Topic: Experimenting with Jalapeno firmware... - page 18. (Read 62616 times)

Yes indeed - pretty much every mining device out that is quite unreliable at measuring a useful temperature.
So the reported before and after numbers wont necessarily give the the correct answer.

I'll make sure to keep my top down then. geesh, BFL Prudes.

https://forums.butterflylabs.com/announcements/3282-bitforce-sc-firmware-version-1-2-5-a-3.html#post45157


Just thought I would make people aware of that.

oh right, sorry.
I was planing to connect (some of) the sensors (and fets) to the main heatsink with the help of a cooper plate and forgot its not the case with the original board

I couldn't tell you for sure. The temps seem about the same, but they are measured from the PCB and not the chips.

I used a ton of paste and torque. Good enough

The problem also is, that with the time the paste will dry and every gap may cause local overheating and looser the heatsink - higher the chance.
Can you tell the difference in temp with pads (before) and paste (now)?

It has always been my understanding that thermal paste is superior to thermal pads. But, the only reason I had to use thermal paste is because the thermal pads on mine completely fell apart after removing the heat sink about 10 times. I was also worried about the chips not being coplanar and damaging them, but it seems to work. I'm not using much torque when tightening down the heatsink.

It's not going to be such an issue with teh jallies with only 2 chips, you get far better odds of them being near enough coplanar to use thermal paste for an advantage in thermal interface thickness.

With 8 chips under a sink though, it might get a little more critical to keep the pads.

Personally, I'll be checking mine with a straightedge to see how they lie, before considering pad replacement. It's not outside the bounds of possibility that I could consider making custom shims for mine, on a per chip to fit basis.

One thing that could happen is that for a chip that meets the sink at a slight angle, the paste fillets it, and seems good enough at first, but as it dries out in service it creeps further up the crack leaving half the chip out of thermal contact.

If I were putting together my own BFL based hardware, I miiight consider pressure mounting the ASICs to a chunk of copper, with a thermal epoxy (Which BTW suck as much as any other TIM) then mounting them to the board after, but that could turn into a huge PITA.

If the board were clear of other components though, you could shim each chip, then lap the tops of all the shims coplanar but that's probably impractical.

Also the less thermal interfaces the better, but also if you can get them thinner than they would be otherwise, that's good too, swings and roundabouts.

I think 0.5 should be fine, after all many people are OK with paste, so it seems the chips are well aligned.

When i get my Jalapeno i will be able to measure the original pads, but will still try the 0.5 first even i have 1.0, 1.5 and 3.0 at hand too.

My idea was to somehow connect the FETs to the main heatsink too (hence the additional pad sizes and some cooper plates collected), but it seems they are adding dedicated heatsinks to them now ... will see

Thinner (0.5) is better for thermal conductivity, but I never used a pad that thin.
I am a bit worry about its...whats the word...flimsiness ?

Ok you think will better 1mm or 0.5mm?

I ordered this one: http://www.frozencpu.com/products/17499/thr-181/Fujipoly_Ultra_Extreme_System_Builder_Thermal_Pad_-_60_x_50_x_05_-_Thermal_Conductivity_170_WmK.html?tl=g8c487s1288

Would that be good enough?

Don't forget the thickness - pads are thicker, so even with higher thermal transfer figure, they are worse than a few microns of paste.
In the Jalapeno case, the pads are preferable, because you may damage the chips when tightening the heatsink with just paste and if the two chips are not perfectly aligned the paste may even not be enough to make good contact with the heatsink

Yes, I was being slightly too hasty Thanks for responding.

The pads used have a thermal transfer figure in the teens (14?) paste is rarely even 10 ...

Edit: from BFL when I was there, the pads heat dissipation number was '14' (whatever that means) where as thermal grease is like 7, 8 or 9 (9 is really expensive)
That is of course information told to me by BFL, but it was a topic of discussion one night at dinner and I'm sure it's correct.

3.2.1 isn't the newest cgminer.

3.3.1 is.

2013-06-18 Was when I added the 1.2.* firmware support.

3.3.0 was even after that.

Thermal paste.

Pads are an easier solution, but they are also slightly worse at transfering heat from the chip to the heatsink.

Paste gets "in there" in the small grooves on the heatsink and transfer heat better.

However, often pads are more than enough. The only reason to change to a thermal paste, would be to try and squeeze the temperature down by a few degrees.

Final what is better thermal pad or thermal paste?

Strange, on newest cgminer for windows, I got no visible text errors when mining.

Now I'm getting this on MinePeon (cgminer 3.2.1):
[2013-07-04 13:11:17] BAJ0: incorrect data count (3) will use -6 instead from (INPROCESS:10x00)
 [2013-07-04 13:11:17] BAJ0: invalid nonce (INPROCESS:10x00) will try to process anyway
 [2013-07-04 13:11:17] hex2bin str truncated
 [2013-07-04 13:11:17] BAJ0: invalid nonce - HW error
 [2013-07-04 13:11:17] BAJ0: incorrect data count (7) will use -6 instead from (COUNT:20x00)
 [2013-07-04 13:11:17] BAJ0: invalid nonce (COUNT:20x00) will try to process anyway
 [2013-07-04 13:11:17] hex2bin str truncated

Tons of HW errors now.

Mysterious.

Solved: Needed to update MinePeon to get a newer cgminer.

Read here for more information: https://bitcointalk.org/index.php?PHPSESSID=ri7gap7378k7nj53su6eh48fo1&topic=236875.msg2525699#msg2525699