Pages:
Author

Topic: KnCMiner Jupiter Miner First Impressions - page 13. (Read 65993 times)

sr. member
Activity: 462
Merit: 250
October 14, 2013, 01:51:40 PM
Those regulators are speced for a max current of 40A each....
Has anyone pointed out that's a good suspect for the VRMs and caps failing?
The capacitor were failing because the hashicg processors are loded in pulses with interruptios, as teh work is momentarily interrupted, the ATX PUS (Corsair) in question causes a spike in the supply voltage, which is short, but deadly. A combination of mechanical load, load current and tenmperature causes ultimately electrical breakdown of the input MLC capacitor.

Please not that only one specific PSU was doing this. The named ATX PSU was able to blow up a board that was completely OK with the coolermaster PSU.

please explain what I just posted.. then please leave this thread
full member
Activity: 151
Merit: 100
October 14, 2013, 01:50:57 PM
Those regulators are speced for a max current of 40A each....
Has anyone pointed out that's a good suspect for the VRMs and caps failing?
The capacitor were failing because the hashicg processors are loded in pulses with interruptios, as teh work is momentarily interrupted, the ATX PUS (Corsair) in question causes a spike in the supply voltage, which is short, but deadly. A combination of mechanical load, load current and tenmperature causes ultimately electrical breakdown of the input MLC capacitor.

Please not that only one specific PSU was doing this. The named ATX PSU was able to blow up a board that was completely OK with the coolermaster PSU.
sr. member
Activity: 462
Merit: 250
October 14, 2013, 01:50:12 PM
explain this big guy



pick which one was 'upgraded' to .95 and which one is still running .93 and overclocked

here is the one overclocked (I don't care how you define it, it is baking, running hot, or whatever, it is PRODUCING better)



and you can see it has one board with 8 VRMs and running cooler with .93 while the other 4 VRM boards are running hotter

so leave this thread ok?

full member
Activity: 151
Merit: 100
October 14, 2013, 01:46:37 PM
.95 firmware lowered the voltage on the 4 VRM boards which prior to its release, were left to overclock after 4 were removed without any testings of consequence.

The performance of .93 & .94 outshine the performance of .95 on boards with 4 VRMs.  I have posted those pictures as well and others have switched back too.  People rather run the boards with higher voltage/amps (overclocking but you hate that term) and cool them to prevent as much of them failing than to run a gimped .95 that does nothing to keep more cores running.

You get more WU with .93 & .94 than with .95.  The cooling helps reduce HW errors regardless if you ignore that reality.
Dear sir, may I direct you to some dictionary that will explain to you what a clock signal is?
All the boards have the same clock and PLL settings, trust me. You are using the word "overclock" in a mistaken belief of its meaning.

The individual chips exhibit up to 40% process variations, which is normal.

What the error is is that the VRM modules can be programmed to a certain output voltage which they fail to respond correctly to.

Please, study the behaviour and manufacture of MOS transistors to see the error in your thinking, it is impossible to  manufacture identical chips, chips across the die will exhibit different capacitance and different speed, both independently.

You may understand it easier when I remind you that the chips are not sorted in ANY way prior to package assembly, thus random chips are selected (but most likely in serial order in which those were on the die.)

Once again, the chips were never tested or put into different batches. That was to save us week(s) of production time, not to mention costs and development costs of the testing scripts.

When you ordered from KnC, all of this was already public knowledge. If you are unhappy with your unit you are free to sell it for more than you have bought it -- on e-bay for example.

P.S. Many people report way higher production rate under the 0.95 BIOS. So you assume that just something works in your set-up better that means the whole world revolves around you. Wrong, sir.

P.P.S. Would you like some overclocked Fish&chips?
legendary
Activity: 1792
Merit: 1047
October 14, 2013, 01:24:50 PM
As you can tell the performance is best with v0.9.3 with the case off and an the added fan only if you can maintain an ambient temp.

Below is the comparison on work accepted comparing v0.9.4 and v0.9.5 with case cover off with added house fan pushing air downward in the center of the unit with ambient temp of 17°C.

Included is the results after 13 hours with case cover on with the two intake fans swapped out running in an ambient temp of 19°C.

13hours - Case On - new intake fans - ambient room temp 17°C Firmware v0.9.4 Results


Compare

13hours - Case On - new intake fans - ambient room temp 17°C Firmware v0.9.5 Results


Compare

13hours - Case On - new intake fans - ambient room temp 19°C Firmware v0.9.5 Results





In order to take advantage of the new higher CFM Fans.
For the remaining 11 hours, I will lower the ambient room temp to sub 17°C. This will help gauge if the lower temperature within the case while the case cover is on will increase performance.

