KnCMiner Jupiter Miner First Impressions

[DeathAndTaxes]

How they come back the dead core?
What they do?

Maybe if we put more power psu like 1000w will have solution?

The power supply can't supply more power than the DC modules can convert.

4 DC modules @ 40A each = 160A.  Chip nominal voltage is 0.750 VDC so that is 120W output per board.  By spec the DC converters are about 90% efficient that means 133W input.  If you assume they are running hot and thus less efficient maybe 150W input.  4x 150W = 600W of 12VDC.  Throw in another 6W per fan * 6 fans + 10W for the host you are looking at <650W.  On the firmware which overvolts the chip to 0.9V you are still talking <770W.

Now the 8 VRM version might be able to push the chips harder given enough power.

[1713935246]

1713935246

[1713935246]

1713935246

[DPoS]

What do you mean by 'baking'?

He means remaining on 0.93 firmware.  The VRMs are running very hot but they provide a higher throughput (even when derated for higher HW error% and reject%).

yes, the one miner I didn't touch which has one lucky board with 8 VRMs runs 70 watts less and is still on .93 and ~520 Gh

The other is a mess now and but I have it running .94 and back to ~410Gh
I have some more theories to prove out so I will try a few things later tonight with this one

[FeedbackLoop]

Can anyone running the factory 0.90 (no upgrades ever made) please share a full image of their firmware?

[Sitarow]

@ Sitarow:  let us know which front fans you get, I am shopping as well.   You have the good early miner so I would say no one will have better conditions than you for jupiters.
That is why I am planning on making two saturns from my one jupiter.  If I can't get another controller from KNC, I will atleast cut the case in half (front and back) then put the back part 90 degrees like |_ so the ASIC cables can still connect from the separated case.  I can make a makeshift grill to hold new case fans in front of it


@ jelin:  It is a 850psu that was running 770 watts stable.  Some say they got their dead dies to come back, we'll see if I play around with that as solo sometime

http://www.microcenter.com/product/410331/Hale90_Series_V2_850_Watt_Modular_Power_Supply


@ ecliptic: I would like KNC to explain the decision more.  The VRMs had their caps and everything around removed from the board so they sure saved a few nickles at our expense.  I am still baking my 4 VRM boards since they give better performance than using .95

I had posted early in this thread as to what ones I would be trying out first.

Here is a photo of both the one I am going to use compared to an avalon fan.



You can read this post here for more info.

https://bitcointalk.org/index.php?topic=306969.msg3303326#msg3303326

You can find the avalon fans at the link below

http://www.amazon.com/gp/product/B0045733JE/ref=oh_details_o00_s00_i00?ie=UTF8&psc=1

[sickpig]

Can anyone running the factory 0.90 (no upgrades ever made) please share a full image of their firmware?

No unfortunately. So you think that the 0.90 they provided is different from the original?

[Wesly]

An Update on my Jupiter: After taking some advice on the KNC forum I updated to .95, rebooted then did a "hard reset" as outlined in the KNC manual. I saw .95 wasn't working out and switched to .94. I'm now stable at .94 with a cool room where before the hard reset it would drop ghs to unacceptable levels. I'm getting this on Eligius.st right now
3 hours   473.45 Gh/s   1190528
22.5 minutes   475.64 Gh/s   149504
256 seconds   551.80 Gh/s   32890
128 seconds   523.45 Gh/s   15600

My only issue now is this module marked with a "3" sticker on it. My other three modules have "1" and are fine.

Temperature sensor: 31.5 C
Die ID    Cores ON    Cores OFF    %
0   48   0    100
1   0   48    0
2   48   0    100
3   48   0    100
DC/DC ID    ON/OFF    Status    Input Voltage    Output Voltage    Output Current
0    ON    OK    11.8 V    0.744 V    34.9 A (26 W)
1    OFF    OFF    11.9 V    0.712 V    0 A
2    ON    OK    11.9 V    0.815 V    15.1 A (12.3 W)
3    OFF    OFF    11.9 V    0.825 V    0 A
4    ON    OK    12 V    0.745 V    38.6 A (28.8 W)
5    OFF    OFF    11.9 V    0.695 V    0 A
6    OFF    OFF    11.8 V    0.739 V    0 A
7    ON    OK    11.9 V    0.732 V    40.2 A (29.4 W)

Same here. 3 board with a '1' sticker on top. The fourth is marked with a '3'.
Bertmod show me that one my asic slot has a die with all 48 cores disabled.
So i thought that knc did this  because they didnt have enough well functioning
chips for everybody.

Last piece of info. All my board have 8 VRM. I'm not able to run .92 e .93.
.90 run only for the first few mins, then upgrade to .91. Both .94 and .95 with almoste the same perf

Same with my Jupiter, 3 good boards with white label #1 and 1 bad board with label #3 with one of the dies that has all 48 cores off.  So all 4 of us (mininganon, jelin1984, sickpig and me) seem to have the same issue.  If I upgrade my Jupiter to anything other than firmware .91, then this entire board is deactivated and not hashing at all.  I think it is safe to say that it is confirmed KNC has been testing each hashing board separately and intentionally mount 1 bad board with 3 good boards per Jupiter.  Now the question is if there is anyway to reactivate some of the cores in the non-hashing die.

P.S. One more observation I notice is KNC seems to intentionally place the bad board close to the front of the case where 2 case fans are.  This might have done to ensure it has lower temp and a better chance to hash better.

[sickpig]

An Update on my Jupiter: After taking some advice on the KNC forum I updated to .95, rebooted then did a "hard reset" as outlined in the KNC manual. I saw .95 wasn't working out and switched to .94. I'm now stable at .94 with a cool room where before the hard reset it would drop ghs to unacceptable levels. I'm getting this on Eligius.st right now
3 hours   473.45 Gh/s   1190528
22.5 minutes   475.64 Gh/s   149504
256 seconds   551.80 Gh/s   32890
128 seconds   523.45 Gh/s   15600

My only issue now is this module marked with a "3" sticker on it. My other three modules have "1" and are fine.

Temperature sensor: 31.5 C
Die ID    Cores ON    Cores OFF    %
0   48   0    100
1   0   48    0
2   48   0    100
3   48   0    100
DC/DC ID    ON/OFF    Status    Input Voltage    Output Voltage    Output Current
0    ON    OK    11.8 V    0.744 V    34.9 A (26 W)
1    OFF    OFF    11.9 V    0.712 V    0 A
2    ON    OK    11.9 V    0.815 V    15.1 A (12.3 W)
3    OFF    OFF    11.9 V    0.825 V    0 A
4    ON    OK    12 V    0.745 V    38.6 A (28.8 W)
5    OFF    OFF    11.9 V    0.695 V    0 A
6    OFF    OFF    11.8 V    0.739 V    0 A
7    ON    OK    11.9 V    0.732 V    40.2 A (29.4 W)

Same here. 3 board with a '1' sticker on top. The fourth is marked with a '3'.
Bertmod show me that one my asic slot has a die with all 48 cores disabled.
So i thought that knc did this  because they didnt have enough well functioning
chips for everybody.

Last piece of info. All my board have 8 VRM. I'm not able to run .92 e .93.
.90 run only for the first few mins, then upgrade to .91. Both .94 and .95 with almoste the same perf

Same with my Jupiter, 3 good boards with white label #1 and 1 bad board with label #3 with one of the dies that has all 48 cores off.  So all 4 of us (mininganon, jelin1984, sickpig and me) seem to have the same issue.  

it seems so.

If I upgrade my Jupiter to anything other than firmware .91, then this entire board is deactivated and not hashing at all.  

my miners can't digest 0.92 and 0.93 due to the fact the cgminer restart every few secs.
0.94/95 both "works". The label #3 slot works as with 0.91 with a die completely disabled, but the other 3 work normally.

I think it is safe to say that it is confirmed KNC has been testing each hashing board separately and intentionally mount 1 bad board with 3 good boards per Jupiter.  Now the question is if there is anyway to reactivate some of the cores in the non-hashing die.

P.S. One more observation I notice is KNC seems to intentionally place the bad board close to the front of the case where 2 case fans are.  This might have done to ensure it has lower temp and a better chance to hash better.

yeah seems logical thing to do.

Unfortunately I wasn't aware of bertmod while using 0.91 so I had no data about number of disabled cores. The only thing I know about 0.90 and 0.91 is:

- 0.90 makes my jup drains almost 3.3 Ampere @ 220 V  ~>730W
- 0.91 makes it consumes a lot less energy (~ 2.2 Ampere)

[Paladin69]

I'm having the opposite problem with my bad Jupiter.

I have the three good ones have a sticker 2 on them.  The bad one is a sticker of 1.

My other two Jupiters running perfectly fine though have the same configuration.

1 2
2 2

I have 4 VRM's per board fyi.

[sickpig]

I'm having the opposite problem with my bad Jupiter.

I have the three good ones have a sticker 2 on them.  The bad one is a sticker of 1.

My other two Jupiters running perfectly fine though have the same configuration.

1 2
2 2

I have 4 VRM's per board fyi.

mine has 8 VRMs per board.

did you have any chance to use bertmod  to get an idea of cores status on your board?

if not look at this mini-guide to get asic status without the need of installing the bertmod fw

https://bitcointalk.org/index.php?topic=306969.msg3320948#msg3320948

[DPoS]

Major update: (a VRM comes back to life, lots of rambling, no sleep, sorry)

Since I really want to split up jupiters into saturns some day, I ran some more solo and paired tests.   It is true that when you run one or two modules they seem to fly.
It seems the controller board has some explainin' to do.   I was able to get ~270Gh paired and ~135Gh solo but when putting those three together it was a disaster.
Cores flopping on one chip that seems really picky when not running solo or paired. Instead of a total over ~400 it would be under ~360 or worse for the three boards

Then as I was putting everything back in resignation, setting up my auto-restart script again for .94 and settling for ~410Gh... I saw the dead VRM  show output current again (instead of < 1)

quickly I updated to .95 (yes that is what started this mess, but now I believe I could preserve it by not overclocking it anymore)
also, that board was a #1 sticker board which usually have perfect ASICs in them (hardly any cores go down) so it was good to see that VRM wake the hell up again and use more the 3 dies

and here I stand with ~480Gh running the safer .95 which was only getting ~360Gh before.
I am not back to the ~520 it was running overclocked, and it is tempting, but I'll just take the lower wattage and being close to what I had for now.

time to get some sleep


(ps - if you ever want to save a file on the miner, put it in /config since most everything gets rewritten on reboot)

[AFox]

You might want to give 0.93 a try.

I think that gives the best performance so far for Saturns.

I'm trying it out now, 0.93 seems good. I'll add it to my stats after running it 24h.

Well, I cannot add any stats, CGMiner was restarting every 2/3 hours. Looks like it's back to 0.94.
I have a board 1 and 2 with 4 VRMs per board.

[DPoS]

You might want to give 0.93 a try.

I think that gives the best performance so far for Saturns.

I'm trying it out now, 0.93 seems good. I'll add it to my stats after running it 24h.

Well, I cannot add any stats, CGMiner was restarting every 2/3 hours. Looks like it's back to 0.94.
I have a board 1 and 2 with 4 VRMs per board.

.93 restarts often if you have hardware issues..  .94 will just let you slowly go down.  That is why I wrote a auto-restart script that would run every hour or so when I was forced to run .94 instead of .93 (had a board with a dormant VRM that wouldn't let me run .93 anymore)

but it came back to life now (see above post) so I am running the safe and slow .95 since that miner has had stability issues.
 

[sickpig]

Major update: (a VRM comes back to life, lots of rambling, no sleep, sorry)

Since I really want to split up jupiters into saturns some day, I ran some more solo and paired tests.   It is true that when you run one or two modules they seem to fly.
It seems the controller board has some explainin' to do.   I was able to get ~270Gh paired and ~135Gh solo but when putting those three together it was a disaster.
Cores flopping on one chip that seems really picky when not running solo or paired. Instead of a total over ~400 it would be under ~360 or worse for the three boards

Then as I was putting everything back in resignation, setting up my auto-restart script again for .94 and settling for ~410Gh... I saw the dead VRM  show output current again (instead of < 1)

quickly I updated to .95 (yes that is what started this mess, but now I believe I could preserve it by not overclocking it anymore)
also, that board was a #1 sticker board which usually have perfect ASICs in them (hardly any cores go down) so it was good to see that VRM wake the hell up again and use more the 3 dies

and here I stand with ~480Gh running the safer .95 which was only getting ~360Gh before.
I am not back to the ~520 it was running overclocked, and it is tempting, but I'll just take the lower wattage and being close to what I had for now.

time to get some sleep


(ps - if you ever want to save a file on the miner, put it in /config since most everything gets rewritten on reboot)

by chance did you measure BeagleBloard load (using uptime) while you test the "Saturn" configuration?

[jelin1984]

sickpig


what psu you have for jupiter?

also what fimware now you run at jupiter?
you have 48 core off at board with sticker number 3?

[whisper]

I have this

[sickpig]

sickpig


what psu you have for jupiter?

also what fimware now you run at jupiter?
you have 48 core off at board with sticker number 3?

psu: cooler master v850
fw:   0.95
yes I have 48 core off at with sticker number 3

[sickpig]

I have this

ASIC slot 4   
Temperature sensor: 32.0 C

Die ID   Cores ON   Cores OFF   %
0   48   0   100
1   0   48   0
2   47   1   97.9
3   48   0   100

DC/DC ID   ON/OFF   Status   Input Voltage   Output Voltage   Output Current
0   ON   OK   11.9 V   0.738 V   38 A (28 W)
1   OFF   OFF   11.9 V   0.714 V   0 A
2   ON   OK   11.9 V   0.821 V   14.8 A (12.2 W)
3   OFF   OFF   11.9 V   0.816 V   0 A
4   ON   OK   11.9 V   0.742 V   38.5 A (28.6 W)
5   OFF   OFF   11.9 V   0.71 V   0 A
6   OFF   OFF   11.9 V   0.74 V   0 A
7   ON   OK   11.9 V   0.752 V   41.3 A (31.1 W)

[jelin1984]

how do 0.95 and only 48 core

when i install 0.95 firmware my jupiter
have all the asic chip off

you how do these with firmware 0.95( i mean only 48 core off)

[sickpig]

how do 0.95 and only 48 core

when i install 0.95 firmware my jupiter
have all the asic chip off

you how do these with firmware 0.95( i mean only 48 core off)

dunno really. it simply works like that. I didn't anything fancy just upgrade from .94 to .95.

anyway if your miner work better with 0.91 stick with it.

[jelin1984]

the core with 0.95 run asap

or you wait some hours before wake up the dead core?