An Attempt to create a more complete freq value chart for the ant miner chip.

ZiG

sr. member

Activity: 406

Merit: 250

Quote from: philipma1957 on July 14, 2014, 08:20:07 AM

I put these links in one spot they will be helpful for the s-1 maybe even the s-3

good links to help with over and under clock of the s-1. also for under volting the s-1

original undervolt- under clock link

https://bitcointalksearch.org/topic/guide-undervolt-antminer-s1-119wgh-at-the-wall-526060

youtube over clock link

https://www.youtube.com/watch?v=72FU4Z_n0B4

youtube for under volt

UnderVolt Antminer S1 Video-Guide
http://youtu.be/Cj9hgoUnpTc

https://bitcointalksearch.org/topic/guide-undervolt-bitmaintech-antminer-s1-video-guide-688408
and the thread for that on bitcointalk

S3 is a "little" bit DIFFERENT...chapter 3 of this...:

https://bitmaintech.com/files/download/BM1382_Datasheet_v3.0.pdf

ZiG

whonesta

sr. member

Activity: 379

Merit: 250

Quote from: philipma1957 on July 17, 2014, 11:29:44 PM

please try to underclock to 215.625

and to 206.25

the above codes were from my hand calculations, GH speeds added. 215.625 gives me 1100Gh and less than 1.1% error rate

https://bitcointalksearch.org/topic/m.6485048

so 206.25

and 215.625 were both developed by klondike_bar

option 'freq_value' '6206' #215.625M
option 'chip_freq' '215.625'
option 'timeout' '18' maybe 17

#option 'freq_value' '5002' #206.25M
#option 'chip_freq' '206.25'
#option 'timeout' '19' maybe 20

these 2 can be tested

Neither clock freq worked with the antMiner S3, nor did 275 fyi Wink

philipma1957

legendary

Activity: 4102

Merit: 7765

'The right to privacy matters'

I put these links in one spot they will be helpful for the s-1 maybe even the s-3

good links to help with over and under clock of the s-1. also for under volting the s-1

original undervolt- under clock link

https://bitcointalksearch.org/topic/guide-undervolt-antminer-s1-119wgh-at-the-wall-526060

youtube over clock link

https://www.youtube.com/watch?v=72FU4Z_n0B4

youtube for under volt

UnderVolt Antminer S1 Video-Guide
http://youtu.be/Cj9hgoUnpTc

https://bitcointalksearch.org/topic/guide-undervolt-bitmaintech-antminer-s1-video-guide-688408
and the thread for that on bitcointalk

windpath

legendary

Activity: 1258

Merit: 1027

387 is awesome!

I have 6 Ants, one of them is missing a couple capacitors, and ran with a VERY high error rate at every setting I tried, was over 1% @ factory setting, and over 3% at other speeds...

Set it to 387 yesterday to give it a whirl...

Error rate: 0.00082799134 %

Calculator: http://www.coincadence.com/antminer-s1-hardware-error/

My worst Ant as far as errors goes has now set the bar for all the others

philipma1957

legendary

Activity: 4102

Merit: 7765

'The right to privacy matters'

Quote from: colinrgodsey on May 15, 2014, 09:31:17 PM

I'll see if i can make a table. I finally tested that code and it should work (will verify against some actual data), can probably just make a script to dump it out to CSV quick.... can throw that into excel or whatnot.

would be nice of you to do so. thanks.

colinrgodsey

newbie

Activity: 39

Merit: 0

I'll see if i can make a table. I finally tested that code and it should work (will verify against some actual data), can probably just make a script to dump it out to CSV quick.... can throw that into excel or whatnot.

philipma1957

legendary

Activity: 4102

Merit: 7765

'The right to privacy matters'

Quote from: colinrgodsey on May 08, 2014, 11:59:49 PM

Quote from: -ck on May 01, 2014, 06:58:50 AM

Cgminer calculates values on the fly for any frequency for the U1/2s with the --anu-freq command. You can either run a U1 with cgminer and various frequencies in debug mode to see what values it generates, or you can audit the code that I wrote that generates it.

annnnd im an asshole for not reading the thread well enough. sorry ckolivas =\

no you are not an a-hole.

I am just a bit too extended time wise to punch in all the freqs and test out the new cgminer.

I will get to it one day soon or as BFL would say in two weeks or so.

I wanted to come up with an easy plug-in chart and frankly The chart I have so far is easy but still short number wise.

I did a lot of testing with freq 387 and freq 393.75 some miners do like one over the other.

I also have a miner that prefers 375 over 350 or 387 or 393 or 400 .

I have to have a few hours alone time to test cgminer and have not had the empty time slot open for me.

colinrgodsey

newbie

Activity: 39

Merit: 0

Quote from: -ck on May 01, 2014, 06:58:50 AM

Cgminer calculates values on the fly for any frequency for the U1/2s with the --anu-freq command. You can either run a U1 with cgminer and various frequencies in debug mode to see what values it generates, or you can audit the code that I wrote that generates it.

annnnd im an asshole for not reading the thread well enough. sorry ckolivas =\

colinrgodsey

newbie

Activity: 39

Merit: 0

the newer version of cgminer has some formula for this (all possible values)

Code:

static void anu_find_freqhex(void)
{
float fout, best_fout = opt_anu_freq;
int od, nf, nr, no, n, m, bs;
float best_diff = 1000;

anu_freqfound = true;

for (od = 0; od < 4; od++) {
no = 1 << od;
for (n = 0; n < 16; n++) {
nr = n + 1;
for (m = 0; m < 64; m++) {
nf = m + 1;
fout = 25 * (float)nf /((float)(nr) * (float)(no));
if (fabsf(fout - opt_anu_freq) > best_diff)
continue;
if (500 <= (fout * no) && (fout * no) <= 1000)
bs = 1;
else
bs = 0;
best_diff = fabsf(fout - opt_anu_freq);
best_fout = fout;
anu_freq_hex = (bs << 14) | (m << 7) | (n << 2) | od;
if (fout == opt_anu_freq) {
applog(LOG_DEBUG, "ANU found exact frequency %.1f with hex %04x",
opt_anu_freq, anu_freq_hex);
return;
}
}
}
}
opt_anu_freq = best_fout;
applog(LOG_NOTICE, "ANU found nearest frequency %.1f with hex %04x", opt_anu_freq,
anu_freq_hex);
}

if thats too hard to read, here's the converted snippet from ScalaMiner (also probably hard to read... also untested, sorry)

Code:

def getAntMinerFreq(freq: Double) = {
var bestDiff = 1000.0
var bestOut = freq
var exactFound = false

for {
od <- 0 until 4
no = 1 << od
n <- 0 until 16
nr = n + 1
m <- 0 until 64
nf = m + 1
fout = 25.0 * nf / (nr * no)
if math.abs(fout - freq) <= bestDiff
if !exactFound
} {
val bs = if(500 <= (fout * no) && (fout * no) <= 1000) 1 else 0
bestDiff = math.abs(fout - freq)
bestOut = fout

val freqVal = (bs << 14) | (m << 7) | (n << 2) | od;

if(freqVal == fout) {
exactFound = true
bestOut = fout
}
}

bestOut
}

helipotte

hero member

Activity: 650

Merit: 500

Pick and place? I need more coffee.

Quote from: Profetu on May 07, 2014, 03:55:12 PM

Hi guys, what values should I use so I can safely power 2 ants with a 750W PSU? I have so many PSUs I don't want to buy more Cheesy

Could I do something like 160 GH for 250W?

Sounds about right. I have two Ants per CX750M and run mine at 300Mhz. I also undervolt to about .90-.95V. This results in about 220W of power draw per Ant. They hash

at about 153Gh. I would imagine 325Mhz would result in close to what you want, power wise. I calculate 8x325=2600Mhs x 64 =166400Mhs, so 166Gh and about 250W.

I run mine at 300Mhz because it seems like the best tradeoff between power use and hashrate. The temps went down quite a bit too Vs. stock speeds.

Profetu

full member

Activity: 204

Merit: 100

Hi guys, what values should I use so I can safely power 2 ants with a 750W PSU? I have so many PSUs I don't want to buy more Cheesy

Could I do something like 160 GH for 250W?

philipma1957

legendary

Activity: 4102

Merit: 7765

'The right to privacy matters'

Quote from: -ck on May 01, 2014, 06:58:50 AM

Cgminer calculates values on the fly for any frequency for the U1/2s with the --anu-freq command. You can either run a U1 with cgminer and various frequencies in debug mode to see what values it generates, or you can audit the code that I wrote that generates it.

I am waiting for a merged usb2 bridge for my 19 port miner. this will allow me to do up to 1.8 amps or 9 watts to a u-2

I will then play with your program. will be nice if I get it to do this. I can run a lot of air on the u-2 miner it should not over heat.

freebit13

hero member

Activity: 616

Merit: 500

I got Satoshi's avatar!

I notice that even though the 387 has a higher HW error rate, you can see that it has more accepted shares and the diffA is almost identical between the 2... with 20W less power, it might actually be the better setting

Nice work!

philipma1957

legendary

Activity: 4102

Merit: 7765

'The right to privacy matters'

this shows 2 s-1's

one at 387 one at 393

the 393 has the better hw rate and hashes a bit more but

the 387 uses 20 watts less power

these s-1's have variance

and a really full chart will help us with clocking.

BTW that one ant miner has a really good hw rate 3 bad and more then 2.5 mill good
It will be fun testing the s-1's .

Please send me freq-- hex codes

freebit13

hero member

Activity: 616

Merit: 500

I got Satoshi's avatar!

Quote from: philipma1957 on May 02, 2014, 07:20:07 PM

and 387 looks like a good choice for an s-1

I will slap a k-watt meter on it later

image removed...

Nice one, it is pretty stable and reduced the HW errors... it might be a good idea to list the HW error rate in the charts too as it really starts to make a difference above 387.

philipma1957

legendary

Activity: 4102

Merit: 7765

'The right to privacy matters'

Quote from: crazyates on May 02, 2014, 09:14:37 PM

Quote from: philipma1957 on May 02, 2014, 07:20:07 PM

and 387 looks like a good choice for an s-1

I will slap a k-watt meter on it later

We run ours at 393, which pull 410W (+/- ~4W depending on unit) on a 80Plus Gold PSU.

I run most at 393. I wanted to get 20 or 30 more values to test..

I am curious about various freq and the hex . an easy pop in a number chart up and down the freq would help to test as many as poosible.

crazyates

legendary

Activity: 952

Merit: 1000

Quote from: philipma1957 on May 02, 2014, 07:20:07 PM

and 387 looks like a good choice for an s-1

I will slap a k-watt meter on it later

We run ours at 393, which pull 410W (+/- ~4W depending on unit) on a 80Plus Gold PSU.

philipma1957

legendary

Activity: 4102

Merit: 7765

'The right to privacy matters'

my k -a-watt meter was running 1.5 s-1's on an evga it read 589 adding the full one clocked at 387 freq jumped power to 977 so this is pulling 388 watts at freq if I get 201gh and 388 watts steady this setting is better then the 393 setting most of us use for s-1's

This is what I want the thread to fill in as many numbers as possible. I found when I used to mine btc with gpus small shifts in settings could make big watt gains.

It could be that 387 is better then 393 for most of our miners. at least most s-1's

philipma1957

legendary

Activity: 4102

Merit: 7765

'The right to privacy matters'

and 387 looks like a good choice for an s-1

I will slap a k-watt meter on it later

philipma1957

legendary

Activity: 4102

Merit: 7765

'The right to privacy matters'

Quote from: freebit13 on April 30, 2014, 12:55:51 PM

You can add this to the list for the S1. It ran better on an older unit that would throw a lot of HW errors at higher frequencies.

freq_value 5e82
chip_freq 387
timeout 37

thank you for the input I set a new ant miner s-1 to it here is a screen shot. so another one that will work . to all I will do more updates soon and thanks for info

Topic: An Attempt to create a more complete freq value chart for the ant miner chip. (Read 9331 times)