Topic: Hacking The KNC Firmware: Overclocking - page 7. (Read 144362 times)

Stop potting before you get cubic shape.
More surface area is better.
A thin cover on everything would be ideal.

Potting is a one way street!!!

YMMV

You might check about chlorine and copper. No idea personally.
You can just place a radiator in the pool water for inexpensive heat exchanger.
Then your pool plumbing can be done with a hose cutter and screw driver frugally.

PVC thru the attic or copper or tough garden hose.
Does your attic ever freeze?

I'm contemplating burying a small coil of copper in the yard for heat exchange.
I have a high water table here.

Can start on a budget with garden pump and a bucket.
Good to have backups too, save the air cooling toys.

I live in a shoebox with single window unit A/C and the as delivered had me hiding in other room.
I cool with outdoor Illinois air whatever it is.
Today I start with 82F/28C/300K ambient.

Watch out for cheapo $20-25 waterblocks on ebay.
(good? bad? depends if ya know ahead of time)
The blue/black metal ones, the fittings are not G1/4 so whatever you get is what you have forever.
The ones for GPU with 90 installed do not come with CPU mounting.
DO NOT TRY TO TURN THE 90 BARBS they are glued in.
(unless you want to clean and reglue threads)
Straight hose barbs are great but hose has minimum bend radius, will it fit?
They are not flat. Lapping WB takes some time and cost of wet sandpaper 400-1200 as needed.
I like em for value, have to add labor, what is time worth? fun?
I doubt customer satisfaction is high for most first time customers with theese.
You have been warned

I DID get some clear ones from ebay that look real good (also needed minor lapping)

Next stop for decent WB is about $40, times 5 and most people are done calculating.
Pumps, radiators, fittings, adds up fast, how committed are you?
Alphacool rotating 90's are shit avoid at all cost! One of me ASIC went swimming.
EDIT Alphacool makes great radiators.

Also keep in mind airflow for VRMs.
WB obstructs air and mounts block more.
I go with high mounts.

FrozenCPU has cupons on web if you google.

YMMV


EDIT
A pool COVER pump is about the right size for watercooling.
Triple check for leaks if you use much higher flow and pressure.
Some of the cheapo flex fittings may explode downstream from a pool filter pump.
It is the better problem to have. RE: too high/low pressure!
Great idea, another reason to envy peeps with pools as I sweat by the hot ASICs

I found the former storage location of some magic smoke.
The smoke was gone when I got there but the evidence remains.

http://i.imgur.com/8ihuFBc.jpg

Result of plugging PSU's in too quickly when using hand did load sharing circuit.
If I wait 4 seconds OK. Power on at same time OK (end of power outage).

My interim solution will be a PSU wall switch/breaker to keep them together at power on.
Hope to some day use brains in PSU for protection and PSU feedback.

The good news, The $50 PSU took the hit for my blunder and the five 2000$ ASIC did not.

YMMV

sounds interesting.. i look forward to your write up! this would certainly void the warranty.

i am (seriously) considering piping water in a loop from my pool and using that as a 30,000 gallon heat sink

i even bought a variable speed pool pump so i can run it 24/7 at a moderate flow rate. just need to work out a few kinks linking WC parts together and how to safely run water through my attic.

After 2.5 days running with Volt-tuned 500 Mhz, this is what currently is possible: 3.500 Th/s

What do you think about potting the entire thing then putting it into running water? A giant cheap water cooler

Thanks for having a look! A good mystery indeed.. Will tinker with airflow some more this week, I've already plugged the leaks at the top (with no measurable benefit unfortunately).

If that were true wouldn't the low V going from #0 (number one's buddy for first die) take a much longer route to same quadrant of ASIC? That distance is so critical that they make sure VRM has output on correct side for shortest run.
Also not symmetrical left/right possibly indicating low V distance priority.
The low V high current traces might be different on that side.
The goal may not have been fully met for some reason.

1 and 0 are likely top right and directly below in that order IMneverHO.

That Iron looks comfy!
Pictured tip has cylindrical room to add more mass with big wire and pliers if ya ever needed such a thing.
A well tinned tip KNOWS somehow if you have a spare tip and lasts longer in self preservation BTW
Finagles Law

YMMV


EDIT the one in the lower right is turned sub optimally for low V, possibly optimized for 12V only?
Temps marginally say 2 is a tad warmer than 3 sample size 5.

Same trend, glad I had other reason for mod.
Voltage is same everywhere on PCB to 2 decimal places on 20V scale.
12.01VDC delivered @ connector and VRM. my rail is 12.09 so even 2x12AWG is hungry wire.
I measured from rail to VRM filter caps and connector to caps all side VRMs. modded and unmodded boards.

