xyzzy099
Legendary
Offline
Activity: 1065
Merit: 1077
|
|
December 13, 2013, 02:53:09 PM |
|
Kaboom! http://s28.postimg.org/v71hnbm71/SAM_1801.jpg PSU: OCZ ZX 1250W. Found the Jupiter shut down. Tried to start it and nothing happened. Removed cables, put them back and poof fireworks. Waiting for my RMA. Hashing with 3 boards and not whining. Has anyone tested soy's theory about the 3.3v filter caps on the PSU being the issue causing max current to be delivered and blowing caps? I wanted to but I have no way of measuring more than 10A without frying my mulitimeter so its not a good test. By the by I received my RMA board back yesterday and everything works great! I did put a 2w 18Ohm resistor on a 3.3V line just in case but who knows really. Thanks again KnC! I would suggest if you shut down your PSU for anything, unplug it and give it 10 minutes before starting again, just in case. A bit of paranoia is better than straight up gambling I think. I'm not familiar with the theory you mention, but the only PSU connection to the hashing modules is +12V, so I don't see how caps on the 3.3V rail could cause this problem. Clearly the actual problem is in the KnC board design. If such current in-rush is possible with a standards-compliant PSU - and it is, as we have seen this with many different well-known PSUs now - their board should should be designed to manage it.
|
Libertarians: Diligently plotting to take over the world and leave you alone.
|
|
|
Miz4r
Legendary
Offline
Activity: 1246
Merit: 1000
|
|
December 13, 2013, 03:00:38 PM |
|
I've been messing around with the tuning suite on my October batch Jupiter and managed to get all the cores working now and hashing 6% faster @560 Gh/s avg at the pool. But I'm slightly worried about the voltage and amps one of my dies needs to be working optimally without shutting down some if its cores. Below is a pic of my most troublesome board, especially Die2 had problems with usually about 12-20 cores shutdown all the time. Now all cores are working fine, but is 0.873 V output voltage and 51.7 A safe to keep it running on 24/7? It's the only die that requires this much to be functioning normally.
|
Bitcoin = Gold on steroids
|
|
|
Puppet
Legendary
Offline
Activity: 980
Merit: 1040
|
|
December 13, 2013, 03:24:11 PM |
|
But the only way to force my miners to get IP addresses in a row is by setting them all static. Unfortunately the router decides to change that later on for no reason If you use static addresses, make sure you select addresses that are outside of the DHCP range.
|
|
|
|
Bogart
Legendary
Offline
Activity: 966
Merit: 1000
|
|
December 13, 2013, 04:02:10 PM |
|
Might sound like a really daft question, but when I RMA a board, do i include the full board with the heat stink attached, or separate it and send?
Just the board and nothing else Thanks, so i take the heat sink off the top and just mail the board? Any tips for what to package it in so it is secure? Bubble wrap it and then place it inside a bubble wrap envelope But before you do that, put the board inside an ESD-safe bag. I'd still use a small box rather than an envelope, personally, with bubble-wrap inside the box.
|
"All safe deposit boxes in banks or financial institutions have been sealed... and may only be opened in the presence of an agent of the I.R.S." - President F.D. Roosevelt, 1933
|
|
|
Cablez
Legendary
Offline
Activity: 1400
Merit: 1000
I owe my soul to the Bitcoin code...
|
|
December 13, 2013, 04:04:06 PM |
|
I'm not familiar with the theory you mention, but the only PSU connection to the hashing modules is +12V, so I don't see how caps on the 3.3V rail could cause this problem. Clearly the actual problem is in the KnC board design. If such current in-rush is possible with a standards-compliant PSU - and it is, as we have seen this with many different well-known PSUs now - their board should should be designed to manage it.
Here is what he had said on the matter. It is untested so do not take it as gospel. But in terms of losing a board for a week or more I think I will play it safe. For this reason I think the plugs having both a jumper and a power resistor to the 3v line are a very good bet. I know you followed instructions, but did you put in the jumper then fire up the supply and check that the voltage was there on the 12V, then shut down the supply, plug the power supply into the Jupiter, then switch on the supply?
If my theory that the current ramp, preventing current surge producing a voltage spike sufficient to blow the caps, the current ramp to ramp up current on the 12V line is actually controlled by a voltage ramp of 0 to 1V produced on the 3.3V line (because the 3.3V will be up to regulated voltage before the 5V or 12V) then putting in the jumper, turning on the supply to check if the jumper is correct and voltage being supplied, then shut down before plugging into Jupiter, since the supply expects that it is plugged into a motherboard having a load on the 3.3V line, it expects that when the supply is turned off the 3.3V line filter capacitors will drain out thru the motherboard. That can't happen without a motherboard or a resistor to ground from the 3.3V line as well as the jumper.
