Topic: Hacking The KNC Firmware: Overclocking - page 32. (Read 144362 times)

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.

testing now 

  useful data thank you

The accident that turned on the VRMs was that I loaded a November FW (0.99.2-E) onto an October Jupiter (bad idea). The machine did not hash with the November firmware, I realized my mistake and put back the proper October firmware. Once it started working again I saw that all 8 VRMs were running.

A pure accident.

My guess is that the November firmware has configured the GE VRMs as if they were the Ericsson units that are in the Novembers. Both VRMs talk the same SMBus protocol and are somewhat compatible. For some reason once they are turned on the October firmware do not turn them off again.

The whole process does not feel very safe.


I am using a 0x0271 setting.

Machine is totally stock, I haven't even opened the box.


When these machines were delivered they ran with all 8VRMs turned on at 0.9V. They ran like that for a couple of weeks when they were new with no major problems. It was not until FW 0.93 (or was it 0.95?) they disabled half of the VRMs.

A speculation to why they turned off 4 of the VRMs is that they ran out of components and had to manufacture boards with only 4 VRMs, and to make things "fair" for different customers they decided to run all boards with just 4 VRMs. They gimped the 8 VRM boards.
Another option is that it is unsafe to run with 8 VRMs. That there is something wrong with the boards.

Great place share!

edit: Did you just add the x86 line with clock speed multiple times for each VRM to achieve this?

and sadly if KNC were able to get enough VRMs in the beginning that whole ruse of 'we don't use 8' nonsense would of been avoided and all those lame firmwares would not of been needed either during the scramble to make 4 work..

very nice to see what was intended

Very good.

A few quest:

- are you able you able to turn on all the vrms at will? If yes how?
- what oc settings are you using?
- any mod to get better cooling?

Aha. Thanks tolip_wen. Seems so obvious now

Wow, nice! Since you say old Jupiter, this is October, right? I have an all 8VRM October Jupiter - would you describe how you managed to turn on the VRMs?

1GHz baby!

I accidentally managed to turn on all 8 VRMs on my old 8VRM Jupiter.
Since they were all on and we can control the voltage with the tuning firmware I decided to try some more overclocking.

When running these with all VRMs on they have a base voltage of about 0.9V, so it had to be lowered in the advanced tuning page to get a reasonable voltage. -0.5V works fine.

Currents and temperatures are reasonable.

Total power consumption on the inside of the VRMs is 850W. I guess that will be 950-1000W at the wall. Since my PSU is only 1050W I do not dare to go higher.

The November units will be crazy once you get the tuning FW!

http://i.imgur.com/3w5XCx0.png
http://i.imgur.com/2Ajx7N9.png

How in a November Jupiter change frequency? Will Explain me, please....

Below 1B1 is 1A1 (675)
Below 1A1 is 191 (650)

There is something to be said for the efficiency increase at lower speeds.
I keep it in mind as the efficiency  tradeoff for speed becomes moot at some difficulty level.

Above 251 is 261

ignore the last digit

 950 MHz would be 25 (251)
 975 MHz would be 26 (261)
1000MHz would be 27 (271)
Increase by one

It's hexadecimal so A comes after 9 (instead of 10)


TANGENT alert!
To calculate the value in the last column
Again ignore the last digit, multiply by 2 and add 1
(determined experimentally)

ex:
using 825MHz because the hex and decimal LOOK the same and the math works.
825.0 MHz    201        412

201, drop the 1
20 x 2 = 40
40 + 1 = 41
add a 2 to the end for last column
412 is the same as 201

Think of it as 20 divided by 1
or
40 divided by 2
(plus 1 for the div by 2 instance)

Most should have a calculator with hex mode, try it on a wierd one with a letter!

Can't seem to find the rest the chart below anywhere. Anybody know the settings for below 1B1 and above 251? I'm having good luck underclocking some dies and overclocking others. Thanks.

its not the unit...its the psu...more likely the cables...switching them with brand new pcie cables now n crossing fingers...when i have the unit on idle volts are all almost 12v...wish me luck

EDIT: Changed all pcie cables with 1 new and i think 1 hardly used and bam all back upto 11.8v or more YAY!!! back to 241 woohoo...well hopefully these cables last awhile cause i don't have anymore thank god i didn't fry my unit yet

Thank you kind sir!!

I hope not but perhaps you are just burning them out? Hoping it's just the PSU and the other will fix it..

tryed a different jack on psu...no change...went back to stock til i can figure this out...even on stock 2 modules are lower input volt then the rest...gona try different psu when i get home

got it thanks!

empty?  usually means you opening it in wrong directory (opening a non existing file will create it and be empty)

anyway, best to copy it to /config and make and keep your changed copies there

i was having the same issue, i was able to overclock my october batch to 680.... but it suddenly started doping ghs so i swapped the power supply and tried to overclock it again but when i do vi /etc/init.b/cgminer.sh i dont see txt in there anymore its empty. im  stuck anyone care to help?