AntMiner S1 Underclocking, a thorough discussion

[CHAOSiTEC]

I'm thinking it's something with values over 300, because I've put in an aslo of numbers between 200 and 275 and it gave good data on all them.

Additionally, ignore that thing I said about 768.75MHz. I had transcribed the binary wrong; it does actually come out to the correct frequency. But an OD of 2 is impossible for frequencies above 250MHz. I reckon that's where the issues lies; I need to put in an extra check for that condition.

hmm, 268.75 returns 1502, but should be 1505..

but yeah everything up to 250 is correct, above, everything which is divideable with 12.5 is correct

[sidehack]

Alrighty, try it now. The problem was it had a check for lowerbound BS criteria (2^OD < 300) but not upperbound (2^OD > 1000) and somehow every test case I gave it didn't violate that. It checks for that condition now, and if it's met it halves the numerator (M+1) and halves the denominator (2^OD, by decrementing OD). The problem there is, if M+1 is an odd number, the halving automatically truncates to M/2 so it rechecks the M/2 and M+2/2 conditions (for the case of f=384MHz, M+1=123 so (M+1)/2=61.5 which truncates to 61 but rounds up to 62) for which one is closer and sticks with the M+1 with the least error. I don't guarantee it works for everything now, but it should work for every frequency for which the initial algorithm calculates an OD of 0, 1 or 2?

[CHAOSiTEC]

Alrighty, try it now. The problem was it had a check for lowerbound BS criteria (2^OD < 300) but not upperbound (2^OD > 1000) and somehow every test case I gave it didn't violate that. It checks for that condition now, and if it's met it halves the numerator (M+1) and halves the denominator (2^OD, by decrementing OD). The problem there is, if M+1 is an odd number, the halving automatically truncates to M/2 so it rechecks the M/2 and M+2/2 conditions (for the case of f=384MHz, M+1=123 so (M+1)/2=61.5 which truncates to 61 but rounds up to 62) for which one is closer and sticks with the M+1 with the least error. I don't guarantee it works for everything now, but it should work for every frequency for which the initial algorithm calculates an OD of 0, 1 or 2?

fantastic, now it works with my odd test cases :-)

i tried myself to get the calculations to work about a month ago, and got stumped on the same, and had given up on it..

(well, im not the best cpp programmer in the world, but i can read the code, and do a little here and there..)

is it ok with you if i implement it into the cgminer i hacked for connecting s1 blades to cp2102s?

thread: https://bitcointalk.org/index.php?topic=671128.0

[sidehack]

I may have to read that thread sometime. I've been avoiding it for a while, though I think Novak was paying some attention. In any case, yes I don't mind you using that code. I like commenting the heck out of stuff in case someone else picks it up, and I hope it's actually good enough. There may still be some cases that break it; I should probably write a testbench at some point that feeds it every value from 1 to about 400 and backsolves the output for correctness.

Or I was thinking earlier, an easy way to do it would be to iteratively generate a list of all valid hex codes and then check for the real frequency closest to the desired. But that's kinda cheating. But it'd guarantee the solution.

[CHAOSiTEC]

I may have to read that thread sometime. I've been avoiding it for a while, though I think Novak was paying some attention. In any case, yes I don't mind you using that code. I like commenting the heck out of stuff in case someone else picks it up, and I hope it's actually good enough. There may still be some cases that break it; I should probably write a testbench at some point that feeds it every value from 1 to about 400 and backsolves the output for correctness.

Or I was thinking earlier, an easy way to do it would be to iteratively generate a list of all valid hex codes and then check for the real frequency closest to the desired. But that's kinda cheating. But it'd guarantee the solution.

something like this? :-p

Code:

chip_freq timeout freq_value(s)
37.50 373 0583,0b87
39.06 358 0c07
40.62 344 0603,0c87
42.19 331 0d07
43.75 320 0683,0d87
45.31 308 0e07
46.88 298 0703,0e87
48.44 289 0f07
50.00 280 0783,0f87
51.56 271 1007
53.12 263 0803,1087
54.69 256 1107
56.25 248 0883,1187
57.81 242 1207
59.38 235 0903,1287
60.94 229 1307
62.50 224 0983,1387,4983,5387
64.06 218 1407,5407
65.62 213 0a03,1487,4a03,5487
67.19 208 1507,5507
68.75 203 0a83,1587,4a83,5587
70.31 199 1607,5607
71.88 194 0b03,1687,4b03,5687
73.44 190 1707,5707
75.00 186 0582,0b83,0b86,1787,4b83,5787
76.56 182 5807
78.12 179 0c06,4c03,5887
79.69 175 5907
81.25 172 0602,0c86,4c83,5987
82.81 169 5a07
84.38 165 0d06,4d03,5a87
85.94 162 5b07
87.50 160 0682,0d86,4d83,5b87
89.06 157 5c07
90.62 154 0e06,4e03,5c87
92.19 151 5d07
93.75 149 0702,0e86,4e83,5d87
95.31 146 5e07
96.88 144 0f06,4f03,5e87
98.44 142 5f07
100.00 140 0782,0f86,4f83,5f87
101.56 137 6007
103.12 135 1006,5003,6087
104.69 133 6107
106.25 131 0802,1086,5083,6187
107.81 129 6207
109.38 128 1106,5103,6287
110.94 126 6307
112.50 124 0882,1186,5183,6387
114.06 122 6407
115.62 121 1206,5203,6487
117.19 119 6507
118.75 117 0902,1286,5283,6587
120.31 116 6607
121.88 114 1306,5303,6687
123.44 113 6707
125.00 112 0982,1386,4982,5383,5386,6787
128.12 109 1406,5406
131.25 106 0a02,1486,4a02,5486
134.38 104 1506,5506
137.50 101 0a82,1586,4a82,5586
140.62 99 1606,5606
143.75 97 0b02,1686,4b02,5686
146.88 95 1706,5706
150.00 93 0581,0b82,0b85,1786,4b82,5786
153.12 91 5806
156.25 89 0c05,4c02,5886
159.38 87 5906
162.50 86 0601,0c85,4c82,5986
165.62 84 5a06
168.75 82 0d05,4d02,5a86
171.88 81 5b06
175.00 80 0681,0d85,4d82,5b86
178.12 78 5c06
181.25 77 0e05,4e02,5c86
184.38 75 5d06
187.50 74 0701,0e85,4e82,5d86
190.62 73 5e06
193.75 72 0f05,4f02,5e86
196.88 71 5f06
200.00 70 0781,0f85,4f82,5f86
203.12 68 6006
206.25 67 1005,5002,6086
209.38 66 6106
212.50 65 0801,1085,5082,6186
215.62 64 6206
218.75 64 1105,5102,6286
221.88 63 6306
225.00 62 0881,1185,5182,6386
228.12 61 6406
231.25 60 1205,5202,6486
234.38 59 6506
237.50 58 0901,1285,5282,6586
240.62 58 6606
243.75 57 1305,5302,6686
246.88 56 6706
250.00 56 0981,1385,4981,5382,5385,6786
256.25 54 1405,5405
262.50 53 0a01,1485,4a01,5485
268.75 52 1505,5505
275.00 50 0a81,1585,4a81,5585
281.25 49 1605,5605
287.50 48 0b01,1685,4b01,5685
293.75 47 1705,5705
300.00 46 0580,0b81,0b84,1785,4b81,5785
306.25 45 5805
312.50 44 0c04,4c01,5885
318.75 43 5905
325.00 43 0600,0c84,4c81,5985
331.25 42 5a05
337.50 41 0d04,4d01,5a85
343.75 40 5b05
350.00 40 0680,0d84,4d81,5b85
356.25 39 5c05
362.50 38 0e04,4e01,5c85
368.75 37 5d05
375.00 37 0700,0e84,4e81,5d85
381.25 36 5e05
387.50 36 0f04,4f01,5e85
393.75 35 5f05
400.00 35 0780,0f84,4f81,5f85
406.25 34 6005
412.50 33 1004,5001,6085
418.75 33 6105
425.00 32 0800,1084,5081,6185
431.25 32 6205
437.50 32 1104,5101,6285
443.75 31 6305
450.00 31 0880,1184,5181,6385

thanks to your efforts with testing different voltage settings on the s1, i now know exactly what to set my s1 boards at ( i have 20 ) should give me aprox 900 GHS for aprox same level as S3s :-p

[sidehack]

Yes, something quite a bit like that. Did you backsolve the hex values to make sure they all made sense? My only concern is if the code spits out anything with an OD=3, which can only be valid for frequencies at or below 125MHz. I only check and correct one depth of OD overshoot right now, which would correct OD=2 for frequencies between 250 and 500MHz.

Also, if you have the time to tune instead of setting all banks to the same stock value, I've been finding some banks that worked at voltages lower than my test data. I have one S1 that's pulling 140GH at 152WDC including fan and controller. I think every bank is at 830mV or better with ~1% HW though my test board did that at 850mV. Course one I did today had a few banks that required 860mV to stay clean.

I'm seriously impressed with their chips. 0.8W/GH is pretty good for year-old hardware that shipped at 2W/GH and was already ahead of the curve. I've got 12xS1 on a single DPS2K with power to spare. Figure in four or six months I'll take 'em to the lowerbound and keep 'em going for a while. They're helping fund the museum right now.

Here in another month or so (if I get time; we're starting to get some real work coming in) I plan on doing the same in-depth analysis on AM Tubes.

[helipotte]

I would be interested in the tube experimenting.  I have three I run at around 0.75V that do 730Gh.  This dropped their power consumption by 50W per board.  Heat is much

more manageable as well.  No longer have to use two or more fans.

[sidehack]

It probably won't be quite as straightforward to run as an S1, but... actually, I bet it will be if I can cgminer it. The VRMs are the same 53355DQP as the S1, S2, Cube, AM-V1 and a bunch of other hardware use, so I already know everything required to do the work. Just gotta find the time.

[helipotte]

It probably won't be quite as straightforward to run as an S1, but... actually, I bet it will be if I can cgminer it. The VRMs are the same 53355DQP as the S1, S2, Cube, AM-V1 and a bunch of other hardware use, so I already know everything required to do the work. Just gotta find the time.

Actually Crazyguy released a modified version of Cgminer with Minera for the Raspberry pi.  I use it to run one of my tubes.  I bought a generic cp2102 adapter on Amazon

to use with it.  Works really well with any pool.  Had to install a 1K pull-up resistor on the cp2102 adapter.  Here is the original post:

https://bitcointalk.org/index.php?topic=865503

[sidehack]

Right, I've got a handful of machines in hosting using Crazyguy's image for Pi and Cubie running Prismas. It's handy, but the interface (at least what little I've seen) doesn't provide near as much information as I would require to do the job right. I'd rather use a standalone (AM-specific) cgminer instance I can get full data from on the fly. I've never really been a fan of RPi myself, rather use a real computer if I can.

Regarding that adapter, Novak and I work together (GekkoScience is the business we started together) so I already know about 'em. He used one of my old Block Erupters to prototype on. I'll probably use one of the adapters he's gonna be batching (https://bitcointalk.org/index.php?topic=882348.0, I think about half of them are sold already) to do the testing, if he has any left by then. Pretty handy stuff that guy does.

[helipotte]

I did the block erupter to uart adapter thing too.  Love breathing new life (or repurposing) old hardware.  I am thinking about making a high density miner based on the

S1 hardware.  Would like to get about 20 S1 blades running on one controller.

[sidehack]

The stock S1 controller, by my measures, only draws about 1W. Not too worried about that power overhead. But it would be nice to rig up something with more space-efficient cooling. Might as well use a single combined controller if you can get them stripped down and lumped into a more compact unit.


Oh, if anyone feels sufficiently benefitted by this work to toss donations, use the BTCMuseum address in my sig. We're buying up old hardware to put together a display of mining history, right now it's all funded by Novak and I's donations and mining on the stuff we're collecting.

[helipotte]

I wanted to do it more from a "manageability" perspective.  Having so many individual controllers is a bit of a pain.  Right now I use a proxy for all my S1's.  Would be nice to

have one Cgminer instance for all of them.

[sidehack]

That is also an excellent point. I've got twelve running at the shop right now (underclocking is a work in progress) and it's a bit of a pain to reconfigure them when necessary.

[CHAOSiTEC]

Also, if you have the time to tune instead of setting all banks to the same stock value, I've been finding some banks that worked at voltages lower than my test data. I have one S1 that's pulling 140GH at 152WDC including fan and controller. I think every bank is at 830mV or better with ~1% HW though my test board did that at 850mV. Course one I did today had a few banks that required 860mV to stay clean.

I'm seriously impressed with their chips. 0.8W/GH is pretty good for year-old hardware that shipped at 2W/GH and was already ahead of the curve. I've got 12xS1 on a single DPS2K with power to spare. Figure in four or six months I'll take 'em to the lowerbound and keep 'em going for a while. They're helping fund the museum right now.

Here in another month or so (if I get time; we're starting to get some real work coming in) I plan on doing the same in-depth analysis on AM Tubes.

i have brought a bunch of potentometers that im adding to each blade (4 per section) and i plan on using that to fine tune each section, and adding them all together on the same controller (1 rasbpi) currently i have 5 done (those are pencil modded) so waiting for the last cp2102s to arrive to get the rest done, and waiting for those potentometers to arrive from china...

its amazing how much these chips can be pushed, but then, it just shows the craftmanship of these chips.

[sidehack]

That's how I did the first, with a pot on a big ol' knob for super fine adjustments (I could tune it to 1mV easily). I've got resistors premade now to parallel on the existing 8200s that make 10mV increments, and I've been setting one bank, running for a few minutes, then moving up or down as necessary until the error rate hit threshold. I've a parts order coming Monday with the actual SMD resistors to do the swaps proper, instead of having through-hole resistors hanging off the boards.

I definitely need to read that external-controller thread for these guys and see what work y'all have done.

[klondike_bar]

It probably won't be quite as straightforward to run as an S1, but... actually, I bet it will be if I can cgminer it. The VRMs are the same 53355DQP as the S1, S2, Cube, AM-V1 and a bunch of other hardware use, so I already know everything required to do the work. Just gotta find the time.

I would be inteested to know if the AM-V1 can undervolt, same with rockminer gear.

MY EXPERIENCE WITH ROCKMINER VOLTAGE (relevant to rk-box and r3):
The regulator on these boards is not marked, but looks just like a TPS53355 seen on Antminer S1/S2 and Bitfury boards. However, it is arranged differently, without the same placement of R1/R2 that are used for raising/lowering voltage. (aplying pencil to R1 raises voltage, R2 decreases voltage). In the case of rockminer, no resistors were acessible that would reduce the voltage when resistance was reduced. only the oposite was seen - increase in voltage when R1 and R2 were simultaneously modded. physical replacement of the resistors would be the only solution i think

looks a lot like the TPS53355 regulator in this picture, with two resisters right by it - if so should be quite easy to pencil mod

http://www.ti.com/lit/ds/symlink/tps53355.pdf   - look at the bottom image on page 7, the reference on page 20, and figure 38.
TLDR; Pin 1 has two resistors on it, R2 goes to ground and R1 goes through a small circuit. The choice of values for these gives the output voltage. This is how the Bitfury overvolts, and how the S1 can over/undervolt. looking at the rockminer PCB (would love a high-def image of the section), this is the solution to undervolt for efficiency:

some pencil lead on R2 increases the output voltage, while lead on R1 would decrease it. Presumably the efficiency could be brought closer to the 0.7w/GH achieved by the prisma this way

*note*: Im basing this on a blurry image and dont have a PCB in front of me to confirm that Pin 1 is the right corner of the chip, but the positionng of the two resistors like that seems logical. hopefully ill try myself on monday evening

update: I brought a unit home, and am unable to determine any markings on the regulator component. looks an awful lot like the TPS5355 though, and as there arent many >20A alternatives I would assume that it is.

however, my expectations were not correct. pencil modding either R1 or R2 (or any other resitor in the vicinity of the regulator) has no visible effect. It appears that R1 and R2 in my above image are both brought to ground together.

HOWEVER - I made an interesting discovery - pencil modding BOTH R1 and R2 results in the output voltage increasing, with 1.15V achieved quite easily using a 2B pencil (stock is 0.75V). Obviously this is the opposite of my intentions though.

UPDATE 2: Partial destruction of R1 and R2 (chipping the sides of the resitors with a sharp tool) so that each is roughly 3.3kOhm when measured in circuit results in a regulator output voltage of 0.6V Testing now -> preliminary results are mediocre at best - 16GH. However, its worth noting that this unit has always been a source of problems, and that two other boards in it report 25GH and 58GH, neither of which were modified in anyway. Not sure if its a USB hub issue, RPI, or maybe even something to do with the server PSU (providing 12.03V)

UPDATE 3: replaced a controller on a different board that was causing reboots and often not functional -> 110GH,90GH, 105GH, 70GH. Obviously there is quite a spread here but it seems like the modified board is functional. Because of difficulties operating (or lack of) at 0.6V, pencil mod was added to increase the voltage to 0.7V on the modified (70GH) board. Reject rate is 6.5% at 300MHz

UPDATE 4: 12hr stats: 112,112,111,66 GHash across the 4 boards, with the 65GH board being the one i modified. Total reject % is 4.3.

TLDR; Pencil mod is not easily implemented on the RK-Box. reduction of chip voltage to 0.7V significantly impacts hashrate, and voltage below 0.65V is often non-functional. No efficiency numbers available yet

[sidehack]

I have exactly one AM-V1, but we just got it in for the museum and I don't really feel like messing with it. I have nothing at all from RockMiner. If anyone's got either to sell at reasonable prices I can look 'em over.

[sidehack]

Looks like we're getting some BTCGarden blades in, and possibly a KnC Jupiter module, so I might have some underclock data on those things in a few weeks. Depends on when (and if) stuff arrives, and what kind of time I have between jobs and holidays.

[CHAOSiTEC]

Looks like we're getting some BTCGarden blades in, and possibly a KnC Jupiter module, so I might have some underclock data on those things in a few weeks. Depends on when (and if) stuff arrives, and what kind of time I have between jobs and holidays.

looking forward to it :-)