Topic: Modular FPGA Miner Hardware Design Development - page 31. (Read 119320 times)

I thought that for such small currents you will get a voltage regulator for something like 30 cents?


Well, they don't conform to the standard. The manufacturer puts in a chip that could support USB slave, OTG and host but they officially support only slave. By changing the driver in the linux kernel you can change the operating mode of the chip to host. At xda-developers some people did that. But they have to use a powered USB-Hub inbetween the devices because of the missing power supply from the host.

Of course that won't be a common use case. But if I have an old smartphone that is worth < 20 EUR it might be the most cost effective solution.

Wide input converters are costly, provide some lanes with the standard ATX voltages so the DIMM designer can choose of his requirements. DIMMs for SDRAM are mostly THT, this could be  good for stability. Small heatsinks are not a problem, there are already Dimms with that, the question is how much heat the can dissipate.

From theory you could provide some areas with a defined module height so you could mount one or two metal bars to secure the DIMMs and if proper designed some bigger heatsinks.

I doubt if such  'no power' USB-Hosts are a really cost optimized solution, and  we need a Power-supply for the FPGA anyway, so there should be enough for a small arm system.

Switching between the jtag-lines of the FTDI and the jtag-lines from the is just a question of an simple TTL multiplexer,  you would require  some driving chips anyway, at least for clock and reset and you have to watch out for the propagation delay on the jtag chain. The XDS510 syste mfrom Texas instruments there is a clock feedback  pin on the connctor for this ...
Maybe there is a microcontroller with usb device funcionality that ist comparable to the FTDI price wise, that could bitbang the stream into the FPGA, and store some config data that could reported to the Host.

I agree that the barrel connector cannot hurt. But what is a "standard" laptop voltage? The range seems to span from 17V to 24V (and I probably missed some here). Should the power supply support all?

Concerning DIMMs: is putting any other connectors on a board that is slotted into a DIMM socket safe? I am thinking of the stresses excerted on the socket due to the cables (or the heatsinks, if there are any).

Hi,

the idea behind host powered was simply to save one power supply (for 5V in the case of a FT2232D). How do these small devices provide USB host at all if they don't have 100mA? I thought the standard required at least that (and the chip would need less).

So... you are suggesting to have the daughter boards be also operable stand alone? While that is probably possible, what is then the point of having the backplane at all? Is this the idea of adding a complete computer (ARM or other) to the backplane?

If yes, then maybe O_Shovah as the original poster should clarify what he intended the purpose of the backplane to be: just cost saving or making a stand-along hashing appliance. If it is the latter, than the host-to-JTAG interface may not be common between stand-alone and backplane based.

Having a barrel connector in addition to the molex one definitely can't hurt. I'd favor 2.1mm ID, 5.5mm OD - Tip Positive, as most other devices have, not one of those more recent HP or Lenovo ones.

DIMM sockets certainly have their advantages (like locking clips), but they limit the width of the FPGA boards. For a rack-mount setup it might be desirable to have low profile (2-3U), but long FPGA cards with heatsinks in between and a couple fans on both ends providing airflow through those. PCI-style connectors seem to be suited better for that.

I'd seriously perfer a router-like web interface for that... It needs a network connection anyway, and buttons aren't exactly comfortable to use.

So maybe we can decide on the first few core points. Please tell me if you have any objections and if yes what you would use instead

- Every daughterboard has a power supply Molex 8981 and in addition one connection for a standart Laptop power supply for high power daughterboards.
  The power for lower currents is supplied through the connector Lanes of the daughterboards.

- Each board has one USB mini B port

- The connection between the Mainboard an the daughterboards is provided through DIMM sockets  ( low price, standart part ~3€) [ Number of Lanes to be determined]  

Okay. Then i request an autoupdater to a server where "bootcode" for every daughterboard is stored, so that i just have to insert my wallet address (you know.. by pushing little buttons on the backplane. it would have a little display for this, combined with btc counter!) and keep paying the electricity bills.

Thanks again.

Exactly, the ARM will have to boot the FPGAs, unless they have a configuration flash on the board itself.

I as a noob have a questin here: As different Daugtherboards/FPGAs should be supported, would this ARM CPU program these fpgas? Or how do you do this, as nobody except the backplane knows which types of daughterboards are connected?

Additionally I want to say that I really enjoy this thread here. Thanks for everybody involved. I guess I will personally switch in read-only mode for now, as I have only slightest experience with such electronics.

Depending on how you do this you might want to plug those things directly into power-hungry mining cards, as the 12V rail might not even need to be regulated. It's only the supply for various other regulators.


