The Chili – 30+GH/s BFL based Bitcoin Miner Assembly

[MrTeal]

why use cgminer version 3.4.0
and not the last cgminer 3.6.6.?

Because 3.4.0 works and I haven't bothered to update.

[1715294256]

1715294256

[1715294256]

1715294256

[1715294256]

1715294256

[1715294256]

1715294256

[1715294256]

1715294256

[jelin1984]

Ok I can glue at bottom a heatsink for vrms?
Or make problem at board?

[Keefe]

The bottom heatsink just has some Kapton tape and a little thermal grease on it, while the top cooler is just using thermal grease. Using this setup there is no board flex at all. The thermal grease is just the Antec branded stuff that's inside the box.

Isn't kapton tape a thermal insulator? Are you using the second heatsink purely as a symmetrical backplate, not to remove heat?

[MrTeal]

The bottom heatsink just has some Kapton tape and a little thermal grease on it, while the top cooler is just using thermal grease. Using this setup there is no board flex at all. The thermal grease is just the Antec branded stuff that's inside the box.

Isn't kapton tape a thermal insulator? Are you using the second heatsink purely as a symmetrical backplate, not to remove heat?

Well, Kapton tape is a thermal insulator, but it's also very thin and is a good electrical insulator. There are exposed pads on the bottom of the PCB where surface mount caps could have been added if we found it improved performance, so you have to make sure they don't get shorted out. Kapton actually works reasonably well, I used to use it as an electrical insulator for high power transistors in audio amps between the transistor and the heatsink. You'd probably get a little better performance with a thermal pad, but I didn't have any laying around at home (my roll is at work) so it was about expediency as opposed to optimization.

As you say, it's just to get good pressure on the ASICs without board flex. The fan on the bottom isn't even plugged it.

[daemonfox]

Here are two more pics of my chili

TIMs (those white spacers were just a little too tall so they are not there anymore... they felt solid and prevented the boughing issue but would not allow full force contact when I fired it up.


Back Heatsinks (I also have an 80mm fam blowing over the heatsinks and i run the whole thing standing up on the coollers lond side rather than letting it lie flat on stands


At first I was able to get just below 30 GH/s and I let it burn in over night... steady at 29+ all night. I tried to tweak it slowly adding more force via slowly tightening the screws round robin but I only ended up making it worse. I got a temp warning and shut off mining... backed the screws off and tested... repeated, even disassembled, reseated the cooler, got the screws tightened in unison and slowly tweaked them while watching for boughing... and was able to bring it back to near 30 GH/s with no warnings.

Man, I need a solution like what Teal made but sans a second cooler... like a heatsink with screw holes that fit the chili's closest mounting holes.

[ryanb]

any updates on the second batch?

[daemonfox]

Mr Teal,

A few questions, if they have already been answered, I am happy to read the reply if they can be pointed out.

1. Does the Chili respond to the cgminer commands to adjust temp targets... so throttling at 70C can be raised higher?

2. Is there another application or scripts that can be used to monitor the individual engines clock and temps?

3. Was there ever a full detail of the board layout posted... like what everything on the board is and can do, what connectors can handle and are meant for?

4. Is there an API or w/e that will be usable in the future for something like an app that can be a "Chili Monitor" to show all the live details of the board?

Thanks for all your hard work!

[Spieder]

I've been playing around with a chili I bought on ebay with an Evo 212 until my batch 2 boards are done.

I originally booted it up with the Evo 212 and a 120mm fan blowing sideways over the vrms and caps. In bfgminer, cgminer and bfgminer on a raspberry pi the unit runs @ 37-38 ghz @ 72 C - 10% errors.

If I remove the side fan, the unit quickly readjust to 52 C and drops to about 32 ghz. So cooling those vrms is a key piece to achieving higher speeds. I know there's a lot a variability in performance based on chips but I'd tend to think if your cooling option let's you hash at 72 C then you're close to max. If your hashing at 52 C you may have room for improvement.

From the wall with my pi, I'm reading the system is pulling 175 watts at 32 ghz and 209 watts at 37 ghz. So 15% increase in ghz is roughly 25% increase in power.

Now if I could figure out how to get those errors down I'd be happy because my effective is @ 33 ghz. I'd rather run at 33 ghz @ 175 watts with no errors than 38 ghz @ 209 watts with 33 ghz effective.

[daemonfox]

More testing, I have rotated my Accelero TT II 90 degrees so that the VRMs are exposed instead of hiding behind the heatsink.

Now stable at 31 GH/s... I also took the fan I had blowing over the bottom heatsinks and now have it blowing over the top of the VRMs.

Let me just say... it i obvious this board design is not suited to GPU style coolers if you want anything above the 30GH/s mark. The heatsink of the Accelero TT II is too close to the VRMs. A cooling solution like what Mr Teal implemented the other night is ideal.... get the heat dissipation AS MUCH away from the board as you can and you will love the results.

I will probably be spending the $26 to get the two coolers Mr Teal used recently... will report back on that.

EDIT: Up to 32 GH/s stable.

[Spieder]

More testing, I have rotated my Accelero TT II 90 degrees so that the VRMs are exposed instead of hiding behind the heatsink.

Now stable at 31 GH/s... I also took the fan I had blowing over the bottom heatsinks and now have it blowing over the top of the VRMs.

Let me just say... it i obvious this board design is not suited to GPU style coolers if you want anything above the 30GH/s mark. The heatsink of the Accelero TT II is too close to the VRMs. A cooling solution like what Mr Teal implemented the other night is ideal.... get the heat dissipation AS MUCH away from the board as you can and you will love the results.

