FPGA development board "Icarus" - DisContinued/ important announcement

[hardpick]

"please observe the LEDs to clear the FPGA's actual operation state. if the IDLE indicate LED (LED1 and LED5) light up:

disable the FIFO will help, change some other version driver might help, change to Linux probably solve the problem permanently.

if not, maybe it's miner's bug effects.

now i'm testing a 40 boards rig with 15.2G/s speed, very stable. no error in at least 24 hours."



Hi

is it possible to advise what miner software you are using ?

and if you are using windows or linix

thanks

also my new icacus board are missing leds sd/rx is that normal for 3rd batch
(i guess they are only for debug)

[kano]

Not answering for ngzhang ... but

I'm using cgminer (the version from my git https://github.com/kanoi/cgminer ) and linux (xubuntu 11.04 as per my sig installed to a HDD)

I can provide a linux executable if you are unable to compile it, that I use on my rig that is 2 x Icarus and 2 x GPUs if you want
(my git says 2.3.1 as the version but it's got a few updates pending on the original ckolivas git of 2.3.1)

[antirack]

Thanks for all the Watt/Ampere information. I am going to buy a ATX power supply. I've seen lots of them in local PC shops with 350 to 1200 Watt (I guess if I look at server shops I can also find higher ones). The question is what makes most sense for my little mining farm with future upgrades.

I don't know how many Molex cables the ATX power supply usually has.  Wikipedia says the Molex plug is limited to 11A/132 Watt. So 6 Icarus per Molex. If this is the case, then it doesn't make sense to buy a higher wattage ATX power supply with more watt than I have Molex connectors.

Can anyone confirm this?

ZTEX wrote somewhere in his thread that you can power 30 of his boards with a ATX PSU. He also put up this guide how to build a PSU for a cluster. He also said this in the guide:

GND wires from PSU (black) should not power more than 3 FPGA Boards and +12V wires (yellow) should not power more than 5 FPGA Boards. (GND wires are more critical in order to reduce current flows through USB cable)

by my experience, with low wire quality in usual, i suggest to power NO MORE than 4 icarus with one D type plug (molex plug). even 3 is safer.

I just want some clarification for us guys working with ATX power supplies. On the PSU I have I counted 6 +12V wires (yellow) in total. The PSU has 3 molex plugs in series with only 1 yellow (+12V) wire (and two black GND wires). So to be on the safe side, you are suggesting no more than 3 Icarus per yellow +12V wire?

[ngzhang]

now i'm testing a 40 boards rig with 15.2G/s speed, very stable. no error in at least 24 hours.

master or testing branch?

by the testing branch. i work with [7] for about a week to debug it.

Using MBPM or cgminer?
I will be finished the Linux host tonight. So will let you know if that helps.

Thanks!

i use MPBM testing branch, but i didn't try P2P pool yet.

"please observe the LEDs to clear the FPGA's actual operation state. if the IDLE indicate LED (LED1 and LED5) light up:

disable the FIFO will help, change some other version driver might help, change to Linux probably solve the problem permanently.

if not, maybe it's miner's bug effects.

now i'm testing a 40 boards rig with 15.2G/s speed, very stable. no error in at least 24 hours."



Hi

is it possible to advise what miner software you are using ?

and if you are using windows or linix

thanks

also my new icacus board are missing leds sd/rx is that normal for 3rd batch
(i guess they are only for debug)

hi, the RXD/ TXD LED is not installed because they are hardly resolved by human eye (continues light up). so i mapping the logic to LED 2/3/6/7.

that 8xLEDs can completely display the FPGA work staus.

I just want some clarification for us guys working with ATX power supplies. On the PSU I have I counted 6 +12V wires (yellow) in total. The PSU has 3 molex plugs in series with only 1 yellow (+12V) wire (and two black GND wires). So to be on the safe side, you are suggesting no more than 3 Icarus per yellow +12V wire?

i will give you some picture later.


here is a screen shot, i start a new tower yesterday evening :

[TheSeven]

So there are no more deadlocks? That's nice to hear

[Glasswalker]

Here are some more pics for you. The build is taking a little longer than I wanted, I hope to finish and have it fully operational tomorrow.