My intention was to use this ARM CPU as the host, and communicate via ethernet from there. No need for a sheeva plug.


I don't really see why one would ever want to do that. If the FTDI is the only chip that needs 5V power, bus-powered seems to be perfectly fine to me.

Hi,

I don't know too much about hardware, but I still want to comment on a few points:

1.) Power connector: I hate those molex-connectors but the ability to use a standard ATX PSU without hard-to-get adapters is worth going with it. But I also think that an ATX supply isn't the most efficient choice, when you use only the 12V rails. Also they are rather expensive, usually overpowered and need active cooling.

My suggestion: there are a lot of aftermarket notebook power-supplies, from very cheap (http://www.amazon.com/Notebook-Charger-Adapter-Thinkpad-40Y7659/dp/B002BW0MNU/ref=sr_1_1?ie=UTF8&qid=1309117838&sr=8-1 or http://www.amazon.com/Replacement-Notebook-Adapter-Charger-Inspiron/dp/B004FQPGG8/ref=sr_1_4?ie=UTF8&qid=1309117838&sr=8-4) to better ones with higher efficiency. Cheaper, passively cooled, still enough power. Adding such a connector and a 12V regulation to the motherboard might not be too expensive?

2.) Onboard controller: I like those embedded ARM/MIPS-CPUs but I think they shouldn't be on the motherboard if they are not necessary. If the USB-bandwith is enough to supply all those FPGA with data, then leave the high-complexity tasks to the Host. And it really doesn't have to be a PC. I have a Sheeva Plug which takes less than 5W and has a 1GHz ARM with 512MB RAM. More than enough I think.

3.)

Since you can't power the whole FPGA-array through one USB-port and we will need an external power-supply anyway, I'd suggest to also power the FT2232 from the power supply. Why? There are potential host-devices that can't even handle the minimum 100mA. For example an Android phone with a modified kernel to support host-mode ... would also be a very power-efficient host.

Sorry if some things I said didn't make any sense, I'm really a software guy.

I would still stick with Molex 8981.  People using bench supplies can easily stick a Molex on their wiring, and people using ATX are already set.  If you use a different connector, then everyone has to make custom cables.

And I would go with a 240 pin DIMM socket for the cards.  Figure on about a third of the pins being grounds.  Add a few for SPD, a few more for JTAG, a few more for whatever serial bus you come up with.  Still leaves you with plenty of open pins for power, even if you go with 8 or 10 pins for each likely voltage level.

...which are all way more expensive than a small ARM-based solution for the backplane, which removes the need for a PC.

My suggestion would be the following:
- One of those FTDIs on every FPGA/ASIC board, with access to the JTAG scan chain, the IRQ signal, and possibly an the I2C bus
- A USB mini B connector on each board, mounted in a way that doesn't allow you to plug a cable when the board is sitting on a backplane
- Each board has a JTAG scan chain, an I2C bus and the USB interface to the FTDI connected to the backplane connector. This increases flexibility and you get it almost for free.
- Future backplane implementations may either use USB or JTAG/I2C.
- FPGA designs may choose to do everything via JTAG, or go for I2C for post-boot communication.
- I'd still be strongly in favor of a small I2C EEPROM on each card. The cost for this is neglegible.

I'm not really concerned about the backplane/card implementation. This can easily be improved at a later point. The interface between those is what can't easily be fixed in the future, so this should be designed for maximum flexibility.

@ Olaf.Mandel

I second your communication approach, but would like to wait for some other comments before finalising this design.

Also i would agree to keep both solutions as similar as possible.

As this board should be abled to be operated by a wide range of people i would prefere an ATX power supply over a dedicated one as i see it to be easier to handle. 

How to delete a message? I accidentally created a copy of my previous post and cannot get rid of this text.

I just had a look at different ATX power supplies: while they had 9A of +12V their 5V rail was above 20A! So maybe I picked the wrong voltage?

The reason why I picked one voltage at all instead of allowing the use of both 5V and 12V: some may not want to power their FPGAs from an ATX power supply but use something dedicated. There, a single supply is strongly preferrable.

On that note: these people may not like the Molex 8981 as much. What about the Tyco 284513-2 from the Buchanan series:



It's basically a question what there are more of: ATX power supply users or dedicated supply users?

Seconded. And one PC can run a whole lot of other cards (standalone or backplane). And having an ARM on the backplane but nothing on the standalone cards will split the development effort down the middle.

A cheap computer can be a netbook. It costs only little more than a custom solution and has a lot of added value (and is faster to get). If you want something embedded: there are already embedded solutions that contain ethernet and USB host on a few square centimetres, e.g. FOX Board G20, different plug computers, ...