Quote from: Biffa on October 12, 2014, 10:33:15 AM

Its just wierd the way it performs with the up and down performance, the higher performing one is one rock solid line.

Low Performer:

Code:

----------
AC2DC power=1176[1260], temp=0, leading zeroes=41
L |Vtrm/max |vlt|Wt|Am|DCt|crt|Gh|Ov|ID |aTMP|Freq |Th|Bt|Engn|Wn |ID |aTMP|Freq |Th|Bt|Engn|Wn |ID |aTMP|Freq |Th|Bt|Engn|Wn |ID |aTMP|Freq |Th|Bt|Engn|Wn
0|fff1(fff1)|674|39|57| 75|340|59| 0| 0: 77c 495hz( 0/ 0) 7fff181| 1: 77c 495hz( 0/ 0) 7fff153| 2: 77c 525hz( 0/ 0) 7fff155| 3: 77c 495hz( 0/ 0) 7fff155
| 4: 77c 480hz( 0/ 0) 7fff150| 5: 77c 495hz( 0/ 0) 7fff151| 6: 77c 495hz( 0/ 0) 7fff151| 7: 77c 510hz( 0/ 0) 7fff167
1|ffed(ffed)|664|35|53| 68|350|57| 0| 8: 77c 480hz( 0/ 0) 7fff188| 9: 77c 495hz( 0/ 0) 7fff166| 10: 77c 480hz( 0/ 0) 7fff172| 11: 77c 480hz( 0/ 0) 7fff175
| 12: 77c 480hz( 0/ 0) 7fff161| 13: 77c 450hz( 0/ 0) 7fff183| 14: 77c 495hz( 0/ 0) 7fff185| 15: 77c 480hz( 0/ 0) 7fff168

High Performer:

Code:

----------
AC2DC power=1254[1260], temp=0, leading zeroes=41
L |Vtrm/max |vlt|Wt|Am|DCt|crt|Gh|Ov|ID |aTMP|Freq |Th|Bt|Engn|Wn |ID |aTMP|Freq |Th|Bt|Engn|Wn |ID |aTMP|Freq |Th|Bt|Engn|Wn |ID |aTMP|Freq |Th|Bt|Engn|Wn
0|fff2( d)|677|40|60| 79| de|64| 0| 0: 77c 540hz( 0/ 0) 7fff7202| 1: 77c 510hz( 0/ 0) 7fff6604| 2: 77c 540hz( 0/ 0) 7fff7159| 3: 77c 525hz( 0/ 0) 7fff6955
| 4: 77c 555hz( 0/ 0) 7fff7277| 5: 77c 540hz( 0/ 0) 7fff7179| 6: 77c 540hz( 0/ 0) 7fff7011| 7: 77c 540hz( 0/ 0) 7fff7057
1|fff3( d)|680|39|58| 71| 96|65| 0| 8: 77c 540hz( 0/ 0) 7fff7335| 9: 77c 540hz( 0/ 0) 7fff7479| 10: 77c 540hz( 0/ 0) 7fff7218| 11: 77c 555hz( 0/ 0) 7fff7481
| 12: 77c 540hz( 0/ 0) 7fff7259| 13: 77c 555hz( 0/ 0) 7fff7382| 14: 77c 540hz( 0/ 0) 7fff7176| 15: 77c 570hz( 0/ 0) 7fff7606

I just thought I'd point out a few things on these. First, you'll note that the low performer is only using 1176W (DC) whereas the high performer is using 1254W. This means you're getting 0.857 J/GH (DC) on the slow machine and 0.832 J/GH on the fast one, so your fast unit clearly has superior ASICs. Second, you have slightly higher voltages on the fast machine on average, which is probably responsible for the lower power consumption. There might be a way to increase the voltages that the firmware settles on by changing (increasing) the initial voltages, but you also might brick your miner by trying that, and you'd only gain at most around 80 GH/s and you'd lose efficiency as well.

As for the waviness of the line, if you look at the graphs you'll notice that the one running 1.5.8 has the graph scale go from 1.2 TH/s to 1.5 TH/s, but the one running 1.5.4 has the graph scale go from 0 TH/s to 1.6 TH/s. The reason for this scaling difference is that Zvi made this modification (https://bitcointalk.org/index.php?topic=521520.msg8356336#msg8356336) in the more recent firmware after I showed it to him. Consequently, the graphs on 1.5.8 just show more detail than on 1.5.4.