sgminer: Baikal Giant X10 / N / B - Open Source - Confirmed OC Giant B!

[tboy32c]

So, I'm confused by the mystic line of code in driver-baikalu.c:

Code:

msg.data[0] = (clk == 0) ? clk : ((clk / 10) % 20) + 2;

I decided to run some common settings.  These resulting values in no way correlate, even if the controller board was accepting input:

mhz  | ((clk / 10) % 20) + 2      
-------------------------------
150   |    17
200   |    2
250   |    7
300   |    12
350   |    17
400   |    2

Perhaps someone smarter can enlighten me?

Also, I found that the following line seems to be leftover from an earlier generation of miner (PiZero_170605_V1.4.img):

Code:

miner->clock        = msg.data[3] << 1;

But I'm not sure why they are bitwise left shifting it.

It seems to me like they're using a lookup table of clock settings to get actual speeds (https://bitcointalk.org/index.php?topic=2790982.msg33090927#msg33090927). If true the max speed might be 490, for a 490/440= 11% overclock (vs 440 stock x10).

[1714138952]

1714138952

[1714138952]

1714138952

[1714138952]

1714138952

[adra12]

It seems to me like they're using a lookup table of clock settings to get actual speeds (https://bitcointalk.org/index.php?topic=2790982.msg33090927#msg33090927). If true the max speed might be 490, for a 490/440= 11% overclock (vs 440 stock x10).

Unfortunately I couldn't find any sign of such a lookup table in the X10 STM firmware (not to be confused with the image file, which has nothing to do with the actual firmware).

I uploaded the pseudo code of the X10 firmware, which I used for static analysis:
https://pastebin.com/PQ8e7H7R

[cod3gen]

Nice work on decompile the firmware adra12, will look a little bit on it tonight. What software did you use to decompile it?

[adra12]

For obtaining the assembly code I used eabi objdump, and for generating a more readable C pseudo code I used IDA Pro. Keep in mind that the pseudo code is just a reasonable guess.

[cod3gen]

Its been brought to my attention from multiple points that this version will OC Baikal Giant B!
Compiling and using my version it will deliver ~10-15% more hashes on LBRY than stock version based on pool side results. Great news for Giant B owners, which im not..!! :-S

Some changes in code and it will deliver even more. Will add a new post when i`ve had access to a Giant B.

[CNoteFan]

Its been brought to my attention from multiple points that this version will OC Baikal Giant B!
Compiling and using my version it will deliver ~10-15% more hashes on LBRY than stock version based on pool side results. Great news for Giant B owners, which im not..!! :-S

Some changes in code and it will deliver even more. Will add a new post when i`ve had access to a Giant B.

Is the Giant B firmware actually accepting the "baikal_set_options" clock command or is this being achieved in another manner?

[cod3gen]

Dont know yet, but it probably is. Still waiting for access so i can be sure.

[Mattthev]

Its been brought to my attention from multiple points that this version will OC Baikal Giant B!
Compiling and using my version it will deliver ~10-15% more hashes on LBRY than stock version based on pool side results. Great news for Giant B owners, which im not..!! :-S

Some changes in code and it will deliver even more. Will add a new post when i`ve had access to a Giant B.

Awesome, but we both need Giant N

[tboy32c]

Its been brought to my attention from multiple points that this version will OC Baikal Giant B!
Compiling and using my version it will deliver ~10-15% more hashes on LBRY than stock version based on pool side results. Great news for Giant B owners, which im not..!! :-S

Some changes in code and it will deliver even more. Will add a new post when i`ve had access to a Giant B.

Interesting!

• How many people have confirmed with you that a modified sgminer can overclock a Giant B?
• Did they share their code changes/improvements with you or give you any idea of what they did?
• What method did they use to measure the 10-15% increase in hash rate?
• What do you have in mind for further software changes in order to boost the OC higher?

[goga5]

After the manipulations, my Giant B did not start. I had to return the old sgminer file.

[CNoteFan]

For obtaining the assembly code I used eabi objdump, and for generating a more readable C pseudo code I used IDA Pro. Keep in mind that the pseudo code is just a reasonable guess.

Thank you for this information! 

I was able to use the following command to extract the assembly code:

Code:

objdump -D --target binary -EL -Mforce-thumb -marm -z GX10_V16.bin > GX10_V16.bin.asm

Any chance you could explain how you got the pseudo code from IDA?  I'm running the following version "ida_pro_v7.0_and_hex-rays_decompiler_armx64armx64x86."  I'd like to compare all the firmware images against one another.

[cod3gen]

Incorrect link to STM Chip firmware is now updated. Also included is multiple firmwares i have received from others.

[goga5]

Incorrect link to STM Chip firmware is now updated. Also included is multiple firmwares i have received from others.

