Asic.to Custom Firmware - Optimize Performance & Efficiency S19/S21/T21

[DickyDck]

Yes, I was looking for such thing! Is it really without dev fee? Can you tell me where to get it? Thanks in advance

I could be mistaken, but doesn't Tazers mod (this forum) allow you to adjust frequency with no dev fee? I'm pretty sure Tazer is doing this simply because he is a cool dude (or chick?) and to stuff it in Bitmains face.

[Shad0wSmurf]

Code:

NONCE2_AND_JOBID_STORE_ADDRESS
#define JOB_ID_OFFSET					(0x0/sizeof(int))
#define HEADER_VERSION_OFFSET			(0x4/sizeof(int))
#define NONCE2_L_OFFSET					(0x8/sizeof(int))
#define NONCE2_H_OFFSET					(0xc/sizeof(int))

This is in btmc5 driver and is the masks the boost uses if you can figure out the usage

[Andartis]

I could be mistaken, but doesn't Tazers mod (this forum) allow you to adjust frequency with no dev fee? I'm pretty sure Tazer is doing this simply because he is a cool dude (or chick?) and to stuff it in Bitmains face.

Yes it does, but not by each chip individually. Do you have a no dev fee firmware that can modify the chips individually?

[DickyDck]

I missed the fine details of that post, the part about "each chip" individually. Wouldn't that be a pain to do and monitor? Other than finite control, what benefit would that offer? I'm not trying to be argumentative, I'm legitimately asking if there is a benefit to this, cause if there is I might be interested.

[House Xylander]

Yes, I was looking for such thing! Is it really without dev fee? Can you tell me where to get it? Thanks in advance

If memory serves it was shared on these forums, I try looking when I get home and have a full PC at hand.

[House Xylander]

I missed the fine details of that post, the part about "each chip" individually. Wouldn't that be a pain to do and monitor? Other than finite control, what benefit would that offer? I'm not trying to be argumentative, I'm legitimately asking if there is a benefit to this, cause if there is I might be interested.

I ran it through the Auto-Tune first and found the range that most of the chips showed green on the manual screen. After that, it was just a few chips that would show "red" or "crimson" (if the voltage and Freq where to high for the chip I would drop it 25-50m). When all the chips showed green they would typically have a value of 63-78. Sometimes during a hash a few would turn yellow and then back to green.

Just a hunch but I'm guessing that the red and crimson chip where the ones causing the most  HW errors when setting it over 637m. That is why I listed the range 625-675m

[Andartis]

Some say it is not significant tuning the chips individually. I dont think so and I would like to try it. As House Xylander said, there is an interface, that checks the chips and shows them all coloured from green, to yellow, to red. Then you can pitch up the frequency for all green chips or each one by yourself etc.
I think you can decrese the HW errors by letting the poor chips hash slower and the good ones faster. That should increse the effectivity.

But unfortunately the only firmware I have found so far that supports this, has a dev fee included. So if someone has one without dev fee, that would be great. although now it would be interesting to get a firmware, that also supports asicboost.

[taserz]

Some say it is not significant tuning the chips individually. I dont think so and I would like to try it. As House Xylander said, there is an interface, that checks the chips and shows them all coloured from green, to yellow, to red. Then you can pitch up the frequency for all green chips or each one by yourself etc.
I think you can decrese the HW errors by letting the poor chips hash slower and the good ones faster. That should increse the effectivity.

But unfortunately the only firmware I have found so far that supports this, has a dev fee included. So if someone has one without dev fee, that would be great. although now it would be interesting to get a firmware, that also supports asicboost.

It's not worth using it the dev fee is like 6% and the gain is little to none.

[Hockeybum]

I have limited results as I only have access to a few machines while away from my data center.  I was doing power measurements at the power supply tonight on 1 machine and only letting it hash for 10 minutes before taking readings.  Machine is an April 2017 delivery Bitmain S9 13.5 Th/s, power supply is a Bitmain APW3+-12-1600-A3, also delivered during that time.  Voltmeter and Ammeter are both Fluke.  Again I want to stress that this is only 1 machine and very limited testing since I can't wait all day to take readings.  This power supply is about 92.5% efficient at 100 amps according to this post https://bitcointalk.org/index.php?topic=5045057.0. Numbers are right around a 12.5% power savings and an overall efficiency improvement of 14.5%.  I've only updated 3 machines so far and have have hundreds to go as well.  Hoping to save about $2,500 in power costs per month with this firmware.

Nov 17 Firmware              12.15 vdc  110.6 amps  1,344 watts  13.226 th/s avg  101.6 w/th    at the wall estimated ~1,453 watts
LPM Patch and checked     12.16 vdc  96.7   amps  1,176 watts  13.554 th/s avg  86.8  w/th    at the wall estimated ~1,271 watts
LPM Patch and unchecked  12.15 vdc  110.5 amps  1,343 watts  13.424 th/s avg  100.0 w/th   at the wall estimated ~1,452 watts

[Andartis]

Thats good!
But did we find out what happens, if you replace bmminer again with taserz version?
I saw a drop in Watts but also in hash. I only upgraded without replacing the bmminer again.

[Hockeybum]

Same 10 minute limited test.  LPM makes no difference in BMMINER880 firmware, it would probably show closer numbers with a longer test.

Code:

FIRMWARE      LPM      FREQ  VOLTS   AMPS    WATTS  GH/S RT GH/S AVG    W/TH

BMMINER 880 NO LPM	550	12.16	   91.2	1,109	   11775	11530	96.2
BMMINER 880 LPM	550	12.16	   91.9	1,118	   11764	11815	94.6
BMMINER 880 LPM	625	12.15	   102.5	1,245	   13226	13001	95.8
BMMINER 880 LPM	750	12.15	   124	        1,507	   14184	14370	104.8
BMMINER 880 NO LPM	750	12.15	   123.6	1,502	   14140	14176	105.9
BMMINER 880 NO LPM	600	12.15	   100.2	1,217    12729	12892	94.4

[Shad0wSmurf]

Same 10 minute limited test.  LPM makes no difference in BMMINER880 firmware, it would probably show closer numbers with a longer test.

[...]

Asic boost isn't enabled within rocketships bmminer.. So your lpm tests are completely irrelevant

[Hockeybum]

I did mention that in the above post.  The test was relevant to me so I could confirm that LPM had no effect on the software (which is something I had already assumed and wanted to verify).  If anyone wants to argue that the 10 minute test is irrelevant I would agree with them.

[Dr_Victor]

By the way the last official firmware for S9 gives minus 200 watt with the same (default) speed!

[Andartis]

By the way the last official firmware for S9 gives minus 200 watt with the same (default) speed!

You mean autotune?

[Hockeybum]

A few people have reported here https://bitcointalk.org/index.php?topic=5055138.60 about a decaying hash rate after some period of time with the LPM patch from Bitmain so I've decided to do a 12 hour test.  Not sure when I'll be here to read the measurements but I will try to do at least a 3 hour, 6 hour, and 12 hour.  So far 2 hours in I'm reading 13.607 th/s average at 1,177 watts at the PSU, assuming close to 1,273 watts at the wall.

[Norm MacDonald]

A few people have reported here https://bitcointalk.org/index.php?topic=5055138.60 about a decaying hash rate after some period of time with the LPM patch from Bitmain so I've decided to do a 12 hour test.  Not sure when I'll be here to read the measurements but I will try to do at least a 3 hour, 6 hour, and 12 hour.  So far 2 hours in I'm reading 13.607 th/s average at 1,177 watts at the PSU, assuming close to 1,273 watts at the wall.

Those are some nice Watt numbers. I don't care much where I get my enhanced efficiency, ie;"better profitability", whether it's from drawing less current or Higher Hash rate or both. Frankly, one could argue that dropping the draw at the wall will be less stressful on the units so will not only jump your profitability, but i would assume create more longevity and stable mining.

[MrMaxwell]

I am patiently awaiting the new release to allow over clocking the new Bitmain Asic Boost LPM firmware. Thanks for all you do taserz!

[taserz]

Code:

NONCE2_AND_JOBID_STORE_ADDRESS
#define JOB_ID_OFFSET					(0x0/sizeof(int))
#define HEADER_VERSION_OFFSET			(0x4/sizeof(int))
#define NONCE2_L_OFFSET					(0x8/sizeof(int))
#define NONCE2_H_OFFSET					(0xc/sizeof(int))

This is in btmc5 driver and is the masks the boost uses if you can figure out the usage

I am going to see if I can incorporate it.

Also, I just realized someone is charging money for a tool to deploy bmminer to a bunch of s9's. When I have time I will throw a program together to do this for you guys. Maybe try to learn a new .net language for shits and giggles to make it fun since it should only take about 20 minutes to make anyways.

[Andartis]

[...]

Also, I just realized someone is charging money for a tool to deploy bmminer to a bunch of s9's. When I have time I will throw a program together to do this for you guys. Maybe try to learn a new .net language for shits and giggles to make it fun since it should only take about 20 minutes to make anyways.

Really? Who pays for that? You could also just write a batchfile with the commands. Its like 4 lines of code