If not the the simple solution is to keep the cover off and add the house fan with firmware v0.9.3 as that has the best results if you are able to maintain an ambient temp at or below 20°C.
legendary
Activity: 1792
Merit: 1047
October 14, 2013, 12:43:51 PM
Performance update.

12 Hour hour mark

Firmware v0.9.5 / Case cover on / New 175 CFM Fans / 8 VRM Jupiter / Ambient Room Temp 19°C

Airflow intake and exhaust has increased considerably from upgrading the two stock system fans.

The results




12 Hours case on with new 175 CFM 140mm fans Firmware v0.9.5


Compare

12 Hours open case with added house fan ambient room temp 17°C Firmware v0.9.5


Link to more screenshots for past results.

https://bitcointalksearch.org/topic/m.3324429
sr. member
Activity: 462
Merit: 250
October 14, 2013, 11:50:44 AM
that is pretty nice,  shame you had to make adjustments but looks good!
legendary
Activity: 1792
Merit: 1047
October 14, 2013, 11:44:34 AM

Here is a few photos of the process and how I got the 120mm mounting holes to fit the 140mm fan slots on the Jupiter case.




I forgot that you switched to 120..playing with that grill is a pain...
  I did put different 140's in the front of mine

I used these

http://www.microcenter.com/product/351003/Shark_Blue_Edition_140mm_Case_Fan

Speed    1,500 RPM
Airflow    96.5 CFM
Maximum Air Pressure    1.069 mm-H2O
Noise Level    29.6 dBA

These ones have a switch that puts them operating
Speed 2,000 RPM
Airflow 175 CFM

I do not notice the noise at all.

I compared these with the Avalon ones and these moved just a tad less air. However the air ventilation from the back of the case with these fans on is noticeable and more then 4x the cfm then the flimsy thin plastic stock fans that came with the unit.

Well the fans themselves are 140mm. However the location of the holes are that of a 120mm model fan. The 120mm holes on these 140mm fans are in the proper place when making use of tie wraps to secure the fans to the case.
sr. member
Activity: 462
Merit: 250
October 14, 2013, 11:41:38 AM

Here is a few photos of the process and how I got the 120mm mounting holes to fit the 140mm fan slots on the Jupiter case.




I forgot that you switched to 120..playing with that grill is a pain...
  I did put different 140's in the front of mine

I used these

http://www.microcenter.com/product/351003/Shark_Blue_Edition_140mm_Case_Fan

Speed    1,500 RPM
Airflow    96.5 CFM
Maximum Air Pressure    1.069 mm-H2O
Noise Level    29.6 dBA
legendary
Activity: 1792
Merit: 1047
October 14, 2013, 11:40:23 AM
One thing that I forgot to mention was that while I was trying single and paired (mercury/saturn) configurations, the board that had the bad VRM was off and this time was against the a/c unit airflow I have the miner positioned.  It was off for 30 minutes while being chilled (it was usually on the far side, but I switched it by chance to the near side)

And then when I put it back on with the rest, the VRM awakens on next boot.    Coincidence??  maybe, but happy either way


(why was I taking boards off my miner you ask? I was planning on cutting the miner in half for better airflow but didn't realize that the bottom is way thicker than the lid which cuts easily..  so I had to ditch those plans for now as I rethink)




Coming up on 12 hour mark with case on and operating with the two new intake fans.

I will be posting this new result with previous tests so you can see the performance difference on firmware v0.9.4 and v0.9.5 with case cover off and house fan on.

Here is a few photos of the process and how I got the 120mm mounting holes to fit the 140mm fan slots on the Jupiter case.






legendary
Activity: 2408
Merit: 1004
October 14, 2013, 11:39:38 AM
you how wake up your system
maybe is the same with me
sr. member
Activity: 462
Merit: 250
October 14, 2013, 11:37:47 AM
any step by steps

details how wake up my 48 core off? Smiley Smiley Smiley Smiley Smiley Smiley

if it is showing cores off I don't know.  the one i got back never said any cores was off, but showed < 1 amps (and testing showed no cores were working)
So i didn't have a bad die/cores for that one, just a faulty VRM not providing for it that now is back
legendary
Activity: 1792
Merit: 1047
October 14, 2013, 11:37:01 AM
One thing that I forgot to mention was that while I was trying single and paired (mercury/saturn) configurations, the board that had the bad VRM was off and this time was against the a/c unit airflow I have the miner positioned.  It was off for 30 minutes while being chilled (it was usually on the far side, but I switched it by chance to the near side)

And then when I put it back on with the rest, the VRM awakens on next boot.    Coincidence??  maybe, but happy either way


(why was I taking boards off my miner you ask? I was planning on cutting the miner in half for better airflow but didn't realize that the bottom is way thicker than the lid which cuts easily..  so I had to ditch those plans for now as I rethink)




DPoS sorry for spamming you but if you'll test again your "split my jup to get two saturns" would you mind to check beagleboard system load using uptime command from a ssh session?

I have a theory that implies that the lower perf we get on jupiters configuration is also due to some bugs present in cgminer driver developed by knc that makes the system waste resources in I/O waiting.

Edit: i would it if i had constant phisycal access to my miner

This is what I got when I did a software reset via the web gui.



One can speculate that the 32 bit is inadequate for the work load?

Not sure but perhaps it is limited on the amount of available bits that can is causing the issue? Maybe the result of a software bottleneck?
legendary
Activity: 2408
Merit: 1004
October 14, 2013, 11:35:41 AM
any step by steps

details how wake up my 48 core off? Smiley Smiley Smiley Smiley Smiley Smiley
sr. member
Activity: 462
Merit: 250
October 14, 2013, 11:34:06 AM

DPoS sorry for spamming you but if you'll test again your "split my jup to get two saturns" would you mind to check beagleboard system load using uptime command from a ssh session?

I have a theory that implies that the lower perf we get on jupiters configuration is also due to some bugs present in cgminer driver developed by knc that makes the system waste resources in I/O waiting.

Edit: i would it if i had constant phisycal access to my miner

sounds reasonable, wish i did that last night

I will start taking some data while it is in jupiter mode so we can see any differences for next time


ps - i made a thread showing my simple auto-restart script if anyone needs it   https://bitcointalksearch.org/topic/m.3337175
legendary
Activity: 1260
Merit: 1008
October 14, 2013, 11:23:37 AM
One thing that I forgot to mention was that while I was trying single and paired (mercury/saturn) configurations, the board that had the bad VRM was off and this time was against the a/c unit airflow I have the miner positioned.  It was off for 30 minutes while being chilled (it was usually on the far side, but I switched it by chance to the near side)

And then when I put it back on with the rest, the VRM awakens on next boot.    Coincidence??  maybe, but happy either way


(why was I taking boards off my miner you ask? I was planning on cutting the miner in half for better airflow but didn't realize that the bottom is way thicker than the lid which cuts easily..  so I had to ditch those plans for now as I rethink)




DPoS sorry for spamming you but if you'll test again your "split my jup to get two saturns" would you mind to check beagleboard system load using uptime command from a ssh session?

I have a theory that implies that the lower perf we get on jupiters configuration is also due to some bugs present in cgminer driver developed by knc that makes the system waste resources in I/O waiting.

Edit: i would it if i had constant phisycal access to my miner
sr. member
Activity: 462
Merit: 250
October 14, 2013, 11:13:51 AM
One thing that I forgot to mention was that while I was trying single and paired (mercury/saturn) configurations, the board that had the bad VRM was off and this time was against the a/c unit airflow I have the miner positioned.  It was off for 30 minutes while being chilled (it was usually on the far side, but I switched it by chance to the near side)

And then when I put it back on with the rest, the VRM awakens on next boot.    Coincidence??  maybe, but happy either way


(why was I taking boards off my miner you ask? I was planning on cutting the miner in half for better airflow but didn't realize that the bottom is way thicker than the lid which cuts easily..  so I had to ditch those plans for now as I rethink)


legendary
Activity: 1260
Merit: 1008
October 14, 2013, 08:03:35 AM
has a single KNC rep even acknowledged this thread???

I really don't know, and I think that today they have other priorities, namely being able to meet the October 15th deadline and solve the problems they had with KnC's pool.

Notwithstanding this morning I've sent an email asking to clarify a few issues we discovered in this thread, let see if and what they reply.

As soon as I get something I'll share it here.

legendary
Activity: 1848
Merit: 1001
October 14, 2013, 07:57:30 AM
has a single KNC rep even acknowledged this thread???
legendary
Activity: 1260
Merit: 1008
October 14, 2013, 07:17:36 AM
tracking numbers are being issued now... 

wrong thread ?
Pages:
Jump to: