Open Source Bitcoin ASIC miner project that uses 2x BM1387 (Antminer S9)

[Skot]

The APW3 sounds a lot better if you put a Noctua fan in there.

Do you use any kind of LNA/resistor for it or just run off the nominal voltage?

Just off nominal voltage; it should be 12V just like the fan.

[Skot]

Edit: Mystery solved. Thermal pads seem to work fine thermally and insulate the chips electrically, at the same time.

The IC package is plated and electrically connected to local ground, which means that unless there's electrical insulation between the package and heatsink, you can cause shorts and break stuff. We use a 0.5mm 20W/k silicone pad to avoid shorts.

Doesn't silicone have low thermal conductivity? I would think electrically insulating and thermally conductive are tough material properties.

[n0nce]

Edit: Mystery solved. Thermal pads seem to work fine thermally and insulate the chips electrically, at the same time.

The IC package is plated and electrically connected to local ground, which means that unless there's electrical insulation between the package and heatsink, you can cause shorts and break stuff. We use a 0.5mm 20W/k silicone pad to avoid shorts.

Doesn't silicone have low thermal conductivity? I would think electrically insulating and thermally conductive are tough material properties.

Indeed; all the products I could find made from silicone have thermal conductivity in the single digit W/mK range. It seems impressive to me that sidehack has access to pads with 20W/mK. I'll search online some more, later.

[paid2]

Edit: Mystery solved. Thermal pads seem to work fine thermally and insulate the chips electrically, at the same time.

The IC package is plated and electrically connected to local ground, which means that unless there's electrical insulation between the package and heatsink, you can cause shorts and break stuff. We use a 0.5mm 20W/k silicone pad to avoid shorts.

Doesn't silicone have low thermal conductivity? I would think electrically insulating and thermally conductive are tough material properties.

Indeed; all the products I could find made from silicone have thermal conductivity in the single digit W/mK range. It seems impressive to me that sidehack has access to pads with 20W/mK. I'll search online some more, later.

Couldn't we use this ? https://desktop.bg/thermal_pastes-thermal_grizzly-Conductonaut-thermal_grizzly_conductonaut_liquid_metal_1_gramm_TGC001R

73 W/mk

[n0nce]

Couldn't we use this ? https://desktop.bg/thermal_pastes-thermal_grizzly-Conductonaut-thermal_grizzly_conductonaut_liquid_metal_1_gramm_TGC001R

73 W/mk

It's highly electrically conductive, though. And any liquid / paste-like thermal interface will squish out and have spots where the cooler directly contacts the chip, anyway.
The main problem here is finding an interface that is thermally conductive and electrically isolated.

[Skot]

I got the bitaxe max_v2 board working with a 1.8V usbserial adapter and cgminer! not doing too bad either:

Code:

 1: GSF 3NNU10x0: BM1397:01+ 400.00MHz T:400 P:400 (32:16) |  100% WU:100% | 366.4G / 259.3Gh/s WU:3621.9/m

Also some basic drivers for the fan speed and temperature:

Code:

I (1148101) i2c-test: Vcore: 1400 mV
I (1148101) i2c-test: Fan Speed: 5004 RPM
I (1148101) i2c-test: Chip Temp: 50.88 C

A DIY miner hacker from twitter came on the discord to tell us he has made (CPU) mining firmware for the ESP32! https://github.com/BitMaker-hub/ESP32_NerdMiner

[n0nce]

I got the bitaxe max_v2 board working with a 1.8V usbserial adapter and cgminer! not doing too bad either:

Code:

 1: GSF 3NNU10x0: BM1397:01+ 400.00MHz T:400 P:400 (32:16) |  100% WU:100% | 366.4G / 259.3Gh/s WU:3621.9/m

Also some basic drivers for the fan speed and temperature:

Code:

I (1148101) i2c-test: Vcore: 1400 mV
I (1148101) i2c-test: Fan Speed: 5004 RPM
I (1148101) i2c-test: Chip Temp: 50.88 C

A DIY miner hacker from twitter came on the discord to tell us he has made (CPU) mining firmware for the ESP32! https://github.com/BitMaker-hub/ESP32_NerdMiner

That sounds awesome! Great job with that hardware; from what I picked up, it is mostly fine except a small change on the PCB, right?
I finally got an ESP32-S3 dev board in my hands and will get my Bitaxe (older revision without onboard ESP chip) soldered soon^TM. Then I'll get to work on that firmware.

