Topic: Antminer S5 - Underclock - Undervolt - Best J/GH - page 11. (Read 31123 times)

"Ran some tests today on the DPS-600 PSU with a voltage adjust pot.(-12V to the Voltage input pin, Pins 5 to 9)  Bottom line not much use with an S5. "

One is not enought but two is. Quick search from net gave me a price of over 50$ per Mean Well 5V supply and You need at least two of them. Efficience is about same and You can get pair of DPS-600s with wires at 40$.   
17$ per DPS-600 shipped; RC servo extensions for on-off pins, trimpots and current share; molex-PCI-E connectors for power feed. 

DPS-600 is easy to hack You don't need to open it.

Like RichBC said (dejavu), at 10.5V it support 200W~, not the full rating from when it's ran at 12V. So you'd need two for a S5. For a S3 and a S1 its plenty.

DPS-600 has a current share pin so You can connect them in parallel to S5.
10.5 V 40A should be enought for undervolted/clocked S5.
Small DPS fans stay quiet when power is low and You still have a under 40$ power source.
Maxed DPS fans have a very annoying high tone whine.

I have 3 S1 undervolted with DPS-600s as livingroom heaters 40db at 1m.

Have measured the efficiency of the MeanWell HRP-150-5 5V PSU's with 2 connected in Series. BTW there seem to be no problems connecting them in series although I am being careful to adjust them both to the same voltage. The DC output is isolated from the metal case.

Efficiency measured at 81% from 9V to 10V & 83% from 11V to 12V. So a bit lower than the quoted 87%, but I always seem to find that. Could be my measurement technique or perhaps the Manufacturers are a bit generous?

Here is the setup with the PSU's connected in series, with some cooling from the Miner exhaust, although they seem to run quite cool without a fan. Will measure the actual temperature they get to at some point. 5V PSU is rated at 70 Deg with a sensor on the Power Stage heatsink. Ambient needs to be below 45 deg for full rating.



Measurements of Hash & J/GH to follow.  

Rich

Yes agreed I would need to get down to around 9.5V, trouble is it won't go that low & even if it did there would then be < 200W available as power will continue to fall. So an interesting experiment and good to get some real at the wall numbers with reduced voltage. All my earlier measurements were measured on the input to the miner, so taking no account of PSU efficiency.

I am now testing with two 5V PSU's in series, have adjustment from 9V to 12V. Good news is that they work, will be Tomorrow before I have some results. One feaature that holds things back is that on both my S5's one hash board better than the other, so you have to set the voltage for the worse one. At lower voltages seems to be about 0.3V difference for the same error rate. Have not attempted to understand why one is better than the other, but would only take one chip in the chain to upset things.


Rich

Wow, thanks for testing that. To have a 0.426J/Gh with an ~80% efficiency PSU is quite something. Also the temperature should drop @ 10.6V (less noise). I think its worth the effort. I will do some testing soon (no time for that at the moment, so in 2 weeks I guess).

Hmm need something better, this is nearly the same as just running the S5 on stock settings on a Platinum or such PSU. Thanks for testing that unit.
Maybe two units would work, one per PCB?

At 200w you'd need to go down to 9.5v~ iirc thats around where the power drain is under 200w and is about the same hashrate as a S3.

HP DPS-600 / Series ESP135 Model PS-3601-1C - (HP Spares number 338022-001) PSU Test

Ran some tests today on the DPS-600 PSU with a voltage adjust pot.(-12V to the Voltage input pin, Pins 5 to 9)  Bottom line not much use with an S5.

The pot allows adjustment from 12.2V to 10.5V. (If adding a pot put a 150 Ohm resistor in series so that that is as low as you can adjust. I think this is how I blew the first PSU?)

Measured the efficiency of the PSU and at 12V it was 82% Good news was that this only reduced to 81% when set to 10.5V

The Bad news is the the output power is significantly reduced when the voltage is reduced. The rated power from 12V is 560W, this reduces to 220W at 10.5V This is unfortunately not quite enough for an S5 with both hash boards connected. I did however test to see what the best with 2 hash boards that could be achieved.

This was at 11.6V and 350Mhz (Yes the S5 does not need 12V at the rated frequency) This consumed 559W at the wall for 1150GH giving 0.486J/GH (with 12V 0.51J/GH)

With a single hash board connected best was. 10.6V and 300MHz which consumed 211W at the wall for 495GH giving 0.426J/GH


So it's not a bad PSU with 80+ efficiency and can be voltage adjusted. However because the possible output power reduces with voltage it unfortunately does not have enough power below 11.6V for the S5.


Rich

all my testing will be on the s-7.

I have done a fan mod on the one I have at my house.  huge improvement in sound.

links to two of these psu's

http://i.imgur.com/5JvfR8m.jpg
http://i.imgur.com/kh94dIs.jpg
http://i.imgur.com/2vniinV.jpg
http://i.imgur.com/mMM3ydE.jpg
http://i.imgur.com/jjrUOa4.jpg
http://i.imgur.com/v058Wvk.jpg
http://i.imgur.com/Qc6o4pK.jpg
http://i.imgur.com/lLgxB28.jpg
http://i.imgur.com/vPGNi5W.jpg
http://i.imgur.com/sHBG6GR.jpg


I will post a few photos of the  fan mod I did


a 90 mm noctua fan


please cover the ac in or you can kill yourself

 

 I'm not sure but I think that's the same RSP-320-12 I mentioned as one of my original 2 links, just from Jameco instead of from TRC.
 (if it's not, then I have changed my mind since then on the "optimal" one as it's the one in my booklinks.)
 Mouser also carries them, I have noticed, but for a couple bucks more, and at least 1 seller on Amazon in the same price range.


 I dunno if they have enough isolation to use them in series, DO let us know how that works out.

Looks good Phil, Meanwell seem to have a very big range of PSU's. I picked up, at a price I could not resist last night, 4 x 5V Meanwell PSU's from ebay. Am going to try them in pairs as they should give adjustment from 8.6V to 11.6V.

 


Rich

I found another mean well  this one is decent price and should be able to run the s-5 at freq 175

88% efficient  can go to 10 volts



 http://www.trcelectronics.com/View/Mean-Well/RSP-320-12.shtml


https://www.trcelectronics.com/ecomm/pdf/rsp320.pdf

Voltage control is achieved by reducing the 12V supply voltage. As the chips are powered in a string, like Christmas Tree lights, this reduces the core voltage to each chip.

Rich

 Just where in the hardware for the S5 is "voltage control" supported?

Seem pretty deliberate to me when the hardware design support voltage control but the option get locked in newer firmware, even though if you dig enough you can unlock it again.

 1) you're not losing that much efficiency. Many of the Meanwells I've seen could be rated as "gold" by ATX standards, even the lower ones are commonly 88%.
 
 2) Haven't found a stepdown YET that's specified to run LONG TERM at the power draw of an S5 per hash board, a couple talk about "higher possible with a bigger heat sink" but no actual spec for that.

 3) The PSUs I already have will probably be getting transitioned to running an S7 or three....



 Side effect of their "keep it cheap" string design, doesn't seem to be deliberate.
 The old "pencil mod" was about changing an on-board regulator that doesn't exist on their current designs - and was STILL far inferior to Spondoolies methodology.




 Wish I could find a place with affordable rent and enough power/cooling in Chelan or Douglass counties.
 Then again, if I had $50k+ to put into this I wouldn't have to worry about the rent as much.

 8-O

I agree that Bitmain seems to be making it more difficult to undervolt than they could. It is unclear to me if this is driven by cost reduction and simplifying support, or if it is driven by a deliberate strategy to have miners become uneconomic sooner than they could?

However as a hack I think it was and will continue to be possible. S1 yes easy to undervolt with either the Pencil Mod or a resistor change. S3 no change only a simple resistor change needed. S4 designed with undervolting built into the firmware, but as you say later removed. Still unclear why this was done, however from a hack point of view you just use the old firmware.

So moving on we now have the miners with a string supply. Also let's remember that they initially told us with the S5 that with a 9V supply you would see 0.2J/TH. This was later removed however my current theory is that the initial design was flawed and when the voltage to the string was reduced the string stopped hashing. On the latest revision of the hash board V1.91 they have made a change to add an oscillator for every  4 chips in the chain, and this "seems" to allow undervolting.

If we now look at the S5+ and the S7, which I have only seen pictures of, there appear to be oscillators for every stage of the chain. So, yet to be proven? I think that they will both undervolt.

The efficiency gain you can get with undervolting is however reducing as they push the factory release of the product less hard. S1 could be doubled, S3 and S5 each less and the S7 has been released with the chip core voltage at the mid point, however still gains to be had.

So undervolting may not be as simple as the pencil mod and Bitmain may not be encouraging it. However I do not think they are deliberately making it difficult, they are just not spending any money on making it easy.  

Rich

It seem to me Bitmain are trying to make their miners NOT undervoltable, basically unhackable. S1 pen moddable, then S3 no longer, then S4's volt control get disable in later firmware, etc.

Maybe we can find a way to crack the S5, but if they do, they might "patch it" next time... maybe keep the good solutions under wrap?

I only tested volt limits I didn't make any other testing.

Power usage from wall was 700W with 381Mhz S5.

Yes I saw that you could connect the pot such that you can go over & under Voltage, but have no need to increase the voltage so just put it between 5 & 9. Do you find that when you turn the volts down, with no load on the PSU, the fan speed increases?

Rich