Topic: Hacking The KNC Firmware: Overclocking - page 23. (Read 144343 times)

Used compressed air around the boards, heatsinks and VRM's. got the vacuum cleaner out and cleaned the well, used the brush head and got right in there.

Can feel the extra airflow, temperatures have reduced by about 10-5c and temps have reduce in my mining room, which is what i wanted as it has been to hot the last few days, i had to run all my 4vrm board without overclock as it was jsut to hot, i think today i will be fine though.

How to manually set frequencies in v1.0 firmware?
There is no "0x86" lines in new cgminer.sh 

The mounting place is same end of PCB but not 'right next to', opposite end on SAME EDGE OF PCB
If you don't KNOW, better you ignore the suggestion.
Unlikely to revive a die with another power connection.


The VRM are big silver colored things.

2 along each of 4 edges of ASIC for 8 VRM unit
for 4 VRM unit only along the sides and not front or back of PCB  (front is where power and data connect)
4 VRM units have empty spots to mount additional parts.
Shiny places with nothing mounted.


You can look in the web interface and determine how many DC/DC converters.
VRM (voltage Regulator Module) is just a different name for DC/DC converter.

I am ignorant in the practice of reviving a sleeping die, all mine are OK.
My suggestion for a sleeping die is and always has been to try lower clock settings.
I have no idea if this works as reports are scarce and lacking in details regarding clocks.

For the record I refer to them as 'sleeping' as I have heard of success in awakening.


YMMV

Help me please

271 - 1000MHz
281 - 1025MHz
291 - 1050MHz
  - 1075MHz
  - 1100MHz

@ Tolip Wen: You mean the 6 solder points right next to the PCI-e adapter already on the ASIC? Think that could help bring up a slow die?

Another question: on 8VRM October boards, is it totally obvious that there are 8 and not 4? I have an October board and it only looks like there are 4, but what do I know (? - enough about Linux, but not squat about EE).

One more, I just received a board that had gone dead and been RMA'd, and the $%^&&) board ran on four dies for maybe an hour after receiving (KNC sent it back with the voltage cranked and running > 200watts, and it took a few boots to get that under control); then die #2 started refusing to come up. I've tried everything I can think of and will email KNC on Monday, in the meantime I'm down once again to completely sub-optimal hashing and just beyond frustrated and pissed off at this situation; it cost me almost $200 to send that board back and now it's slower than the last effing board (unless I ran it at 75C) before it went tits up.

Any suggestions for a die that looks fine under bertmod but won't come up over an amp or so? Tried everything I can think of - fans all over, air direction directly to VRMS, used to have to crank the temp way up and they'd kick on, but so far no change at a reported 71C, and honestly now I'm a little nervous running another board that hot after learning that the VRMS are frequently MUCH MUCH hotter than the indicated temp. under advanced.

Half venting, half looking for suggestions here.. =/

I thought of a POSSIBLE use for the PCIe extensions.

Step one, cut them in 1/2.
(this should be done even if not reusing them!! IMneverHO)

Step two, solder the white end to an OCT module.

Step three, plug in 2 PCIe 6 pin power connecters from PSU to each module.

YMMV

I did it...switched to bfgminer n turned that core off and sure nuff....0% hw error for 43min Yay...

http://imgur.com/J5OhEXN

at 48min mark...got 3 hw error but % is still 0 lol well i just wanted to know for sure that i would be right if i disabled that core...well i guess back to cgminer i go as i really don't need to lose that core for good since it still produce good hashes too...So can i claim the lowest hw error rate from knc unit? hehe...I wanna at least be top at something in my life

When I first powered it on, without air tunnel, I ran it till it hit 70C.

I made a tunnel and shoved a couple pieces of foam in there (temporarily) to direct airflow.

Runnin between 48C-58C
Moderate OC, I keep each ASIC/module  < 190W.

Have IR gun, I can find hotspots on bottom of PCB +20C compared to LM75(temp part on PCB)
The VRM are the hottest parts.

YMMV

You can do this easily with bfgminer..

PONCHIK, as mentioned by KNC staff on their support forum it is not possible for nov. units

How in a November Jupiter up a voltage of SPI?

Hi all

Is there any way  to push nov. unit above 1000MHz?
My 2 miners are running stable 750Gh/s@675W, looks like there is more to squeeze 

Is it possible to run November units with more than 4 boards?  I may just add 1 port to each board and run 5 modules per board.

I know Octobers can so with other October style units but not Novembers.

Would be really awesome if so!  Lol

Like the watercooling setup!!!

Have the parts to do a setup soon but that Jupiters controller board is dead   I literally have the beaglebone now running another system. LOL  If any of you all have one with or without the BBB on it send me a PM / Price LOL!

Anyone know how to make cgminer disable a core permanently? I have a feeling I'd be at 0% hw errors if it wasn't for this one naughty core with 170 hw errors an hr...think it would be worth it to just turn it off

see that one line going straight up? ya thats the naughty one so that is why i claim 0% hw errors

http://imgur.com/Fncy3Kg

also nice unit WC'in guy   What temps ya running? got a IR gun? you oc'in?

I like the flexability of 2 Saturns.
I have options

Radiator is Alphacool copper.
 
The waterblocks are cheap ebay junk.
They work OK but I had to deck them (make flat BEFORE lapping)
I did not lap them as the ASIC's need to be decked before lapping ALSO.
The ASIC's have a low spot right in the center.
It's criminal but I just pour on the goo, so far.

Another bad thing about the waterblocks is the fittings are different from all the rest (not G1/4) and are a bit restricted (relatively).
They work fine for ASIC
It's why I did not add 90's

Still tweaking the VRM airflow.

YMMV

fantastic engineering.
I was thinking of doing similar setup but never got around to it.

Are they 2 Saturns? Why not use one controller for the 4 asic boards?
What brand are the water blocks and radiator used?

looks great!

Kinda OT but WTF

Some KnCMiner Hardware Pr0n.

qty 1) 60mm 3x 120 radiator
qty 4) Oct. 4 VRM ASIC
qty 2) conrtoller board
qty 2) PSU
qty 2) water pump
qty 2) ethernet extension from controller boards to side of case.


All squeezed into 1 KnCMiner Hoopiter case.


Enjoy

http://i.imgur.com/vf9eekv.jpg
http://i.imgur.com/GXvNz3g.jpg
http://i.imgur.com/oZW5Bfs.jpg
http://i.imgur.com/uhhZlLx.jpg
http://i.imgur.com/IJzV8fA.jpg
http://i.imgur.com/xoHsaiJ.jpg
http://i.imgur.com/tQSoEJI.jpg



I have yet to cut up the cover and some other fabrication tid bits to direct all air over VRMs.
I plan to put a standard furnace filter as the new top, side actually(rotated 90), They live on the edge! ).
All air will enter thru the side and PSU's heat exit one end and ASICs heat exits other end.

WIth different waterblocks (side exit) could fit 6 or 8 ASIC in there!

YMMV

Does anyone have a spare control board for a Jupiter?  Had a November controller board (BB ok, red light blinks faintly every 3-4 seconds (ASIC's are ok)) crap out last night but all I need is the board.  Do not need the BB unless you need it to go with.

Let me know I'd like to get one out today too.