Author

Topic: At what clock rate does an AntMiner S5 mine with peak efficiency? (Read 1802 times)

legendary
Activity: 4256
Merit: 8551
'The right to privacy matters'
Efficiency doesn't change much with clock speed because although dropping the clock reduces power consumption, it also reduces hashrate, and both are approximately linear effects.  The way to boost power efficiency is if you could also undervolt- something that apparently does not work well at all on the S5.

The BM1384 chip on the S5 theoretically allows significant undervolt but even apart from the difficulty of finding high efficiency 8-10V power (required as the S5 is a string miner) the S5 has something which evidently just doesn't work if you drop the voltage much.  Sidehack has said he suspects the (chained) clock line, at any rate some of the chained signals and the related I/O voltages are most likely suspect.  As far as I know no one has taken the time to mod an S5 to be able to undervolt or could tell you what is going wrong for certain.

--
novak

My results are 0.44w/GH par 11.1v
Under 11v. Miner starts but dies after a few minutes.
At 10v and lower, it doesn't start hashing., but you have access to the web interface.
Stock efficiency with the same PSU is 0.54w/GH


I suspect the s-6 will allow good numbers which is why they did not alter the s-5 to allow for a 10 volt psu to attach to the s-5 and work.
full member
Activity: 173
Merit: 100

My results are 0.44w/GH par 11.1v
Under 11v. Miner starts but dies after a few minutes.
At 10v and lower, it doesn't start hashing., but you have access to the web interface.
Stock efficiency with the same PSU is 0.54w/GH


Yeah I don't think I've heard anyone claiming they would run under 10.8V.  You have access to the web interface because it doesn't affect the controller it's probably BM1384 chip I/O of some kind or other.

We have a BM1384 on our breakout board running at 0.625V right now, which would be 9.375V into an S5.  It's not a chip limitation, it's a miner design limitation.

--
novak

Good to know that it's the design. Maybe a simple trick could allow lower voltage to the S5, then.
I'll have a look at an S5 board. I have one to RMA anyway, so it'll be removed from the unit today.

BTW, if the design wasn't good for lower voltage, it could have been upgraded for later batches, since it seems minor changes were made like changing the data conector from 16 to 18 pins...

We have poked at them a bit but done no real work to try to underclock.  I wish they had fixed it later on, I'm fairly sure it is not that big of an issue- most likely some of their chained signals don't work perfectly if the input voltage is too low.  Since the comms on the S5 are chained if one chip in a string drops out they all do.

--
novak
hero member
Activity: 840
Merit: 1000

My results are 0.44w/GH par 11.1v
Under 11v. Miner starts but dies after a few minutes.
At 10v and lower, it doesn't start hashing., but you have access to the web interface.
Stock efficiency with the same PSU is 0.54w/GH


Yeah I don't think I've heard anyone claiming they would run under 10.8V.  You have access to the web interface because it doesn't affect the controller it's probably BM1384 chip I/O of some kind or other.

We have a BM1384 on our breakout board running at 0.625V right now, which would be 9.375V into an S5.  It's not a chip limitation, it's a miner design limitation.

--
novak

Good to know that it's the design. Maybe a simple trick could allow lower voltage to the S5, then.
I'll have a look at an S5 board. I have one to RMA anyway, so it'll be removed from the unit today.

BTW, if the design wasn't good for lower voltage, it could have been upgraded for later batches, since it seems minor changes were made like changing the data conector from 16 to 18 pins...
legendary
Activity: 1232
Merit: 1000
The efficiency doesn't change as much with the frequency settings as one might like.  I think it is more useful as a heat management tool.

I agree heat and noise at the higher frequencys
full member
Activity: 173
Merit: 100

My results are 0.44w/GH par 11.1v
Under 11v. Miner starts but dies after a few minutes.
At 10v and lower, it doesn't start hashing., but you have access to the web interface.
Stock efficiency with the same PSU is 0.54w/GH


Yeah I don't think I've heard anyone claiming they would run under 10.8V.  You have access to the web interface because it doesn't affect the controller it's probably BM1384 chip I/O of some kind or other.

We have a BM1384 on our breakout board running at 0.625V right now, which would be 9.375V into an S5.  It's not a chip limitation, it's a miner design limitation.

--
novak
hero member
Activity: 840
Merit: 1000
Efficiency doesn't change much with clock speed because although dropping the clock reduces power consumption, it also reduces hashrate, and both are approximately linear effects.  The way to boost power efficiency is if you could also undervolt- something that apparently does not work well at all on the S5.

The BM1384 chip on the S5 theoretically allows significant undervolt but even apart from the difficulty of finding high efficiency 8-10V power (required as the S5 is a string miner) the S5 has something which evidently just doesn't work if you drop the voltage much.  Sidehack has said he suspects the (chained) clock line, at any rate some of the chained signals and the related I/O voltages are most likely suspect.  As far as I know no one has taken the time to mod an S5 to be able to undervolt or could tell you what is going wrong for certain.

--
novak

My results are 0.44w/GH par 11.1v
Under 11v. Miner starts but dies after a few minutes.
At 10v and lower, it doesn't start hashing., but you have access to the web interface.
Stock efficiency with the same PSU is 0.54w/GH
full member
Activity: 173
Merit: 100
Efficiency doesn't change much with clock speed because although dropping the clock reduces power consumption, it also reduces hashrate, and both are approximately linear effects.  The way to boost power efficiency is if you could also undervolt- something that apparently does not work well at all on the S5.

The BM1384 chip on the S5 theoretically allows significant undervolt but even apart from the difficulty of finding high efficiency 8-10V power (required as the S5 is a string miner) the S5 has something which evidently just doesn't work if you drop the voltage much.  Sidehack has said he suspects the (chained) clock line, at any rate some of the chained signals and the related I/O voltages are most likely suspect.  As far as I know no one has taken the time to mod an S5 to be able to undervolt or could tell you what is going wrong for certain.

--
novak
donator
Activity: 4760
Merit: 4323
Leading Crypto Sports Betting & Casino Platform
The efficiency doesn't change as much with the frequency settings as one might like.  I think it is more useful as a heat management tool.
legendary
Activity: 4256
Merit: 8551
'The right to privacy matters'
very very flat.

 .49 with  900gh say freq  250

.51   with 1326gh say freq 412

these are from memory I used high quality plat psu's and an evga 1300g2 to test.
legendary
Activity: 1218
Merit: 1003
You can calculate it yourself.

Hash Rate: 1155 GH/s   Power Efficiency: 0.51 J/GH (at the wall)

Power Consumption: 590 W (at the wall)

Hash Rate: 1300 GH/s   Power Efficiency still the same : 0.51 J/GH (at the wall)

Power Consumption: x W (at the wall)     

x = 664 W
sr. member
Activity: 751
Merit: 253
Good question.  I have some, ran some brief tests and did not see significant gains so I just left at stock.  I am sure those S5s have some room.  I just changed frequency.  I am sure it is a combination of volts and freq but just too time consuming for only a couple units.  If I had a 1000 units I would be tweeaking every setting.  So If I spent 100 hrs figuring this out, I might not want to just share that info for free.

BUT I hope someone does.  Just give us some idea of potential.   Like what is max hash and the power consumption at that rate.

look forward to some info
sr. member
Activity: 507
Merit: 253
At what clock rate(s) does an AntMiner S5 mine with peak efficiency, where efficiency is measured in terms of hashes / Watt?

Has anyone here performed such a measurement? If so, could you please post your data? thanks
Jump to: