FYI: Think twice about replacing the thermal paste

[EETanner]

I'm no hardware hacker, but I thought I was decent enough to put some antech on the darn thing.  After pulling the case off the jally, pulling the heat sink, removing whatever thick stuff they had on there, applied the antech and put it back together.

It was running at 5.1GH/s (with the case removed and the fan turned upside-down)

Now it's running at 3.5GH/s.

In short.  If you are getting good speeds, think about possibly not messing with the thermal paste

[1714627894]

1714627894

[1714627894]

1714627894

[ElitePork]

Replace the thermal pad and used thermal paste myself. from 7.9GH/s to 8.1GH/s. Based on 24 hours mining.
Something u did must be incorrect... Do a ZXV to your jallies and check? There's a .NET tool to use to check if you want to...

[bcp19]

I'm no hardware hacker, but I thought I was decent enough to put some antech on the darn thing.  After pulling the case off the jally, pulling the heat sink, removing whatever thick stuff they had on there, applied the antech and put it back together.

It was running at 5.1GH/s (with the case removed and the fan turned upside-down)

Now it's running at 3.5GH/s.

In short.  If you are getting good speeds, think about possibly not messing with the thermal paste

Did you use isopropyl alcohol and a lint free cloth to remove the old paste?  Some cleaners can leave behind a residue that will affect thermal conductivity.

[Walking Glitch]

You may have incorrectly mounted the heatsink back on. I've seen this with people building computer, they put a better heatsink and fan on, and do their own thermal paste, but they can't push their CPU any faster than with the stock cooler and thermal pad. 20 times out of 10 it's because either the heatsink was mounted incorrectly, it was not making even half contact with the chip due to something being messed up with the mount, or the cooler was too loose, causing the above problem.

EDIT: It could also be that you just used too much thermal paste. People don't actually need as much as they usually think they do. It took my two tries to get the amount just right the first time I installed a CPU+after market heatsink and fan, however, I am a very quick learner and have an extremely accurate spatial judgement, so YMMV.

[mazedk]

+1 for the "try again" option.

Its possibly a poor pasting job - either to much, or too little. And also make sure its properly cleaned before you reaply the thermal paste.

I'v done quite a few GPU re-pastes and if you do it properly, it'll help you quite a bit.

[Walking Glitch]

+1 for the "try again" option.

Its possibly a poor pasting job - either to much, or too little. And also make sure its properly cleaned before you reaply the thermal paste.

I'v done quite a few GPU re-pastes and if you do it properly, it'll help you quite a bit.

^THIS^ Dropped the temp on my stock HIS 6850 from 105 while Playing SR3 down to about 60C. It was a HUUUUGE fucking mess that they made with it.It also helped that I tightened the heatsink down so it was making good contact with my proper amount of thermal paste. Didn't matter a few week later though, their terrible fan seized up, and re-oiling it made no difference like it did on some of the other ones I had that happen to.

[EETanner]

Broke it down again.  Cleaned off the paste with 91% isopropyl alcohol and q-tips.  Re-applied, put it back together and BAM!

1.3GHS

Seems like every time I touch this thing it gets worse

Someone said "Replace the thermal pad and used thermal paste myself."  On my jally there is a aluminum pad on the bottom of the board, however there is no pad between the heat sink and the chips.  Is there supposed to be?

Thanks all.  I think now I'm going to drink.

[testerx]

Maybe your particular unit has too much of a gap between the heatsink and the chip for paste to work well, if that's the case then go buy a new thermal pad.

[EETanner]

"Maybe your particular unit has too much of a gap between the heatsink and the chip"

Is there supposed to be a plate between the chip and the heat sync?  My unit simply doesn't have that. 

I was concerned that there was a good connection.  So I had screwed it on fairly tight.  Went through the whole damn process _AGAIN_.  I found after only having the screws secure but not tight got me back up to ~3.5GH/s.

Still not where I was before I started messing with it.  Perhaps the antec isn't good for this type of chip?  Can someone recommend better thermal material?

[Unacceptable]

Try Ceramique,non conductive & thick 

http://www.newegg.com/Product/Product.aspx?Item=N82E16835100018

Or find out how thick the TIM pad was,if there was one installed.

[testerx]

"Maybe your particular unit has too much of a gap between the heatsink and the chip"

Is there supposed to be a plate between the chip and the heat sync?  My unit simply doesn't have that. 

