Hacking The KNC Firmware: Overclocking

[hostmaster]

  useful data thank you

[1714222057]

1714222057

[1714222057]

1714222057

[1714222057]

1714222057

[elasticband]

testing now 

[elasticband]

DO NOT ATTEMPT THIS


Three attempts with a single 8vrm board for tests and each time i got the same result, a CRAZY buzzing noise from each VRM.

I have had this before from a board that got RMA'ed.


edit: few minutes after powering off i hear this clicking noise, like a switch being flicked. I also heard this on the faulty 4vrm board.

[corrow]

Anyone can confirm this? 

[crashoveride54902]

DO NOT ATTEMPT THIS


Three attempts with a single 8vrm board for tests and each time i got the same result, a CRAZY buzzing noise from each VRM.

I have had this before from a board that got RMA'ed.


edit: few minutes after powering off i hear this clicking noise, like a switch being flicked. I also heard this on the faulty 4vrm board.

did you remember to turn down the volts?

[jelin1984]

How much we must down the volt?

[CYPER]

How much we must down the volt?

Forever and to infinity 

[jelin1984]

Thanks my friend

[mekadeka]

Any ideas on how to reduce errors? Does tuning do the trick?

I've been getting some crazy machine-side speed reporting from an October Saturn. The KNC GUI is showing 398gh. SSHing into the machine is showing 398/355.

Poolside, however, I'm apparently only getting 227? Can this be accurate?  I presume the problem is errors, but I don't know...

[j03]

Any ideas on how to reduce errors? Does tuning do the trick?

I've been getting some crazy machine-side speed reporting from an October Saturn. The KNC GUI is showing 398gh. SSHing into the machine is showing 398/355.

Poolside, however, I'm apparently only getting 227? Can this be accurate?  I presume the problem is errors, but I don't know...

Just my 0.02: I noticed better hash rate matching between the pool and client with cgminer 3.9.0 on firmware 0.99.1-tune. As far as reducing errors, I've had good luck with not overclocking, or underclocking, some dies and overclocking others. Use the bash script that CYPER posted in this thread a while back in order to set per die clock settings. I still use Bertmod ZPM fix to analyze individual module/die performance.

[ImI]

Any ideas on how to reduce errors? Does tuning do the trick?

I've been getting some crazy machine-side speed reporting from an October Saturn. The KNC GUI is showing 398gh. SSHing into the machine is showing 398/355.

Poolside, however, I'm apparently only getting 227? Can this be accurate?  I presume the problem is errors, but I don't know...

Just my 0.02: I noticed better hash rate matching between the pool and client with cgminer 3.9.0 on firmware 0.99.1-tune. As far as reducing errors, I've had good luck with not overclocking, or underclocking, some dies and overclocking others. Use the bash script that CYPER posted in this thread a while back in order to set per die clock settings. I still use Bertmod ZPM fix to analyze individual module/die performance.

may i ask what the Bertmod ZPM fix is? I only know bertmod 0.4 so far.

[Biffa]

Any ideas on how to reduce errors? Does tuning do the trick?

I've been getting some crazy machine-side speed reporting from an October Saturn. The KNC GUI is showing 398gh. SSHing into the machine is showing 398/355.

Poolside, however, I'm apparently only getting 227? Can this be accurate?  I presume the problem is errors, but I don't know...

Just my 0.02: I noticed better hash rate matching between the pool and client with cgminer 3.9.0 on firmware 0.99.1-tune. As far as reducing errors, I've had good luck with not overclocking, or underclocking, some dies and overclocking others. Use the bash script that CYPER posted in this thread a while back in order to set per die clock settings. I still use Bertmod ZPM fix to analyze individual module/die performance.

may i ask what the Bertmod ZPM fix is? I only know bertmod 0.4 so far.

If I remember correctly its a hack someone called ZPM did to make 0.3 work with some later firmware.

[j03]

may i ask what the Bertmod ZPM fix is? I only know bertmod 0.4 so far.

