Topic: X6500 Custom FPGA Miner - page 27. (Read 220050 times)

no, but i didn't make e long term test

on my wrt54gl i wasn't able to install python because of diskspace, maybe it would be possible to mount an usb stick
on my buffalo wrz-hp-ag300h-eu it runs fine, cpu usage was about 40%

Yeah the hardware is pretty sick for the price (the concept is also very very cool) but yeah I'll probably not be able to get my hands on of these ;(

And yes! I would totally love to see how you got MPBM working on OpenWRT!!! I can always get my hands on a TP-Link 1043nd and then run my X6500 off the USB port! Quick question: Have you noticed any mh/s difference between running an X6500 off an ARM platform vs a regular PC?


Cheers,

nbtcminer

USB can't easily be daisy chained (would require a 2-port USB hub on each board, and you may only chain something like 6 hubs total (including hubs that are possibly integrated in the PC).

Hm, very interesting. The results I got from MPBM show a slight bias towards FPGA0, fizzisist's x6500-miner results show a bias towards FPGA1. Might after all be related to how x6500-miner deals with the FPGA alternatingly while MPBM manages them in parallel or something like that.
An aluminum heatsink certainly can't keep up with a copper one, but I don't think it really matters much whether the heatsink is like 2°C cooler if the temperature difference between the die and heatsink is like 20°C because of the FPGA's nasty package. So yes, it might improve things, but it might just not be worth it, some airflow around the ground plane of the PCB might have a bigger effect. (All guesstimations, feel free to prove me wrong )

The way the X6500 communicates with the controlling PC works well for regular bitcoin mining, but introduces some latency (a couple of milliseconds) that hurt P2Pool a lot at its very high block rate. So you should expect degraded efficiency when using an X6500 with P2Pool. This might be addressed in a future board revision. (The miner software might also have an impact, but it isn't all that big.)

Should work with MPBM and possibly also with X6500-miner if libd2xx is available for ARM platforms and the board can handle it performance-wise.
I'd say that with MPBM up to 4 boards should work well, if you have more you might run into performance issues.

i don't think you have one of those
first round of those to public will be out of stock in hours

but i hope it supports opencl

if you want to use an ARM-CPU you can do this, i managed to get MPBM running on openwrt

@All / Fizzisit:

Could we use something like this to host a bunch of X6500?

http://www.raspberrypi.org/

the x6500 miner software seems to have some trouble with p2pool, after some hours of work with avarage 5% rejected 0 stale shares the miner just pulls blocks but doesnt provide any shares, have to restart the software to make it work again. does someone have this aswell with p2pool ?

@TheSeven:

I'm glad you noticed the FPGA 0 / 1 invalid thing too. I didn't think much of it at first but I think it might be more hardware related then anything (I experienced the same issue of FPGA 1 generating more invalids then FPGA 0, which usually had 0% invalids). However having said that I'm at about 0.2% invalids on FPGA 1 and 0 % on FPGA 0 after 4 days of 24/7 mining so I'm not really too concerned about it at the moment.

BTW with regards to the plastic surface on the FPGA; do you think the anodized aluminum will do as good of a job as a full copper heatsink? I think a copper heatsink + good thermal compound (and/or epoxy) + active cooling should do the trick of eliminating overheating as a source of mh/s degradation.

Cheers,
nbtcminer

Wow, those towers are lol.

Hi fizzisist,

could a fully passive cooler like this, one per fpga, be enough?

http://www.frozencpu.com/products/12775/vid-165/Deep_Cool_Nbridge_8_Northbridge_Chipset_Cooler.html?tl=g40c16s500&id=sLNsfH2v

spiccioli

How difficult would be to daisy chain the USB, and power?

Heh you should have come up with all these suggestions a month ago when the new board was being designed.   I for one much prefer power one side, usb the other. The rev3 board, for me is completely optimal.

Is there a way to put a male USB connector to the board next to the female connector, so that multiple boards could be daisy chained together?

all connectors on the same side would make my setup much easier

but as soon is i have enough btc i will buy a seconed board anyway

Standardized chipset mounting holes will allow for water cooling too.
Might make it easier to develop some sort of backplane too, so you could just plug the board into a "dock."

I think I'm going to have to save up my BTC to get one of these.

I'm not sure what you saw, but I have some hard numbers to look at. I just took the trouble to compile all of the test results I have for the boards. I saved the "Run Summary" that prints out when you quit x6500-miner for every board that shipped out of here. Mind you, some of these tests were done before my cooling setup was really nailed down. Most tests were done at 180 MHz. Some runs also were noted as "bad" and I changed something, like reseating a heatsink or gluing a heatsink on, then retested. My notes show all of this. If we go too far back, x6500-miner wasn't recording the invalids rate, so this only includes about 2/3 of the boards.

Here are the raw results for all tests:

FPGA 0:
  Accepted + Rejected: 171902 shares
  Invalid: 1322
  -> 0.77% invalids
FPGA 1:
  Accepted + Rejected: 171557
  Invalid: 2975
  -> 1.73% invalids

As you can see, on average, FPGA 1 is actually worse. This makes sense to me, because I've always had FPGA 0 closest to the fan (air flowing left-to-right).

If I go through and remove those ones that I noted were not good and later improved them, taking only the last test with that board, I get the following:

FPGA 0:
  Accepted + Rejected: 149807 shares
  Invalid: 1154 -> 0.77%
FPGA 1:
  Accepted + Rejected: 150775
  Invalid: 1620 -> 1.07%

I noticed something similar as well while testing MPBM on the cluster. On most boards, there were no invalids at all while the ambient temperature was rather low, except for two boards. On one of them, FPGA0 got like 2% stales while FPGA1 didn't have any, and on the other one it was like 0.5% on FPGA0 and zero on FPGA1. As ambient temperatures increased all boards produced invalids at some point, but this seemed to be biased towards the FPGA0 side as well. FPGA0 on that "misbehaving" board was at like 5% invalid at that point, most other FPGAs still <1%.
No idea what exactly is going on here, but it indeed seems like there's something different between the two FPGAs, either with power supply, cooling or signal routing.
I think I can definitely rule out that this is a miner issue, as both FPGAs use exactly the same code base in MPBM. However it might still be interesting whether this happens with x6500miner as well, as this might be somehow related to communication behavior, so different software might be affected to different degrees.


Oh noes! The current heatsink looks way better
But seriously, I don't think using better heatsinks will help much here. The real bottleneck is the thermal bridge between the FPGA's die and the heatsink, which is obstructed by that nasty plastic package.

@Fizzisist:

This is kind of a minor thing, but would it be possible to move the USB connector to the same side as the barrel / molex connector? That would make it much easier for enclosure /air flow designs.

Spent a little time drawing some components:



I'll do a little more work on this, then post the Sketchup file so that anyone can play around with it. I imagine it would be helpful when designing an enclosure or heatsink or whatever.

Yep, it says on the board, next to the FPGAs. I really don't know what's going on here. It seems to me that your fan setup is good, and you say the heatsinks are on good. Airflow on the bottom of the board would be good, but I think you have plenty of air. Does the PCB itself feel warm? That would be an indication that either your heatsink isn't on right (the heat is going into the PCB instead of the heatsink) or you should think about getting airflow on the PCB itself. If that feels cool, too, then something else is going on. Do you get invalids when running at 180 MHz?

there is a text printed on the board