[hZti]

Couldn't we use this ? https://desktop.bg/thermal_pastes-thermal_grizzly-Conductonaut-thermal_grizzly_conductonaut_liquid_metal_1_gramm_TGC001R

73 W/mk

It's highly electrically conductive, though. And any liquid / paste-like thermal interface will squish out and have spots where the cooler directly contacts the chip, anyway.
The main problem here is finding an interface that is thermally conductive and electrically isolated.

There are many reviews on this topic on youtube, where they compare liquid metal to different thermal paste. In general the answer is that a very good thermal paste will give you a similar (slightly worse) result as the liquid metal. The thermal past will however be way more easy to use and not short your circuit or do other bad stuff. Liquid metal may be an option if you want to build just one or two of the miners for yourself, but it would never be an option for a larger scale production.
See videos like this: https://www.youtube.com/watch?v=5a9VcG4j8-k

[Skot]

There are many reviews on this topic on youtube, where they compare liquid metal to different thermal paste. In general the answer is that a very good thermal paste will give you a similar (slightly worse) result as the liquid metal. The thermal past will however be way more easy to use and not short your circuit or do other bad stuff. Liquid metal may be an option if you want to build just one or two of the miners for yourself, but it would never be an option for a larger scale production.
See videos like this: https://www.youtube.com/watch?v=5a9VcG4j8-k

The issue with paste is we can't guarantee it's completely insulating the heatsink from the ASIC.

What about those old school mica heatsink pads? How's their thermal conductivity? I googled around and couldn't find any numbers.

[hZti]

The issue with paste is we can't guarantee it's completely insulating the heatsink from the ASIC.

What about those old school mica heatsink pads? How's their thermal conductivity? I googled around and couldn't find any numbers.

This is what the website says of thermal grizzly:

Is Thermal Grizzly thermal paste electrically conductive?
"Conventional" Thermal Grizzly thermal paste, such as Aeronaut, Hydronaut, Kryonaut and Kryonaut Extreme is not electrically conductive. Only Thermal Grizzly Conductonaut (liquid metal) is electrically conductive.

From my own experience I can say that there is no issue with the electrical conductivity. I have done some custom water-cooling project, where I basically drowned some small components in thermal paste because it was the only possibility. I never had any issues form it. I can however not say what kind of thermal past it was. It would in general at least be an option that could be tried out.

Or do you think the metal of the cooler will short out the asic?

If this is the issue then maybe the best solution is indeed a thermal pad.The best ones should be: https://kriticalpads.com
However they are expensive and compared to paste pretty bad. (20 W/mK)

[n0nce]

Or do you think the metal of the cooler will short out the asic?

That's the issue all along, indeed.

These chips expose their local ground on the top. Since the grounds have a voltage potential between them though, a common heatsink would allow undesired current flow, i.e. magic smoke.

If this is the issue then maybe the best solution is indeed a thermal pad.The best ones should be: https://kriticalpads.com
However they are expensive and compared to paste pretty bad. (20 W/mK)

20W/mK is good. Sidehack uses that, too, as quoted by me earlier.

[hZti]

What could also be worth to check out is to coat the cooler with a non conductive coating:

What Does Non-conductive Coating Mean?

Non-conductive coatings are specialized coatings that limit the transfer of heat or electricity to the metal substrate with the coating applied. Most non-conductive coatings are organic in nature and achieve their insulative properties by being devoid of conductive metals such as copper, zinc and nickel. Classes of non-conductive coatings include:

Polyurethanes
Epoxies
Ceramic coatings
Polymers
A non-conductive coating may also be known as an insulating coating or insulation coating.

It will obviously mess with the cooling, but the thermal paste has such a high conductivity, that even a 50% decrease will still make it better and way cheaper than the thermal pad.

[TalkativeCoin]

The APW3 sounds a lot better if you put a Noctua fan in there.

Do you use any kind of LNA/resistor for it or just run off the nominal voltage?

Just off nominal voltage; it should be 12V just like the fan.

Thanks! I was looking maybe to use a bit less voltage to quiet it down, without too much of a performance hit/compromise.

[hZti]

Thanks! I was looking maybe to use a bit less voltage to quiet it down, without too much of a performance hit/compromise.

You can simply put in a fan that has a lower RPM rating. I use for example a noisblocker in my APW3 and it is not possible to hear it. However the cooling ability is much smaller than the original Fan. This only works, since I use less power than the APW3 can supply.

[TalkativeCoin]

Thanks! I was looking maybe to use a bit less voltage to quiet it down, without too much of a performance hit/compromise.

You can simply put in a fan that has a lower RPM rating. I use for example a noisblocker in my APW3 and it is not possible to hear it. However the cooling ability is much smaller than the original Fan. This only works, since I use less power than the APW3 can supply.

Yes, that is an option too, but Noctua's fans usually come with Low Noise adapters (basically an in-wire resistor that lowers the voltage to 7V or 5V), so I was thinking maybe trying that one out and seeing if there's a noticable performance loss.

[hZti]

Yes, that is an option too, but Noctua's fans usually come with Low Noise adapters (basically an in-wire resistor that lowers the voltage to 7V or 5V), so I was thinking maybe trying that one out and seeing if there's a noticable performance loss.

I can tell you that the airflow with a slow standard fan will be only a slightly fraction of that of the original fan. You should decide for yourself if you are willing to take the risk of overheating. I don"t actually know if there is a good overheating protection built in.
I use my APW3 until around 1.000 Watts with the slower fan, but it depends a lot also on room temperature and different factors.

[TalkativeCoin]

Yes, that is an option too, but Noctua's fans usually come with Low Noise adapters (basically an in-wire resistor that lowers the voltage to 7V or 5V), so I was thinking maybe trying that one out and seeing if there's a noticable performance loss.

...but it depends a lot also on room temperature and different factors.

Definitely! I'll play around with the setup and see what's the best optimum. Thanks for the input!

[Skot]

I got a proper INA260 from Mouser, and combined with getting the correct endianness in the driver, I can now get Voltage, Current and Power measurements from the bitaxe! This test code is in the ESP-Miner `i2c_test` branch https://github.com/skot/ESP-Miner/tree/i2c_test

Code:

0: GSF 3NNU10x0: BM1397:01+ 400.00MHz T:400 P:400 (32:16)  |  100% WU:100% | 363.1G / 264.0Gh/s WU:3687.4/m

I (2345491) i2c-test: Fan Speed: 3508 RPM
I (2345491) i2c-test: Chip Temp: 54.38 C
I (2345491) i2c-test: Current: 2013.75 mA
I (2345491) i2c-test: Voltage: 4911.25 mV
I (2345491) i2c-test: Power: 9980.00 mW
I (2345501) i2c-test: Vcore: 1367 mV

I've also fixed all of most all of the HW bugs found in the bitaxe `max_v2` rev and created a new `max_v2.2` branch. https://github.com/skot/bitaxe/tree/max_v2.2

[n0nce]

I got a proper INA260 from Mouser, and combined with getting the correct endianness in the driver, I can now get Voltage, Current and Power measurements from the bitaxe! This test code is in the ESP-Miner `i2c_test` branch https://github.com/skot/ESP-Miner/tree/i2c_test

Code:

0: GSF 3NNU10x0: BM1397:01+ 400.00MHz T:400 P:400 (32:16)  |  100% WU:100% | 363.1G / 264.0Gh/s WU:3687.4/m

I (2345491) i2c-test: Fan Speed: 3508 RPM
I (2345491) i2c-test: Chip Temp: 54.38 C
I (2345491) i2c-test: Current: 2013.75 mA
I (2345491) i2c-test: Voltage: 4911.25 mV
I (2345491) i2c-test: Power: 9980.00 mW
I (2345501) i2c-test: Vcore: 1367 mV

I've also fixed all of most all of the HW bugs found in the bitaxe `max_v2` rev and created a new `max_v2.2` branch. https://github.com/skot/bitaxe/tree/max_v2.2

These are some awesome updates, great work Skot! Small personal update from my side: I am going to try hot-air soldering my old bitaxe board to be able to write ESP firmware for it. If that doesn't work, I will need some more time to acquire and modify a small oven for soldering. I could still already work on the Stratum part, though.

If we're confident that the max_v2.2 hardware is correct, how about creating a new thread? Such that other interested people can find it?
I'm thinking something like: [In Development] Open Source BM1397 Bitcoin ASIC miner project with a nice start post that contains all information to get started and build it.

[sirmuhamad1]

Hi
I want Register-Map for BM1397 & Bm1398 & BM 1387
If this chipss supoort verison rolling 16
PLZ help