Topic: X6500 Custom FPGA Miner - page 24. (Read 219954 times)

Yes, exactly. Thanks for adding this clarification.

It wasn't to "make it work".  The Rev. 2 boards pulled power from the 5v rail as designed.  The molex mod was to switch the power feed from the 5v rail to the 12v rail.  This was done because most of the modern PSUs can supply far more power on the 12v rail than the 5v rail.  They will work on either the 5v or 12v supply.

Ok that clears it up,  I was reading your first post and it was talking about modifying a molex connector to make it work.  I have a bunch psu's so power won't be a problem.  Thanks.

Holy crap! $17 for a connector with a whole 2 pins connected? People pay for that!?
Yeah, think I'll make do with a paperclip and some electrical tape.
Sheesh.

No modification needed to the connector itself, but the PSU needs to have the ENABLE pin pulled to ground. This is what a motherboard will do, to turn on the PSU when it wants it on (when you press the power switch) but if you're using the PSU standalone, you need to do that yourself. This is what the dummy plug from Cablesaurus does for you. The molex connector itself can be used as is.

By the way, the original (rev 2) X6500 used the 5V rail from the Molex connector, but the new X6500s (rev 3) take power from the 12V connector.

Can I just plug the molex adapter from the psu in or does it need to be modified.

This one works well: http://cablesaurus.com/index.php?route=product/product&product_id=50

To be clear, though, you can use either a wall-wart adapter or a 4-pin Molex connector from a standard ATX power supply. The wall-wart should work well for running a single board, but I recommend using a standard PSU if you have more than one because they are so much more efficient. You can run a lot of X6500s on this one, for example, with great efficiency: http://www.newegg.com/Product/Product.aspx?Item=N82E16817371033

If the PSU isn't also powering a computer, you'll need to modify it or pick up one of the dummy connectors from Cablesaurus: http://cablesaurus.com/index.php?route=product/product&product_id=45

In the description it says to use a wal-wart power adapter.  Does anyone have a direct link to the one that will work.  Or are these molex connectors on Cablesaurus ready to go without modification.
 8-Slot 4-Pin Molex Splitter http://cablesaurus.com/index.php?route=product/product&path=10&product_id=51.

BTW, I'm runny my ARM stuff on a Pandaboard (Ubuntu Oneiric) as well.

@ fizzisist :

correction:  180Mhz not 190 i mixed that up with the icarus.

I may conduct measurements in linearity with different bitstreams.
Maybe i got a very good sample of the chip.

Im just very busy at the moment(office)

Thanks, O_Shovah, and amazing work on the ARM stuff! Definitely start up a thread on it, there is a huge amount of interest, believe me.

My only hesitation with your power consumption data is that it was so damn good! You got under 15 W, and that's way less than any of the other FPGAs have been claiming. I'm not saying I doubt you, but I'm hesitant to start advertising such great efficiency until I have some independent confirmation.

As soon as I get the new boards, I'm planning to do some careful measurements as a function of clock speed for: power consumption, temperature, and invalids. What clock speed were you using when you measured the power? You said above 190 MHz, but we had only 180 MHz and 200 MHz at that time, I think.

Thanks!

Hello fizzisist

I am sorry for not being around for so long,but i was busy hitting on my ARM board.

Finally i manged to get etherything to run. "gumstix" ARM board + one x6500 + one Icarus  with bitcoin-qt ( it finally compiled )
p2pool and the Modular python bitcoin miner by TheSeven.

Hopefully i will find the time to open a thread for posting a "HowTo" for this.
I also wil get a panda board this week so i wil add a system for it soon.
And as i am curently running with 600Mhz i am positive i will also get it to run on rasberry pi as soon as i get my hands on one.

So far from my side.

Regarding the thermal images.

Of course you may use them.
I propose to use the same conditions as with the icarus people.
eg see this http://en.qi-hardware.com/wiki/Icarus

So that would be:

Initial author:
Jens Schmelkus

Licensing:
Creative Commons Attribution Share Alike (CC BY-SA) 3.0

Setup:
Thermographic pictures of the "x6500" (rev1.0) board taken from top, 45° top-down and bottom. Full speed operation(190Mhz).Room temperature 21°C.
Cooled by 80mm fan @ 11V. No additional air movement. Taken with NEC IR Thermoshot F30.


I also recommend you to use the power consumption data i measured. I consider them quite accurate.


Also if you or anybody else would like to have some thermal imaging made. Just contact me

Manufacture on the next batch of boards will finish in the next few days. After that, it'll be about a week of testing and packing before Cablesaurus has any in stock again. I'll be posting updates here after the boards arrive (and pics, of course!) and after some initial tests are done.

Aiming for the first wave to be at Cablesaurus by the end of next week (3/16). We'll do our best to hit that goal!

Does anyone know when the next shipment will be available.

On an unrelated note, I ran the 180mhz bitstream for about 8 hours with no invalids so this afternoon I switched to the 200mhz bitstream.  FWIW these are unmodified and are still using the thermal adhesive tape for the heatsinks that they were shipped with.  The room is about 60F (16C) which is probably helping a lot.

I've had p2pool running for the rest of my farm so I'd probably need to isolate the FPGAs to their own p2pool instance to get clean data from the p2pool side.  The rejected numbers I saw were just in MPBM - when total accepted shares got to 80 there were 40 rejected.  This is with the FPGAs on the same machine as p2pool / localhost connection.

50% is a damn lot. Both the board interface and MPBM itself currently aren't really optimized for P2Pool, but I wouldn't expect that there are more than like 30% stales. I hope we get down to those usual 10% one day...
How did you determine the number of stales? How many of them are dead on arrival? Can you paste the p2pool output, ideally an excerpt containing some dead on arrival stales, to get an idea what the timing of those looks like?

fizzisist - I don't mind if you use the pic of my setup but as mentioned I don't plan on running them like this for long and have plans for a much improved layout.  I have everything running on linux and my only problem was that I had missed the step to run "rmmod ftdi_sio" prior to loading the bitstream because I was too eager to get started!  PEBKAC

TheSeven - I'll definitely be tinkering with MPBM and p2pool once I have some time.  Right now I'm running MBPM on the same machine as my p2pool+bitcoind instance and each x6500 is connected to native USB port, so that should rule out most latency issues.

edit - I was able to set up p2pool using the defaults but was seeing 50% of shares rejected.  After reading some comments on the MPBM thread it seems this may be expected behavior.

Just uncomment the P2Pool example at the bottom of the default config file.
This is in a different blockchain definition, in order to decouple its long polls from the other pools.

I'd be very interested to see more MPBM P2Pool results. My own attempts didn't really work well, but the main reason for that was probably bitcoind running on a screwed up box with too high latencies. (I didn't have space to store the whole bitcoind blockchain locally on the little ARM board that was running MPBM.)

Welcome!

That's a sweet setup! I can't tell you how much I love seeing these pics. And, I'm really happy to see you starting with 166 MHz. If anyone is setting up their X6500s for the first time, I strongly recommend starting with a very low clock speed just to see that it works and your cooling is adequate. Then, slowly ramp up the clock until you find a good one (less than 1% invalids ideally). Of course, this will be a lot easier when our new firmware is ready, and you can easily change the clock in 2 MHz steps without reprogramming. On top of that, the software will automatically reduce the clock if the FPGA gets a lot of invalids or the temp gets too high (on boards with temp sensors).

On another note, we're working on a new website, and one part of it could be photos of different people's setups. Would you guys be interested in that / willing to share your photos and maybe a few details on what you used? This could be nothing more than what you have already posted about it, or it could be a lengthy write-up if you so desired.

Skimming through the thread, I see photos from:
Qoheleth
O_Shovah (sweet thermal pics)
bitcowok (and here)
99Percent
mike2kt
SamHa1n
and now coretechs

Would you folks be ok with your pics posted on fpgamining.com? If so, and you want to have a little blurb or some other info about it you would like to share, please let me know!

Thank you!