Based on fan direction and twist of airflow,
the right may be in the wind shadow of hot heatpipes.
Might get more air downstairs on one side too.
Downwind are the hottest, pre heated coolant.

Cardboard origami twist reducing baffles to test?
(small glass xmas ornament box seperator like thingies)
Plug leaks past top of heatsink foam! no question wasted air
Tape across some ASIC heatsink fins to trade ASIC air for VRM air?
Restrict one side till even?

A warmer side of ASIC would do something too.
The heatsink is about 35mm^2 on about a 40mm^2 heatspreader.
The alignment holes are larger than alignment screws.
Did we all happen to tighten screw closest to connector before farther one or vice-versa?
The top of the ASIC are not perfectly flat, high in middle mine are.
Can see the PCB warp, one side might warp less?.
I recall the bumps being oval but not orientation.
Was just glad it was high and not low, less to remove to make flat.

Without bfgminer, looking for core trends is painful at best from cgminer output.
Can find hotspots with bfg on 28 easy.

MOST chip lids are not flat, but most not over 300W thru less than 16cm^2 heatspreader either.

A VRM reflow issue on that side of PCB? It's gotta drive reflow setup guys nuts them things.

Good mystery!

YMMV

Yep... I managed to "plumb" the plug on top of the  PCB-pins... but I know, it could be done much better with a hotter iron.

I will buy something like this:
http://www.ersa.com/art-0710cd-358-5046.html

I now think that numbering is from right-top (1), left-top(2), right-middle(3), left-middle(4)...and than 5--8 on the front of the Fan....But we need to test that still (no time yet...)

I've got the exact same behavior.. DC/DC 0-3 are easily 10C hotter then DC/DC 4-7 and I don't have any special features beyond an external box fan. I know you don't think it is, but a voltage drop in the board itself delivering lower input voltage to the VRMs would cause higher current/heat.. maybe the bus is a little undersized that feeds those 4.. only a guess. Could also be a firmware issue.

tolip_wen.. would be very interested to know if you see this with your VRMs and the dedicate power runs?

16W may not melt thru all solder in the via/hole.
Quite acceptable for your method but hard to get it just so.
You must have the knack for it and more patience than me.
If you did it with 16, 40 will be enough for you forever, exception big power circuits.
Heavy tips help lots too if ya git er done quick, some spare heat!
(PSU, thumper amps, soldering AC wiring, etc.)

If it works it works.

A bit of time/frustration vs additional tool money, depends on budget/frequency of use.

YMMV

I think air soldering is better in that case.

First PCB I did with a Radio Shack 40W iron with a huge perfect tip.
It worked OK was prepared for a struggle and it was.
I raised the PCB with some hot air too.
If it did not flow quickly I stopped and let the tip reheat.

I destroyed the tip and flipped on the workstation to 896F wattage unknown but 70-100 ish.
WIth the ground pins you are also fighting the top, bottom, middle gnd??? layers.
One reason I nixed the plastic, end of mostly straightened pins was much easier target.
I was also dealing with 2 long 12AWG heatsink wires per 3 pins (thourogly fluxed and pre-tinned).

I'd suggest 75+ if you have a choice, use what you have if reasonable.
I don't recommend guns but cheap spares for rude big stuff are.
http://www.homedepot.com/p/Weller-75-Watt-Soldering-Gun-Kit-7200PKS/100143633
http://www.homedepot.com/p/Weller-100-140-Watt-Soldering-Gun-Kit-8200PKS/100085564

I have an old 200/260 monster that I inherited, not too big for the power connections if ya don't lurk forever.

OBTW
Got me VRMs 70s mid80s instead of 80s, upper90s, and ker plink 0 watt ambient PCB temp.
Still just nekkid in the breeze but I added a playing card sized piece of cardboard to WB on upwind side to offset the WB obstruction.

Weeee, now I could coast on cooling till faster clock magically appears.
I still have one more speed on AC box fan.

YMMV


EDIT
If anyone is interested I had an idea for stock parts survival lunch.
I put it there as it probably applies to all nep owners not just OC folks.

https://bitcointalksearch.org/topic/m.7727618

Still no high a lot

My, cgminer Average with running it 2 days: 3.483 Th/s  and 48667 WU

final which is avg speed of neptune when push it at 500mhz?

That's also my experience: the solder on the PCIe connectors doesn't melt well (or not at all)... the first additional plug I soldered on the backside of the board, just melted the plastic of the plug itself while the original PCB-solder was still hard... I had to plug in a PCI-cable into the plug first, to function as a heat sink. This worked more or less, but still wonder how well (or bad) the solder joined...maybe I need to redo it with a bigger Iron?  What wattage did you use?