I will probably be spending the $26 to get the two coolers Mr Teal used recently... will report back on that.

EDIT: Up to 32 GH/s stable.

That's interesting, I would have thought the airflow down on the board would have helped. Even though my chips are running at 72 C, the Evo 212 is only warm to the touch on the base plate. The cooling fins are barely warm at all.

How hot to the touch is the TT II? If it's just warm like the Evo then maybe there's something else going on other than the heat getting pumped onto the board.

I noticed Mr. Teal said he used paste, not pads for his experiment. I'm thinking the pads may also be a large factor. Maybe I'll break out my old thermodynamics book and figure out what effect those pads would have on something that would normally pump 240 watts. I can guarantee that TT II is going to loose that 240 watt rating with another thermal resistor in play.

So I'm thinking it may be less about the TT II pumping hot air on the board and more about the restricted airflow over the TT II with the board behind it.

[daemonfox]

Yes... it is two fold.

Using the pads and basic screw pattern, the board has a boughing issue and can only get so much force without causing the center chips to lose some contact. Add to that the fact that the fins of the TT II are within 1 inch of the VRMs and caps, it makes airflow over them impossible.

Rotating it they way I have now has helped... but the determining factor is still going to be force on the chips... I am hoping to find a heatsink block that has matching screw patterns to the TT II and try that on the back with a pad... then all the torque of the screws is to the haetsinks and zero on the board so full force and no boughing.

[one4many]

Hello MrTeal,

....

• Z-Link Chaining Connections
  Similar to BFL's X-Link interface, this still in development interface allows boards to be chained together and a single USB connection to be used, keeping wiring clutter in check for large data center applications

....
[/list]

How is this feature coming along?
Is it already implemented and how do I daisy-chain several Chili boards together?
How many boards are the limit in one chain?

Thx

      one4many

[daemonfox]

Anyone mining with a Chili on android?

I have an HTC Thunderbolt, rooted, running IGS and I just bought an OTG USB 2.0 cable for it.

I am planning to use it to mine with the chili and just wanted to see what application anyone might be using on android with this?

[merlin3650]

So here's a quick picture of what I've (on an unfortunate post-Halloween sugar bender) named my "Tower of Power" cooler.
Basic setup is two of the $10 Antec Kuhler Shelf coolers, with a 2" #8 screw connecting the two coolers. There's also two press on heatsinks on the VRMs.



The bottom heatsink just has some Kapton tape and a little thermal grease on it, while the top cooler is just using thermal grease. Using this setup there is no board flex at all. The thermal grease is just the Antec branded stuff that's inside the box.

The total cost of this cooling setup is about $26 including the two press-on heatsinks, though the space efficiency is terrible. I think you could get good numbers with just a simple backplate, as long as you are getting good contact with all the dies. My non-RevB (no engine 0) sample board with 7 ASICs mounted gets 32GH/s, and the other production (RevB) Chili I have configured like this is averaging 39.8GH/s, so I think there is a lot of benefit to looking at your cooling solution. Remember that you want your cooling solution to give good force pressing the ASICs to the heatsink (force is the red arrows, BTW) without doing so asymmetrically which will bend the PCB and give bad contact on the center ASICs.

Where did you find them for $10? Everywhere I have looked they are $50 or more?

[MrTeal]

Where did you find them for $10? Everywhere I have looked they are $50 or more?

Direct from the manufacturer.
http://store.antec.com/Product/clearance_items/kuhler-shelf/0-761345-77081-1.aspx

[MrTeal]

Hello MrTeal,

....

• Z-Link Chaining Connections
  Similar to BFL's X-Link interface, this still in development interface allows boards to be chained together and a single USB connection to be used, keeping wiring clutter in check for large data center applications

....
[/list]

How is this feature coming along?
Is it already implemented and how do I daisy-chain several Chili boards together?
How many boards are the limit in one chain?

Thx

      one4many

It's a low priority firmware upgrade, since it's really only useful for a few users (most of whom already have their units set up). We will continue to work on it, but it will likely be a little while before it gets implemented.

[Lucko]

Where did you find them for $10? Everywhere I have looked they are $50 or more?

Direct from the manufacturer.
http://store.antec.com/Product/clearance_items/kuhler-shelf/0-761345-77081-1.aspx

And if you are not US:)

EDIT: I need a proxy service

[MrTeal]

Where did you find them for $10? Everywhere I have looked they are $50 or more?

Direct from the manufacturer.
http://store.antec.com/Product/clearance_items/kuhler-shelf/0-761345-77081-1.aspx

And if you are not US:)

EDIT: I need a proxy service

I'm not sure about Europe, but I actually ordered another 20 of them (and their last of the Kuhler Flow's) and shipping to me in Canada was only $40. Very reasonable.

[agibby5]

Where did you find them for $10? Everywhere I have looked they are $50 or more?

Direct from the manufacturer.
http://store.antec.com/Product/clearance_items/kuhler-shelf/0-761345-77081-1.aspx

Would it be acceptable to use one of these or must you use two like you posted?

[MrTeal]

Where did you find them for $10? Everywhere I have looked they are $50 or more?

Direct from the manufacturer.
http://store.antec.com/Product/clearance_items/kuhler-shelf/0-761345-77081-1.aspx

Would it be acceptable to use one of these or must you use two like you posted?

Using one would be acceptable.