Topic: Experimenting with Jalapeno firmware... - page 4. (Read 62616 times)

I am happy to give it a try, the JTAG programmer I ordered from eBay hasn't turned up yet, and I don't even know if it's compatible with the ATMEL 32bit chips. So the Arduino might be a safer bet.

Okay, so my Jally is now happily crunching along with the BitForce_SC_1.2.5ck.elf firmware at 8.5-8.9Gh/s, 39C, and about 1% HW errors. Same setup - just converted the firmware and fed it to UrJTAG in the same way as I did last time with the botched-compile firmware. Once it got to "Verifying" (and I slept through the programming, so it was a good halfway through verifying), I canceled it and fired up the board. Took a few seconds, and it finally came-to with 2 LEDs and ready to roll.

So, in the interest of helping the interwebnets, I'm posting a thread with the info for doing it with an Arduino.

Anything with 5 I/O pins would work - the code is entirely made of "bit-banging", which is why it took 2.5 hours (YES! With tightly optimized code!) to program. It's absolutely a "desperate" measure, but from what I've seen, UrJTAG is extraordinarily careful with the commands it performs, checking and double-checking everything to account for "dirty" adapters like this.

Just a whole hell of a lot of modifications to the code, though, which made it program in less than 5 hours as it was originally running. UrJTAG is from LukeJr's git clone, patched with Arduiggler-urjtag's modified code to support the interface. Arduiggler itself was partially re-written to implement bit-banging in a more efficient way. So it's not exactly stock "go grab these things and run it like this"

Little status post here for those curious about the fringe of hacking this thing... last night I spent a few hours compiling/assembling (hardware/software), adjusting, tweaking, and actually successfully programming my Jalapeno using only an Arduino. I think this effectively lifts all limitations on what JTAG programmer you can use

I compiled 1.2.9 with some comment adjustments in the config code. As others have noted, the code seems to be haphazardly thrown together with huge swaths of code commented out (like, in my case, the routines to test chips on startup - commented out by default). It compiled in AVR Studio, then I took the HEX file and made a BIN out of it - then for the next 2.5 hours, the Arduino and the Jalapeno had a conversation with the help of UrJTAG and Ubuntu. Finally, when it moved on to verifying (and a few pages passed by without errors), I killed the program and rebooted the Jally.

It worked! Sort of. It boots up, lights up the 2 LEDs for the two chips, and responds to ID commands (ZCX, etc), indicating the new firmware is there. Chips now running at 300MHz, which is far more than I expect it to run at (but with poor commenting in the code, I couldn't tell what I was setting it to). But that's all it will do now. The moment I fire up a miner, the Jally knocks out one of the chip LEDs and goes into a hard-lock until I power-cycle it.

Not bad for the first guy flashing a Jally using only an Arduino and about 30% custom code made up on the spot (on both UrJTAG's source code and in Arduiggler's code). When I get back to it today I'll give it another go.

Damn shame the "set frequency factor" (ZVX + 0x04 + 4-byte code) command interpreter function is broken - always saying "INVALID DATA" no matter what it's given, probably due to a broken wait/timeout loop mechanism (killing the command entry too early like it does with Z*X commands if not copied/pasted all at once). Probably could use that command to get the firmware working as it is now

Sounds good. Looking through the code yesterday I can see that compiling JP or LS has some impact. I'll try it again once the chips come in (Thursday or Friday is my guess). Power supply may take a bit longer; so I'll just do two chips and run on the crap power supply for a few days.

C

I don't know. I never complied the firmware without any changes. But, you should try setting it to little single at least.

If you still have trouble, I can PM you a link to the firmware I am using.

Interesting. I remember that; was it necessary to get it to not flash?

Maybe the issue is little_single; it's looking for a pair of chained boards. Hm.

I told someone the changes I made to stock 1.2.9 a while back...

So I'll document the saga of trying to boost this Jally with more chips. First, code issues.

I'm still waiting on the chips to arrive, but since I have been running CK's 1.25 elf (which gives me .5% errors, 40c temps with the case closed, and a nice little 7.4gh speed on my crap chips) I will need to upgrade to 1.2.9 to use the new chips with the old ones. So I figured I would test upgrade before putting the chips on.

Glad I did.

First I flashed Tarkin's mods. Looks like he is using the 1.2.9 base code, with serious tweaks. Oddly enough I can't recompile it, AVR blows up looking for a /dev hard disk on my Windows box. Might be a code dependency in there, I'll look. But the supplied elf does work and loads up.

8.1gh speed, but with 30% errors as reported by BFG. WOW. More importantly the long term stats (third of the three percentages) is only 6.8%, I think this means I am hashing but am getting *less* throughput because once again my chips suck. I should just float them off and run with all new chips. Maybe when they hit a buck a chip or something.

So I went with the stock BFL 1.2.9. Compiled it on Atmel, no errors, hundreds of warnings and all that. Flashed it to the jally, hooked it up, no hashing. Front light would come on, then after a few seconds flash fast.

Oh that did a number on my heart. Tried recompiling and loading, still fast flash. So I reloaded ck's 1.2.5 and all is well again.

Questions:

Tarkin: Did you code a hard dependency in your changes? I'd like to remove it if so, seems to be blowing up Windows 7 and Atmel.

CK: Have you done whatever you did for 1.2.9? I love your changes the most, sending you some bitpennies in thanks.

Anyone: Why the hell is stock 1.2.9 blowing up?

Worst case I could run Tarkin's elf, but I am really glad I did this before soldering on new chips. Moral is to always do things one step at a time :-)

C

Solved it. I changed the cables, so it was probably a faulty one that generates the above error message. I also used this cabling/power up sequence:
Connect all pin cables unpowered, power up jalapeno (no usb), then power up raspberry pi. -I noticed that if you power up jalapeno after raspberry pi the HDMI signal blanks for a moment-

You've pretty much maxxed that one out... You might get more by disabling some of the more rigorous checking done in the firmware on bootup, but chances are if the engine is defective you'll just end up with a higher HW error count.

GetInfo=DEVICE: BitFORCE SC0x0a
FIRMWARE: 1.2.90x0aIAR
Executed: NO0x0a
CHIP PARALLELIZATION: YES @ 20x0aQUEUE
DEPTH:400x0a
PROCESSOR 3: 13 engines @ 277 MHz -- MAP: FF8F0x0a
PROCESSOR 7: 15 engines @ 250 MHz -- MAP: 7FFF0x0a
THEORETICAL MAX: 7351 MH/s0x0a
ENGINES: 280x0a
FREQUENCY: 274 MHz0x0a

I can`t go more then 7.2 Ghash.
Please advise how to enable engines on first processor or freq or second.

Thank you!

be careful i burned my Raspberry when mixed up pins look at this pic
http://www.cl.cam.ac.uk/projects/raspberrypi/tutorials/turing-machine/two.html

http://randomcontent.wolfnexus.net/RandomSite/reflashing-a-butterfly-labs-jalapeno-with-only-a-raspberry-pi/

I followed Red_Wolf_2's instructions and when I run jtag>detect I receive "warning: TDO seems to be stuck at 0error: not found: queue is empty"

I am connecting the pins based on the circuit board markings meaning:
 
pin 1 on jalapeno is the closest pin to the nearby corner edge and pin 9 is the last pin in the same 5pin row

pin 2 on the raspberry pi is the closest to the nearby corner edge and pin 26 is the last pin in the same 13pin row

Am I doing something wrong with the pin numbering or cable connections? I am using old internal cd rom to sound card 3/4 pin cables
Could it be a software problem? I think my jalapeno is fairly typical with the cut heatsink and firmware 1.0.0

I used noobs_v1_3_2.zip installation and chose raspbian (according to the release notes date, it is the 25-09-2013)
Installed srecord, git, bison, flex, autoconf, libtool, gettext, automake and python-dev using apt-get
Am I missing a package?
I used su (and even tried sudo jtag) but the same jtag>detect error message appears

BTW run sudo apt-get upate BEFORE installing the python-dev package in order to avoid python.h file missing during make

 
Thank you in advance

Can I use AVR Studio 4.19 to flash the Jalapeno? That's what's recommended to use with my 3rd. party ICE-JTAG adapter I got from eBay.

Just line up the dot on the top of the chip with the silkscreened dot on the PCB. I applied liquid flux (Kester 951) to the board first, then I set the chip on the board and began applying heat.

You inspire me Dana. Well you and greed.

So how did you know which way was "front" with the new chips? Did you use flux to hold them in place before using the air tool?

Just mentally getting ready to get 2 then 4 of these on the board.

C

Yep. I had the power go out at my house yesterday for 10 hours and when I got power back I couldn't get my jalapeno to start up. I guess I pushed that power brick too far. I wired up a barrel jack to my ATX power supply I already had running for my other miners and it's running like a champ again.

And in the "in for a penny" project, I'm just going to buy 4 chips and boost this thing to six chips. Three and three, should be stable on the heat sink.

Why not?

Big problem will be power, that "7a" supply shipped with these things is a joke. So I bought an ATX 400 watt power supply which should be able to put out 17a at 12 volts, that might do it. I'm guessing this thing will be pulling and dissipating 60 watts, might be on the edge for that little fan. Fortunately this is going out to the shed (so I don't burn the house down) and it's getting cold so sub-32 degree temps should keep it under limits.

Maybe. No need to plan for the spring since it will not be economical anyway.

In terms of cost, I'm spending $120 total for 16gh capacity. In theory I should make a max profit of $120 if I start mining in a week.

What's the worst that can happen?

C