Hello! Tell me please, in this way, I get the firmware of my STM chip ?   https://bitcointalk.org/index.php?topic=2790982.msg29642597#msg29642597  

[CNoteFan]

Incorrect link to STM Chip firmware is now updated. Also included is multiple firmwares i have received from others.

Hello! Tell me please, in this way, I get the firmware of my STM chip ?   https://bitcointalk.org/index.php?topic=2790982.msg29642597#msg29642597  

I just created a python script to do it.  Just duplicate one of the python files in /usr/bin/ and paste this code into it.  Then, if successful, then result will reside as /home/baikal/GX10 (Extracted).bin

Code:

#!/usr/bin/env python
import subprocess
import os
import sys
from subprocess import Popen, PIPE
import fcntl
import time
import glob

tmpfile = '/home/baikal/GX10 (Extracted).bin'
USBDEVFS_RESET= 21780

# enter dfu mode
def enter_dfumode():
  subprocess.call('sudo /opt/scripta/startup/miner-stop.sh', shell=True)
  subprocess.call('echo 0 > /sys/class/gpio_sw/PA18/data', shell=True)
  subprocess.call('echo 0 > /sys/class/gpio_sw/PA10/data', shell=True)
  subprocess.call('echo 1 > /sys/class/gpio_sw/PA10/data', shell=True)

def exit_dfumode():
  subprocess.call('echo 1 > /sys/class/gpio_sw/PA18/data', shell=True)
  subprocess.call('echo 0 > /sys/class/gpio_sw/PA10/data', shell=True)
  subprocess.call('echo 1 > /sys/class/gpio_sw/PA10/data', shell=True)

def reset_usb(driver):
  try:
    lsusb_out = Popen("lsusb | grep -i %s"%driver, shell=True, bufsize=64, stdin=PIPE, stdout=PIPE, close_fds=True).stdout.read().strip().split()
    bus = lsusb_out[1]
    device = lsusb_out[3][:-1]
    f = open("/dev/bus/usb/%s/%s"%(bus, device), 'w', os.O_WRONLY)
    fcntl.ioctl(f, USBDEVFS_RESET, 0)
  except Exception, msg:
    print ""

def extract_firmware():
  print "Extracting firmware..."
  cmd = 'sudo dfu-util -a 0 -d 0483:df11 -s 0x08000000:leave -U ' + tmpfile
  subprocess.call(cmd, shell=True)
  #n = os.path.getsize(fwfile[0])
  #cmd = 'cmp '+ fwfile[0] + ' ' + tmpfile + ' -n ' + str(n)
  #ret = subprocess.call(cmd, shell=True)
  #subprocess.call('sudo rm -rf /home/baikal/tmp.bin', shell=True)
  #return ret
 
enter_dfumode()
reset_usb("STM32F407")
reset_usb("DFU")

extract_firmware()
exit_dfumode()

print "Done"

[Dlikrot]

Would pay for working x13 ^^

[mindtrip]

I have been tinkering with mine but so far no luck getting the hidden Algo's to work

[rogerthat55]

So I thought I would try it out, the code from https://github.com/cod3gen/sgminer-baikal I first downloaded the current version of OrangePI Zero from Baikal at https://www.baikalminer.com/support04.php and installed it on a new micro SD card, to keep the original safe. Then I followed the steps to install and compile cod3gen versions of sgminer (note: I couldn't even do apt-get install git without first setting up the swap). Now, when I got to the part of "driver-baikal.h" and do changes to #define BAIKAL_CLK_DEF I made it 330 for the X10 (just 10% more than 300). When I finished compiling and started it up, it was reporting the clock at 440MHz, but the hash rate was still normal; no overclock. Just to check, I restored the default pools, and checked all the algorithms, and I didn't see any overclocks.

[Swayzeee]

We need Wolf here

Been trying to get WOLF. on board since Jan. He doesn't want to tell his secrets.

[CNoteFan]

We need Wolf here

Been trying to get WOLF. on board since Jan. He doesn't want to tell his secrets.

Hope someone who figures it out will at least give me a few days to mine and make a little for my effort, not to mention @Cod3gen's hard work.

[cod3gen]

So I thought I would try it out, the code from https://github.com/cod3gen/sgminer-baikal I first downloaded the current version of OrangePI Zero from Baikal at https://www.baikalminer.com/support04.php and installed it on a new micro SD card, to keep the original safe. Then I followed the steps to install and compile cod3gen versions of sgminer (note: I couldn't even do apt-get install git without first setting up the swap). Now, when I got to the part of "driver-baikal.h" and do changes to #define BAIKAL_CLK_DEF I made it 330 for the X10 (just 10% more than 300). When I finished compiling and started it up, it was reporting the clock at 440MHz, but the hash rate was still normal; no overclock. Just to check, I restored the default pools, and checked all the algorithms, and I didn't see any overclocks.

It does not overclock X10 and N. Only Giant B