Ran into a few issues, one of which was the power supply failing. Had to go pick up a different PSU (went with a slighly more expensive model to give us some room to grow)
Also the above mentioned issue with the driver potentially... Linux is expected to resolve the issue, so I moved onto the build and will finish testing on the actual host machine. May have also been related to the psu failing (was giving voltage dropouts on the 12V rail under load)

Anyway, things are rolling along now, so...

Here are some pics of the build along the way (a little more eye candy for you to look at):

A block of 4 being tested together:


My RepRap producing a prototype of the brackets to hold the cards:


The RepRap producing 4 brackets at once:


Getting ready to dry fit the parts in the wooden rackmount box:


Fans mounted, and cards on brackets (still dry fitting):


Close up of the rig so far:


Still left to do:
- Solder 2 more power leads (bundles of several power leads for several cards, back to either a molex, or a PCIE connector to distribute load on the ATX supply properly)
- Adjust the standoffs on the blocks (in this config with the dry fit there isn't enough room for 8 wide without causing airflow problems, if I reverse the standoffs there will be)
- Screw everything down
- Run all the wires and clean it all up
- Install the host OS (Linux) on the computer
- Mount it all up in the rack, and get mining!

I hope/expect to get done tomorrow. Worst case Tuesday if something holds me up tomorrow.

[Doff]

Pretty sweet setup. What I want to know is where you got those extra long standoffs, did you make them yourself?

Here are some more pics for you. The build is taking a little longer than I wanted, I hope to finish and have it fully operational tomorrow.

Ran into a few issues, one of which was the power supply failing. Had to go pick up a different PSU (went with a slighly more expensive model to give us some room to grow)
Also the above mentioned issue with the driver potentially... Linux is expected to resolve the issue, so I moved onto the build and will finish testing on the actual host machine. May have also been related to the psu failing (was giving voltage dropouts on the 12V rail under load)

Anyway, things are rolling along now, so...

Here are some pics of the build along the way (a little more eye candy for you to look at):

A block of 4 being tested together:


My RepRap producing a prototype of the brackets to hold the cards:

The RepRap producing 4 brackets at once:


Getting ready to dry fit the parts in the wooden rackmount box:

Fans mounted, and cards on brackets (still dry fitting):

Close up of the rig so far:

Still left to do:
- Solder 2 more power leads (bundles of several power leads for several cards, back to either a molex, or a PCIE connector to distribute load on the ATX supply properly)
- Adjust the standoffs on the blocks (in this config with the dry fit there isn't enough room for 8 wide without causing airflow problems, if I reverse the standoffs there will be)
- Screw everything down
- Run all the wires and clean it all up
- Install the host OS (Linux) on the computer
- Mount it all up in the rack, and get mining!

I hope/expect to get done tomorrow. Worst case Tuesday if something holds me up tomorrow.

[Doff]

Oh and btw Glasswalker, that Reprap is so cool, thanks for introducing me to new cool things.

[ngzhang]

nice work...
looks like i must post up my 40 cards setup tonight.

ADD:

 

above pictures declared how to setup a 40 boards rig in 30 minutes.



above pictures declared how to use a standard ATX power. notice that the low- quality plugs and wires are burn to black color. it's very dangerous. maybe cause fire.

[Dhomochevsky]

Sent e-mail, maybe I can grab one more from what's left of this batch.

Also, eagerly awaiting more info on the 4th batch.

[ngzhang]

Sent e-mail, maybe I can grab one more from what's left of this batch.

Also, eagerly awaiting more info on the 4th batch.

 
not received any mails .

[ngzhang]

 

the hand made power transfer cable burn up tonight. Almost a risk of fire. 

i must find some new cables...

[Turbor]

Take care. We still need you

[antirack]

the hand made power transfer cable burn up tonight. Almost a risk of fire. 

i must find some new cables...

Can you let us know a little more info? I am trying to avoid exactly that from happening. If it happens to you as engineer, I don't want to know what _can_ happen to my ghetto style ATX PSU and cables (so far no problem with my new ATX PSU though).

[Glasswalker]

The important thing will be to ensure you're not overloading any wires. The setup that ngzhang posted looked a little "ghetto" lol (I suspect he was rushing and just slapped it together, as I've done in the past myself once or twice ) It looked to be hand twisted wire with tape sealing it up (no offence ngzhang, and if I'm incorrect, let me know). This is prone to high resistance at the twisted connection, and will result in a lot of heat generated there (and risk of fire).

Your best bet is to make sure not to exceed the rating (4 pin molex connectors are rated for 11A at 12V or 132W) The Icarus draws about 20W. So that would mean if everything is perfect, you could run 6x Icarus boards. But all is rarely perfect. And Molex connectors are renowned for arcing and other problems (which can generate excess heat). Best to keep it to no more than 4. Also ensure a clean connection (well soldered joints, not just crimped) and so on. I'm making several "Molex to Barrel DC Jack" connectors, which I'm building to a fairly high standard. To allow me to run up to 4x Icarus per Molex. On that 700W supply I'll be using 2 Molex, (8 Icarus total). And I am building adapters from the PCI-E connectors in the same fashion (rated 150W each and designed to deliver high DC current) to power 8 Icarus each. That will give me a total of 24x Icarus boards off that PSU. (24x 20W = 480W roughly, and that supply sends nearly 600W to the 12V rails, so it should be able to handle that just fine)

Anyway, basically ensure clean connections, good joints, and always give yourself plenty of headroom from the rated spec on all your components (keep an eye on the weakest link).

Once you have the rig built, I would check all power lines for heat. If you have wires running hot to the touch that's probably bad.

[rjk]

The important thing will be to ensure you're not overloading any wires. The setup that ngzhang posted looked a little "ghetto" lol (I suspect he was rushing and just slapped it together, as I've done in the past myself once or twice ) It looked to be hand twisted wire with tape sealing it up (no offence ngzhang, and if I'm incorrect, let me know). This is prone to high resistance at the twisted connection, and will result in a lot of heat generated there (and risk of fire).

Your best bet is to make sure not to exceed the rating (4 pin molex connectors are rated for 11A at 12V or 132W) The Icarus draws about 20W. So that would mean if everything is perfect, you could run 6x Icarus boards. But all is rarely perfect. And Molex connectors are renowned for arcing and other problems (which can generate excess heat). Best to keep it to no more than 4. Also ensure a clean connection (well soldered joints, not just crimped) and so on. I'm making several "Molex to Barrel DC Jack" connectors, which I'm building to a fairly high standard. To allow me to run up to 4x Icarus per Molex. On that 700W supply I'll be using 2 Molex, (8 Icarus total). And I am building adapters from the PCI-E connectors in the same fashion (rated 150W each and designed to deliver high DC current) to power 8 Icarus each. That will give me a total of 24x Icarus boards off that PSU. (24x 20W = 480W roughly, and that supply sends nearly 600W to the 12V rails, so it should be able to handle that just fine)

Anyway, basically ensure clean connections, good joints, and always give yourself plenty of headroom from the rated spec on all your components (keep an eye on the weakest link).

Once you have the rig built, I would check all power lines for heat. If you have wires running hot to the touch that's probably bad.

It looks like the actual Molex connectors are burnt. This is likely due to inexpensive underrated pins and tiny gauge wire.

[Glasswalker]

Very true, my previous post is working under the assumption that the manufacturer didn't cheap out and get sub-standard pins/housing. Or under-rate the wire gauge (which admittedly happens more often than not). And as I mentioned, Molex connectors are well known for causing arcing and such due to a poor design. The higher end ones aren't as bad, and that's also why it's important to give yourself a safety margin, and to ensure all connections are good.

Edit: You're right, I somehow missed that in the photo.

That looks like damage to the plug due to arcing (one of the pins not seated perfectly well and either causing high resistance, or actual electrical arcing, generating tons of excess heat).

[ngzhang]

a 18AWG standard wire can pass up to 10A current, and 8A is enough for a safety margin. so i setup no more than 4 boards to one Molex plug.
but i doubt the power transfer cable's quality, so i use them in parallel, 1 wire 1 board, 2 board 1 connector.
and i really soldered them very well.
but they still burn up in a few days...

just because i buy them from 0.5~1 USD/ ea.  

last night after the fire, i find some about 5 USD/ ea, i think this time they will be ok.  

[chungenhung]

Which prolific PL2303 driver do I install? v1.5 or 1.52?
Could someone point me to an example using Icarus with cgminer?
Thanks

[ngzhang]

Which prolific PL2303 driver do I install? v1.5 or 1.52?
Could someone point me to an example using Icarus with cgminer?
Thanks

i use 1.5 now, but please feel free to try any other version, some of them may more stable on your machine.