I was concerned that there was a good connection.  So I had screwed it on fairly tight.  Went through the whole damn process _AGAIN_.  I found after only having the screws secure but not tight got me back up to ~3.5GH/s.

Still not where I was before I started messing with it.  Perhaps the antec isn't good for this type of chip?  Can someone recommend better thermal material?

I've never taken it apart but I don't think these have a plate, but thermal pads are generally about a millimeter thick whereas thermal paste requires really good contact with the heatsink so if you have a unit where the chips and the heatsink don't really touch the thermal pad would have done a better job transferring heat than thermal paste.  It probably varies unit by unit anyway, but chances are that if screwing it tight got you back up to 3.5GH/s then better contact would get you back to your original speed.

Frankly I'm surprised that people manage to use thermal paste and get good conductivity, especially on the single scs where you're trying to make contact with so many chips.

Anyways, taking apart these things is of course going to be at your own risk so if you don't know what you're doing I would advise against it, but if your chip and heatsink aren't contacting well then you could either use a metal shim (more costly and difficulty) or switch to a thermal pad that would buffer the space.  Or maybe the drop in speed doesn't have to do with either, who knows.

Did you notice the temps go up after your tinkering?  If that's the case it's probably the lousy contact.  I'd probably just be lazy and throw some thermal pads back on there. 

[bcp19]

"Maybe your particular unit has too much of a gap between the heatsink and the chip"

Is there supposed to be a plate between the chip and the heat sync?  My unit simply doesn't have that.  

I was concerned that there was a good connection.  So I had screwed it on fairly tight.  Went through the whole damn process _AGAIN_.  I found after only having the screws secure but not tight got me back up to ~3.5GH/s.

Still not where I was before I started messing with it.  Perhaps the antec isn't good for this type of chip?  Can someone recommend better thermal material?

The speed loss from over-tightening would indicate that the added pressure either warped something or you ended up with one side tighter than the other and only partial contact.  I would suggest at this point to loosen all the screws and without breaking the paste, just barely snug the screws and power it up.  Once I had a good rate showing I would then turn one of the screws a 1/4 turn tighter and see if anything changes.  If the speed decreases, loosen that screw back that 1/4 turn and try another.  It will be a slow and painstaking process, but if something is slightly warped, this may help to find a better fit and get your lost speed back.

[ElitePork]

"Maybe your particular unit has too much of a gap between the heatsink and the chip"

Is there supposed to be a plate between the chip and the heat sync?  My unit simply doesn't have that. 

I was concerned that there was a good connection.  So I had screwed it on fairly tight.  Went through the whole damn process _AGAIN_.  I found after only having the screws secure but not tight got me back up to ~3.5GH/s.

Still not where I was before I started messing with it.  Perhaps the antec isn't good for this type of chip?  Can someone recommend better thermal material?

No, you dont need to screw them so tight... Just turn until you feel resistance and stop. If you screw too tight, the thermal paste will be "squeeze" out of contact... also, you should use very little thermal paste... Maybe less than a drop size for each chip...

Edited: have you check on the engines itself? Do a bfgminer rpc or cgminer api call. If you are using windows, red wolf 2 has a program that can help you on that also.

[Ridicuss]

"Maybe your particular unit has too much of a gap between the heatsink and the chip"

Is there supposed to be a plate between the chip and the heat sync?  My unit simply doesn't have that. 

I was concerned that there was a good connection.  So I had screwed it on fairly tight.  Went through the whole damn process _AGAIN_.  I found after only having the screws secure but not tight got me back up to ~3.5GH/s.

Still not where I was before I started messing with it.  Perhaps the antec isn't good for this type of chip?  Can someone recommend better thermal material?

No, you dont need to screw them so tight... Just turn until you feel resistance and stop. If you screw too tight, the thermal paste will be "squeeze" out of contact... also, you should use very little thermal paste... Maybe less than a drop size for each chip...

Edited: have you check on the engines itself? Do a bfgminer rpc or cgminer api call. If you are using windows, red wolf 2 has a program that can help you on that also.

^^ This,
When I seat mine I use only  a torx bit and fingers. No driver. If you only have a driver go lightly.

On another not so popular note,  Are you sure your heatsink is covering both chips?  On one of my Jals the heatsink just barely covers the edge of one of the chips, I mean barely..
If it is not covering both chips completely it may be thermally clocking down.

My 2 satoshi

[EETanner]

Now this is ugly

Code:

bfgminer version 3.1.1 - Started: [2013-09-02 22:58:15] - [  0 days 00:11:01]
--------------------------------------------------------------------------------
 5s: 3.34 avg: 3.16 u: 0.08 Gh/s | A:4 R:0+0(none) HW:23103
 ST: 2  GF: 0  NB: 4  AS: 0  RF: 0  E: 1.32  U:0.4/m  BS:231
 Connected to stratum.bitcoin.cz diff 4 with stratum as user texsantos.asic
 Block: ...fab9e9b3 #255824  Diff:65.8M (470.7Th/s)  Started: [23:06:11]
--------------------------------------------------------------------------------
 [P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
 OCL 0:                |  22.0/ 22.2/ 32.6Mh/s | A:2 R:0+0(none) HW:    0
 BFL 0:  23.0C/26.0C   |  3.65/ 3.14/ 0.05Gh/s | A:2 R:0+0(none) HW:23205
--------------------------------------------------------------------------------

 [2013-09-02 23:02:45] Stratum from pool 0 requested work update
 [2013-09-02 23:03:05] Stratum from pool 0 detected new block
...

Lot's o' HW errors.  So here is the plan for now:

Order some thermal pads and a less annoying fan.  Then see what we get.  I'm pretty sure it's hosed up -- mining pool says "12.622 MH/s".  I wouldn't mind flashin' the firmware either and have been looking for an excuse to mess with the rasp.

Ah well.  It's fun learning/mucking-about with this stuff.  Been quite an adventure.

[iikun]

Is that temp. data accurate?  Seems insanely low

[kjj]

Paste is simply not suitable here.

Paste is intended to fill in microscopic gaps between a single package and a single heatsink.  It can not and will not work on multiple chips with a single heatsink.

The root problem is that multiple chips are never coplanar.  Hell, even single chips are rarely parallel to the board.  If you have a single chip with a single heatsink mounted with springs, you can use paste to fill in the microscopic gaps between them while the springs allow the heatsink to float relative to the board, matching the chip's bulk surface.

If you have multiple chips, they will all be at funny angles, and all at different heights.  A single heatsink will never make good contact, not with all of them, and usually not even with one of them.  For multichip designs, it is customary to back the heatsink off by a millimeter or two, use a rigid mounting to the board, and fill in the substantial gap with a thermal pad.

Something like tflex 600 (PDF) would be appropriate.

[superresistant]

Did you fix it ?

[EETanner]

Got my "Phoby Thermal Pad Xt 1mm" today.  I cut a little rectangle out of it so that it would cover, with one piece, both chips.

Broke the jally down (again) cleaned off the paste very carefully and thoroughly.  I placed the pads between the chips and the heat sink and screwed the heat sync on securely but not tightly.

Note that I didn't put it completely back together.  In addition I have a stronger fan I pulled from a desktop blowing down on it.

A ten minute test:

Code:

 bfgminer version 3.1.1 - Started: [2013-09-17 00:17:15] - [  0 days 00:10:19]
--------------------------------------------------------------------------------
 5s: 1.68 avg: 1.70 u: 0.40 Gh/s | A:55 R:1+0(1.7%) HW:4561
 ST: 2  GF: 0  NB: 4  AS: 0  RF: 0  E: 3.05  U:5.3/m  BS:125
 Connected to stratum.bitcoin.cz diff 1 with stratum as user texsantos.asic
 Block: ...e178b81f #258428  Diff:113M (806.2Th/s)  Started: [00:25:10]
--------------------------------------------------------------------------------
 [P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
 OCL 0:                |  22.0/ 22.0/ 13.9Mh/s | A: 2 R:0+0(none) HW:   0
 BFL 0:  25.0C/27.0C   |  1.67/ 1.67/ 0.38Gh/s | A:53 R:1+0(1.8%) HW:4561
--------------------------------------------------------------------------------

So yeah.  I didn't fix it.  Currently pool reports ~300 mh/s.  I'm believe something irreversibly terrible happened during my mucking about.  It's simply something I'll have to live with probably.  At least now I won't feel weird trying to flash it.

Thanks all for the help/encouragement.  If anything changes, I'll post back.

[goxed]

After removing pads and using Arctic MX2 + Firmware / speed 9 + 30s Freezer trick

Code:

 BAJ 1:  max 48C 3.77V | 9.278G/9.365Gh/s | A:67107 R:396 HW: 390 WU:131.2/m