and would have about 5 btc margin per board.
Are the 4 boards in the K64 already daisy chained together?
Jump to: