I feel that you are not particularly good at electronic equipment, I think you can buy a second-hand S19 series hydro machine for mining, it will be very quiet with liquid cooling equipment

Does anyone know what kind of process the firmware will read the temperature sensor chip model? I don’t know why the ANT 17 series mining machines have the problem of reading the temperature sensor failure. I am a little confused whether it is a software problem or a hardware problem.

Intel uses BGA-packaged asic chips for BTC mining. In Shenzhen, China, some factories have already produced finished products, but intel finally abandoned the asic chip project. Compared with bitmain’s QFN package, intel’s flip-chip BGA may It does not have the advantages of scale and cost. In TSMC, the 14nm process chip needs about 3 million US dollars for one tape-out, the 7nm process chip needs 30 million US dollars for one tape-out, and the 5nm process chip costs 47.25 million US dollars for one tape-out. Not any private person can do this. It is most likely to design a DIY finished product. FPGA can be used to realize basic mining, but the computing power is not enough. It must be as strong as ASIC

S19 PRO 110T, the lowest energy efficiency ratio I tried is as low as 25.66J. At 1976W, the running computing power is 77Th. I guess the post on the top floor is due to the communication problem between his control board and the power supply. The power sent by the control board The voltage difference between the voltage instruction and the actual power supply voltage is too large, so some netizens say that it can be stabilized after replacing the control board. The lowest maintenance cost of this problem is to constantly try various firmware, or buy another control board to try

I mentioned the firmware I use, like you guys mentioned braiins+, I didn't send any firmware to any of you (This firmware is private and currently only available for internal testing and is not open to the public), I just stated my opinion on the issue, Don't think that whoever discusses with you is trying to sell you products

In addition, the problem we discussed is because the new machine of the S19 series in the later period cancels the pic control chip and the switch MOS tube, so there is no way to control the voltage and power consumption of the power board independently by switching the switching frequency of the MOS, and there is no way to independently control the power consumption of the power board. Turn off or turn on a hashboard

This problem is nothing more than finding a hashboard with a better physique, which can work at a lower voltage threshold, or replace those chips with poor physique. The official firmware has many restrictions on the use conditions, while the third-party firmware will open These restrictions may also shield certain non-working chips (there is no technical obstacle), or the firmware adaptively increases the voltage threshold a little bit, you can find a few more firmware to try

Hello, my friend, I think your problem may be that the control board has lowered the voltage in low power mode, and a chip on this chain cannot work at low voltage due to manufacturing problems. Next, so the problem you mentioned will appear

Low power consumption is nothing more than two conditions, one is to reduce the frequency, and the other is to reduce the power supply voltage of the PSU. The official firmware may not have much experience in this area, while the third-party firmware will have richer options. For example, a certain chip cannot If it works at low voltage, then we shield this chip so that it does not work, or increase the voltage of the PSU without increasing the frequency, at least the chain can work without causing a loss of computing power

For example, the hulkos I use has such a function

Thank you very much my friend, I will try talkimg.com

This firmware is developed by us privately. It is currently in the public testing stage. This is the built-in maintenance mode of the firmware. It only supports testing one computing power board at a time, similar to the official maintenance fixture, so my log is only testing one computing power board, this board is a normal operating computing power board

When I was analyzing this problem, I kept trying to start the test program, and I often reproduced the problem that the temp_sensor could not be read. I analyzed the log and found that it would initially read the component model of the temp_sensor, and the hardware used was NCT218. , if it is read correctly, then he will work normally, if the identified component model is inaccurate or not read, then he will replace the identified model with the default TMP451, then he will not be able to calibrate Temp_sensor Get the correct parameters, resulting in a temperature reading error and an error

This test log is the result of testing under the same machine
If it is read correctly, it will prompt:
temp_sensor_type=26 is NCT218

If it is not read correctly, then he will prompt:
temp_sensor_type=0 then he will set this temp_sensor as TMP451 by default

NCT218=type26
TMP451=type0

Also, our firmware does not depend on the PIC firmware in the EEPROM chip attached to the PIC chip. We directly issue the operating frequency of the asic chip and the power supply voltage of the PSU from the hulkos firmware, which completely bypasses the problem of the PIC, and the frequency and voltage are dynamically adjusted, so even if the PIC chip is not flashed with firmware, or our three-chip computing power The PIC information of the board is not uniform, and our hashboard can still work normally, so it has nothing to do with the PIC, which simplifies the work difficulty of on-site operation and maintenance personnel. To a certain extent, you can understand that hulkos automatically sends the PIC program to the hashboard to run

My friend, thank you for your suggestion, I know this company, but I don’t need their components, I have my own channel to get the goods, their prices, you know, I have a lot of scrap machines that I can dismantle Chip, I have carefully studied the maintenance instructions of all series of 17 a long time ago. I even have the original circuit diagram. I have a headache for this damn temp-sensor problem. I agree with your point of view through analysis, it’s his mother There is a problem with the computing power chip, and there is a problem with the welding process of Century Cloud Core. They even sprinkled rosin solution on the 17 series computing power board and re-soldered it. Maybe the board you saw with UV glue is that batch.

In addition, I am using HulkOS, and his logs are not the same as Bitmain’s official firmware. I don’t know how to upload attachments to this forum. It seems that it is not good to post large sections of logs.

Try this to see if it can be opened. This is the log of the same machine. It is normal to read the temperature sensor as NCT218 at the beginning. If it cannot be read later, it will default to TMP451 and start reporting an error.

http://m.qpic.cn/psc?/V53cp4R00xIcKH0EHAvf1OQprC2kRuqc/6tCTPh7N*X6CBkvkDvKlZW7IcKWUta.hnJDG0METJ90Ar0e6SMIWli43jVioiJGuwTqZ4ca3US4144ps1sKKAzUbfzWRp.UfSI4nkFNwh7w!/b&bo=OARSBAAAAAABJ2o!&rf=viewer_4&t=5

http://m.qpic.cn/psc?/V53cp4R00xIcKH0EHAvf1OQprC2kRuqc/6tCTPh7N*X6CBkvkDvKlZTUWsnYdM4Opujj.8Vzl0orUyQ4gUKypHmsJHlflbPfZ0.AxuI4iaZom9zyEezL6eGY2xnfXoy*ztOCOijPtreo!/b&bo=UwQ4BAAAAAABJ2s!&rf=viewer_4&t=5

I have carefully read all the replies under that post, and I have also carefully studied bitmain’s official maintenance instructions and the circuit diagram of S17. The 17 series hosts do not only have a model of TMP451. Among the 17 series models I have encountered so far, there are at least NCT218 and TMP451 two models, I found through the detailed log of my third-party firmware that he will first read the register of the temperature sensor, then obtain the device type of the sensor, that is, the model of the sensor, and then read the temperature correctly, such as The sensor of NCT218 is on the PCBA, and when the firmware cannot read NCT218, it will default to the device model TMP451, and then when it continues, the firmware will report that the reading of the temperature sensor failed

After analyzing the circuit diagram, I found that the temperature sensor has 4 wires connected to the corresponding power chip, which are SCL, SDA, TEMP_P, and TEMP_N. The BJT of the PNP transistor in the chip is read through the latter two signal wires. The temperature of the junction, and then the SDA transmits this data to the rx and TX on the ASIC chip through the I2C bus, and then transmits it to the PIC control chip, and then the PIC chip transmits it to the control board, and the temperature sensor is uploaded to the Whether the signal of the upper end machine is accurate or not becomes the key to whether the temperature sensor can correctly identify

Hi, everyone, I am a miner from China, I have a large number of T17+ and S17+, they often have the problem of temperature sensor reading failure, I want to get your help

I often encounter the problem that a certain computing power board reports that the temperature sensor fails, causing the board to fall off and fail to operate

I found that the 17 series mining machine has two sensors, one is TMP451, the other is NCT218, and the operating voltage range of these two chips is not the same, the former is 1.7~3.6V, the latter is 1.4~2.75V, And when I use the maintenance mode of the third-party firmware, it will show the NCT218 chip when it is normal at the beginning of the test, and it will show the TMP451 chip when there is a problem and the temperature sensor is lost. I don’t know if there is any connection between them.