DIY PCB with AVALON: "The Quarter Stick" - Needs Help!

[Tyger]

a some reverse engineering can get you there. but yes  it will take to long, just w8 for Avalons expention boards

[daemondazz]

It may be wise to limit a board like this to only 40 ASICs to keep power down and have 10 chips per serial line. Smaller, cheaper PCB as well.

Without knowing the exact power draw of each chip, estimates I've read seem to be around 2-2.1W per chip. Maximum draw from a PCIe slot is 25W unless you're a graphics card in a 16x slot, then you can have up to 75W.

So, 25W would likely allow for 10 chips, plus the PCIe bridge and a microcontroller of some description. This would be without needing to use extra power connections like most higher powered graphics cards need to get higher than the 75W.

[BkkCoins]

It may be wise to limit a board like this to only 40 ASICs to keep power down and have 10 chips per serial line. Smaller, cheaper PCB as well.

Without knowing the exact power draw of each chip, estimates I've read seem to be around 2-2.1W per chip. Maximum draw from a PCIe slot is 25W unless you're a graphics card in a 16x slot, then you can have up to 75W.

So, 25W would likely allow for 10 chips, plus the PCIe bridge and a microcontroller of some description. This would be without needing to use extra power connections like most higher powered graphics cards need to get higher than the 75W.

You would add 6 or 8 pin PCIe power connectors just like GPU cards. All miners are set up for that already so it's an easy drop in switch over. ATX supplies have the PCIe power conns. Docs indicate 6.6W / 1GH/s which is 6.6*0.282 per chip = 1.86W. I do believe that is at the wall. Which means you can allow 90% efficiency discount twice when figuring on board regs for 1.2V. eg. 1.86W * 0.9 *0.9 = 1.5W per chip, or 1.25A. So an IR3895 buck regulator (16A) should be capable of powering 12 chips. I'd go with 8 or 10 chips per regulator. 3.3V can be pulled from PCIe bus as it's minimal draw.

No need for a micro-controller here as the ASICs are a serial shift register chain design. By using a serial interface PCIe controller like the MCS9901 you put all the dev work in cgminer driver and don't need to waste time coding firmware for a micro-controller. You'll need to wait for the Avalon design materials and I'm guessing to some extent based on the Icarus and how he did that but I think this will be fairly close.

Current GPU miners could pull out a GPU and replace it with a 40 chip ASIC board giving 11.2 GH/s.  And using 75W instead of 150W. Sweet.

edit: Here's another PCIe serial bridge I found from Pericom. This one is actually carried by Digikey and is in stock at $15 each. Mouser carries it but non-stock.

http://www.pericom.com/assets/Databriefs/PI7C9X7954_db.pdf
http://www.pericom.com/assets/Datasheets/PI7C9X7954.pdf

[atcsecure]

I don't think pci-e is the way to go, something external with ethernet, wifi or usb or any mix and something expandable, but keeping it as simple as possible - at least 3Ghash/s per module

[BkkCoins]

I don't think pci-e is the way to go, something external with ethernet, wifi or usb or any mix and something expandable, but keeping it as simple as possible - at least 3Ghash/s per module

I'm not planning to build with PCIe. Just exploring and posting info for others. I personally think PCIe is too difficult a protocol bus for people without deep experience in it. You can't just connect to it due to complex bus technology and differential signalling. Nowhere near as easy as PCI. But for someone willing to dig in it has the potential to be a fast, easy way to connect. One chip bus interface, pre-made drivers allowing treatment as serial devices could make it the fastest up and running solution. But if something doesn't work right at first you'll have a terrible time debugging and need special equipment to analyze bus issues. Major headache.

Personally, I'm building a standalone board with a simple serial connection and 16 chips. Plus a small PIC Ethernet micro-controller board with ready to use TCP/IP stack. I'm hoping this will get me up and going before others. It's a gamble.

4512 MH/s per board. Up to 8 boards per controller. But I don't have many chips so if it works I may trade assembling boards for more chips.

[atcsecure]

I don't think pci-e is the way to go, something external with ethernet, wifi or usb or any mix and something expandable, but keeping it as simple as possible - at least 3Ghash/s per module

I'm not planning to build with PCIe. Just exploring and posting info for others. I personally think PCIe is too difficult a protocol bus for people without deep experience in it. You can't just connect to it due to complex bus technology and differential signalling. Nowhere near as easy as PCI. But for someone willing to dig in it has the potential to be a fast, easy way to connect. One chip bus interface, pre-made drivers allowing treatment as serial devices could make it the fastest up and running solution. But if something doesn't work right at first you'll have a terrible time debugging and need special equipment to analyze bus issues. Major headache.

Personally, I'm building a standalone board with a simple serial connection and 16 chips. Plus a small PIC Ethernet micro-controller board with ready to use TCP/IP stack. I'm hoping this will get me up and going before others. It's a gamble.

4512 MH/s per board. Up to 8 boards per controller. But I don't have many chips so if it works I may trade assembling boards for more chips.

Sounds simple enough, hope it works... feel like printing off some extra pcb's?

[roybitcoin]

I don't think pci-e is the way to go, something external with ethernet, wifi or usb or any mix and something expandable, but keeping it as simple as possible - at least 3Ghash/s per module

I'm not planning to build with PCIe. Just exploring and posting info for others. I personally think PCIe is too difficult a protocol bus for people without deep experience in it. You can't just connect to it due to complex bus technology and differential signalling. Nowhere near as easy as PCI. But for someone willing to dig in it has the potential to be a fast, easy way to connect. One chip bus interface, pre-made drivers allowing treatment as serial devices could make it the fastest up and running solution. But if something doesn't work right at first you'll have a terrible time debugging and need special equipment to analyze bus issues. Major headache.

Personally, I'm building a standalone board with a simple serial connection and 16 chips. Plus a small PIC Ethernet micro-controller board with ready to use TCP/IP stack. I'm hoping this will get me up and going before others. It's a gamble.

4512 MH/s per board. Up to 8 boards per controller. But I don't have many chips so if it works I may trade assembling boards for more chips.

Hi BkkCoins

Do you think it´s possible make an standalone PIC based miner?  Or we need to relay allways to a host runing cgminer (for example)?

I´m  just started reading about the portocols and I dont understand everything yet 

I'm thinking in building an USB based 10 chips (as proposed by burning) just for learning (and mining of course) but I want to work an standalone concept.

[BkkCoins]

Do you think it´s possible make an standalone PIC based miner?  Or we need to relay allways to a host runing cgminer (for example)?

Yes. It's feasible. The PIC only deals with a small amount of info being relayed between the pool and the ASIC chips chain. I'm 98% sure it can handle this adequately but I could be wrong as I have not written the code yet. If it doesn't have the chops then my second choice would be replacing it with a RasPi. The main reason I'm using the PIC is because I need to make one for another project anyway, so may as well do this first while waiting for Avalon docs and then leverage it here. I have the PICs already and they only cost $3.

Sounds simple enough, hope it works... feel like printing off some extra pcb's?

If it works out I'll post my materials on my github repo. I'm not sure if I'll sell extras. I might use as many as 2-3 but I'll have to order 10 minimum. Some people have asked me if I'd assemble boards for them. I might but have to consider the risk of damaging their ASICs. Nightmare scenario.

[jeprokzki]

is there any diy mining rig builders group here that i can join in? because i dont have a knowledge in building this diy rig i'll just fund them to make me a mining rig

[daemondazz]

Well, my reason for going down the PCIe path is so I can put them into systems I already have spread around the place. Most of them are server class machines and not all of them have spare Molex or PCIe power connectors, so the limit to 25W.

It probably would not be hard to build a much higher density board for people that would be replacing GPUs that do have the power connectors available.

[crazy_rabbit]

If I recall Sparkfun.com will entertain building projects if they think there is a market for them. Why not pitch them on selling a DIY Bitcoin Miner? 

[atcsecure]

Do we have a baseline PCB design? I've gone through this thread but I don't see it... something maybe in freepcb?

[BkkCoins]

Do we have a baseline PCB design? I've gone through this thread but I don't see it... something maybe in freepcb?

There won't be a PCB design until documentation is released by Avalon in May. People are just guessing at various things now.

From what I recall reading Avalon is supposed to be open sourced and will include schematics, pcb files etc. So at that time even if you don't want to design your own custom board you should be able to take the pcb files and send them to a fab to be made as is.

Right now there isn't even chip pin out details.

edit: I had to go check that I wasn't remembering wrong. On the chip order page it states:

8. communication protocol, reference board design provided in early May.
9. everything will be open source from FPGA to PCB design.
10. we do not offer technical support of any kind, this is final.

[atcsecure]

Do you think it´s possible make an standalone PIC based miner?  Or we need to relay allways to a host runing cgminer (for example)?

Yes. It's feasible. The PIC only deals with a small amount of info being relayed between the pool and the ASIC chips chain. I'm 98% sure it can handle this adequately but I could be wrong as I have not written the code yet. If it doesn't have the chops then my second choice would be replacing it with a RasPi. The main reason I'm using the PIC is because I need to make one for another project anyway, so may as well do this first while waiting for Avalon docs and then leverage it here. I have the PICs already and they only cost $3.

Sounds simple enough, hope it works... feel like printing off some extra pcb's?

If it works out I'll post my materials on my github repo. I'm not sure if I'll sell extras. I might use as many as 2-3 but I'll have to order 10 minimum. Some people have asked me if I'd assemble boards for them. I might but have to consider the risk of damaging their ASICs. Nightmare scenario.

I'm willing to have the PCB"s printed and then do assembly.. I plan on ordering a lot of asic's for my own farm

[stripykitteh]

Well, my reason for going down the PCIe path is so I can put them into systems I already have spread around the place. Most of them are server class machines and not all of them have spare Molex or PCIe power connectors, so the limit to 25W.

It probably would not be hard to build a much higher density board for people that would be replacing GPUs that do have the power connectors available.

Just thinking out loud, do you think there would there be a market for PCIe mining cards for casual/novice miners?

[Bicknellski]

We have a variety of PCB designs on the go here... how many different designs are there currently.

Personally I will use a PCB manufacturer here in Jakarta so that I don't have to go through the hassle and pain of importing boards which can be a nightmare. Chips I can pick up and carry in from Singapore where I will have them drop shipped first with an agent.

I am really keen on developing a local bitcoin ASIC miner for Indonesia that is a 2 to 3 chip model, an 8 chip model and a 16 chip model. All I really need is a partner who can get me a working / tested board design and I will do the rest on this end.

If anyone is interested in working on this with me let me know via PM.

[daemondazz]

Just thinking out loud, do you think there would there be a market for PCIe mining cards for casual/novice miners?

I honestly don't know, but we can see

[shivansps]

Well, my reason for going down the PCIe path is so I can put them into systems I already have spread around the place. Most of them are server class machines and not all of them have spare Molex or PCIe power connectors, so the limit to 25W.

It probably would not be hard to build a much higher density board for people that would be replacing GPUs that do have the power connectors available.

PCI-E limit is not 75W for 1.1?

[daemondazz]

x16 graphics cards are allowed up to 75W, non-graphics cards in x16 slots and all cards in lower bus count slots (x1, x4, x8) are only allowed up to 25W.

[Signus]

What frustrates me is that people are assuming that we can take FPGA designs and treat an ASIC design the same way.

We have a variety of PCB designs on the go here... how many different designs are there currently.

Personally I will use a PCB manufacturer here in Jakarta so that I don't have to go through the hassle and pain of importing boards which can be a nightmare. Chips I can pick up and carry in from Singapore where I will have them drop shipped first with an agent.

I am really keen on developing a local bitcoin ASIC miner for Indonesia that is a 2 to 3 chip model, an 8 chip model and a 16 chip model. All I really need is a partner who can get me a working / tested board design and I will do the rest on this end.

If anyone is interested in working on this with me let me know via PM.

I think people are just guessing at the moment. Nobody knows what these Avalon chips need. For those people who have started a design they may be a step ahead of everyone, but they will practically have to scratch most of it once they realized they made a number of incorrect assumptions.

I'm currently waiting on the documentation to be released by Avalon as well before I start any design or BOM. However if anyone would like collaboration on PCB designs if you have an inkling of where to start, put up a GitHub repo and I'd be willing to assist in the process and give my 2 cents.

Just thinking out loud, do you think there would there be a market for PCIe mining cards for casual/novice miners?

I honestly don't know, but we can see

I would imagine so. For casual mining if you're not looking to profit, the hardware is less expensive. However the problem with PCI-E cards is the interfacing. Not as simple as USB.