I haven't used the latest Bertmod versions, but ZPM's fix used to make Bertmod prettier and more useful. If this link doesn't bring you directly to the original post from ZPM, it's post #35 that contains the download link. It still works on the latest firmware:

http://forum.kncminer.com/forum/main-category/main-forum/6183-bertmod-0-2-unofficial-firmware-mod-feedback-thread?p=9442#post9442

[ImI]

may i ask what the Bertmod ZPM fix is? I only know bertmod 0.4 so far.

I haven't used the latest Bertmod versions, but ZPM's fix used to make Bertmod prettier and more useful. If this link doesn't bring you directly to the original post from ZPM, it's post #35 that contains the download link. It still works on the latest firmware:

http://forum.kncminer.com/forum/main-category/main-forum/6183-bertmod-0-2-unofficial-firmware-mod-feedback-thread?p=9442#post9442

thx!

[mekadeka]

Any ideas on how to reduce errors? Does tuning do the trick?

I've been getting some crazy machine-side speed reporting from an October Saturn. The KNC GUI is showing 398gh. SSHing into the machine is showing 398/355.

Poolside, however, I'm apparently only getting 227? Can this be accurate?  I presume the problem is errors, but I don't know...

Just my 0.02: I noticed better hash rate matching between the pool and client with cgminer 3.9.0 on firmware 0.99.1-tune. As far as reducing errors, I've had good luck with not overclocking, or underclocking, some dies and overclocking others. Use the bash script that CYPER posted in this thread a while back in order to set per die clock settings. I still use Bertmod ZPM fix to analyze individual module/die performance.

Thanks very much, j03. I'll try testing that way.
I have not been able to find the bash script, but I'll look again. A couple questions:
1. Is Bertmod ZPM different from v4? Is this from the KNC forums?
2. Have you done any SPI voltage, SPI frequence, or per-die tuning in the KNC Advanced gui?
3. What's your order of procedure? I have:
-flash a KNC firmware
-run enablecores.bin (just out of habit)
-let the machine spin up
-ssh into cgminer
-"vi /etc/init.d/cgminer.sh" to get to clock settings
-adjust clock settings
-type esc, then a colon, and then "wq", then enter
-machine seems to reboot on it's own, i don't restart
-watch the new hashrate
-mess with tuning, even though I don't know what im doing
-compare pool and machine hash rates
-wonder what's going on (LOL)

Could you share? Thank you very much.

[j03]

Thanks very much, j03. I'll try testing that way.
I have not been able to find the bash script, but I'll look again. A couple questions:
1. Is Bertmod ZPM different from v4? Is this from the KNC forums?
2. Have you done any SPI voltage, SPI frequence, or per-die tuning in the KNC Advanced gui?
3. What's your order of procedure? I have:
-flash a KNC firmware
-run enablecores.bin (just out of habit)
-let the machine spin up
-ssh into cgminer
-"vi /etc/init.d/cgminer.sh" to get to clock settings
-adjust clock settings
-type esc, then a colon, and then "wq", then enter
-machine seems to reboot on it's own, i don't restart
-watch the new hashrate
-mess with tuning, even though I don't know what im doing
-compare pool and machine hash rates
-wonder what's going on (LOL)

Could you share? Thank you very much.

Bash script is here. Keep the original clock setting on line 61 and paste the script in beginning on the next line (line 62):

https://bitcointalk.org/index.php?topic=313978.msg4432410#msg4432410

That script is for a 4 module, 4 VRM machine and assumes you modules are plugged in 0 through 3 connectors on the controller board. Change to your setup. Also, you need to add or subtract the number of module IF statements based on your machine. Mine is a Jupiter+2, so I added 2 additional statements for modules 4 and 5 that are plugged into the controller board.

The 4 die clock settings within each module IF statement assumes 4 VRM. If you have 8 VRMs, add 4 more within each IF statement.

1. Not sure about the diffs in Bertmod versions.
2. I haven't touched Advanced Tuning because I also have no idea what I'm doing there
3. My procedure is the same as yours but you forgot /etc/init.d/cgminer.sh restart after :wq the changes. Don't reboot the machine or you'll lose your changes. Also applying changes at Advanced Tuning wipes changes to cgminer.sh. There's instructions in this thread to make your custom cgminer.sh persist on reboot, but I haven't done it.

I currently get around 920GH/s on a Jupiter+2.

[mekadeka]

Bash script is here. Keep the original clock setting on line 61 and paste the script in beginning on the next line (line 62):

https://bitcointalk.org/index.php?topic=313978.msg4432410#msg4432410

That script is for a 4 module, 4 VRM machine and assumes you modules are plugged in 0 through 3 connectors on the controller board. Change to your setup. Also, you need to add or subtract the number of module IF statements based on your machine. Mine is a Jupiter+2, so I added 2 additional statements for modules 4 and 5 that are plugged into the controller board.

The 4 die clock settings within each module IF statement assumes 4 VRM. If you have 8 VRMs, add 4 more within each IF statement.

1. Not sure about the diffs in Bertmod versions.
2. I haven't touched Advanced Tuning because I also have no idea what I'm doing there
3. My procedure is the same as yours but you forgot /etc/init.d/cgminer.sh restart after :wq the changes. Don't reboot the machine or you'll lose your changes. Also applying changes at Advanced Tuning wipes changes to cgminer.sh. There's instructions in this thread to make your custom cgminer.sh persist on reboot, but I haven't done it.

I currently get around 920GH/s on a Jupiter+2.

Thank you very much j03. I'm not clear on exactly what to modify in the bash script for an October Saturn, but I'm getting there.

KNC gui shows ASIC slot #3 and ASIC slot #4 are used. I assume this is 3 and 4 on the control board? There are two free spots to the left and two to the right.

---------------beagleboard side----------------------
 plug        plug        plug       plug        plug       plug
{FREE}   {FREE}   {USED}   {USED}   {FREE}   {FREE}  
                             |            |
                             |            |
                             |            |
                             |            |
                          [ASIC]    [ASIC]
                           4 dies     4 dies
---------------ASIC/FAN side-----------------------


How do you count these spots for the script?


EDIT: I'm thinking it's [ $p -eq 2 ] and [ $p -eq 3 ] based on counting from zero?

[elasticband]

Bfgminer just names your asic 0-3. So working out the control board placement is irrelevant I think. I have not done this so could be wrong though

[mekadeka]

Bfgminer just names your asic 0-3. So working out the control board placement is irrelevant I think. I have not done this so could be wrong though

Was that for me? I'm not sure it's accurate. I'm not using bfgminer, btw. The bash instructions say to count ports starting with 0 in order to ID the "if" statements. Since my hardware is only using 2 of 6 plugs/ports, the bash script needs to know which it is sending instructions to.

[mekadeka]

Does someone have the code handy for a cgminer.sh file that has each board or die for that matter OC'able?  I think I can get more performance out of particular boards but not by doing the entire Jupiter. Thanks.

Paste just below line 61 or whichever line the original 0x86 register is at.
Also change if [ $c -eq 0 ] to your particular board configuration.

http://pastebin.com/vy7jppS3

Cyper, please, does the script require the "DC/DC ID" for each die or the "Die ID"? KNC's GUI shows 1,2,3,4 for Die ID but 0,2,4,7 for DC/DC ID, but I'm not clear on which to use.


       if [ $c -eq ? ]
       then
               cmd=$(printf "0x86,0x%02X,0x02,4xB2" $c)
       fi
       if [ $c -eq ? ]
       then
               cmd=$(printf "0x86,0x%02X,0x02,4xB2" $c)
       fi
       if [ $c -eq ? ]
       then
               cmd=$(printf "0x86,0x%02X,0x02,4xB2" $c)
       fi
       if [ $c -eq ? ]
       then
               cmd=$(printf "0x86,0x%02X,0x02,4xB2" $c)
       fi
fi