So, if you check the jumper was correctly placed by looking for 12V on the 12V line, then shut down, plugged the supply into the Jupiter, and turned it on, the supply 3.3V line never drained and still had full voltage on its filter caps and the timed ramp voltage that controlled the ramp for the 12v current had remained at max, as if already having timed out its ramp, and BANG, instant max current.
|
Tired of substandard power distribution in your ASIC setup??? Chris' Custom Cablez will get you sorted out right! No job too hard so PM me for a quote Check my products or ask a question here: https://bitcointalk.org/index.php?topic=74397.0
|
|
|
xyzzy099
Legendary
Offline
Activity: 1065
Merit: 1077
|
|
December 13, 2013, 04:20:28 PM |
|
I'm not familiar with the theory you mention, but the only PSU connection to the hashing modules is +12V, so I don't see how caps on the 3.3V rail could cause this problem. Clearly the actual problem is in the KnC board design. If such current in-rush is possible with a standards-compliant PSU - and it is, as we have seen this with many different well-known PSUs now - their board should should be designed to manage it.
Here is what he had said on the matter. It is untested so do not take it as gospel. But in terms of losing a board for a week or more I think I will play it safe. For this reason I think the plugs having both a jumper and a power resistor to the 3v line are a very good bet. I know you followed instructions, but did you put in the jumper then fire up the supply and check that the voltage was there on the 12V, then shut down the supply, plug the power supply into the Jupiter, then switch on the supply?
If my theory that the current ramp, preventing current surge producing a voltage spike sufficient to blow the caps, the current ramp to ramp up current on the 12V line is actually controlled by a voltage ramp of 0 to 1V produced on the 3.3V line (because the 3.3V will be up to regulated voltage before the 5V or 12V) then putting in the jumper, turning on the supply to check if the jumper is correct and voltage being supplied, then shut down before plugging into Jupiter, since the supply expects that it is plugged into a motherboard having a load on the 3.3V line, it expects that when the supply is turned off the 3.3V line filter capacitors will drain out thru the motherboard. That can't happen without a motherboard or a resistor to ground from the 3.3V line as well as the jumper.
So, if you check the jumper was correctly placed by looking for 12V on the 12V line, then shut down, plugged the supply into the Jupiter, and turned it on, the supply 3.3V line never drained and still had full voltage on its filter caps and the timed ramp voltage that controlled the ramp for the 12v current had remained at max, as if already having timed out its ramp, and BANG, instant max current.
I don't think it's possible for this to be correct. In most (all?) modern ATX PSUs, AC is converted to +12V, and the +12V rail provides the input power to the DC-DC converters that produce the 3.3V and all the other power rails, so it would not be possible for the 3.3V ramp to control the 12V ramp time.
|
Libertarians: Diligently plotting to take over the world and leave you alone.
|
|
|
DeathAndTaxes
Donator
Legendary
Offline
Activity: 1218
Merit: 1079
Gerald Davis
|
|
December 13, 2013, 04:34:02 PM |
|
Is anyone using Seasonic X1250W supplies for their November Jupiters? I found that after about a minute after bootup (when the Jupiter is about to start hashing), the PSU shuts down. I let the PSU reset itself (takes several minutes), try again, same results. Consistent and repeatable. I tried 2 separate X1250's and 2 separate November Jupiters with the same result.
Obviously 1250W should be plenty, and the Seasonic is a high-quality supply, so I'm not sure what is causing the shutdowns. I normally have the X1250's running 4 BFL SC Singles without any issue, and those take more power than a Jupiter.
I had some Corsair AX860's kicking around so I tried those for the Jupiters. They work well and I haven't had any trouble after 24 hours now.
I am using a SeaSonic 1250W with no issue. Shutdown is likely over current protection kicking in. 1) The SeaSonic has two PCIe connectors one one cable. Don't use two on one cable. You should have 4 connectors plugged into the power supply itself (plus mother board and molex connectors). That means you will have 8 PCIe connectors use only one from each pair. 2) Make sure the paper clip is making a good connection. It may sound stupid but you might want to try either a different paperclip or buying a cheap "PSU tester" to ensure you have a good connection. 3) Check the PCIe extension cables. On one of mine one of the pins was loose. Loose connection = resistant = higher current to deliver the same power to the ASIC. Another way to test this is unplug the PCIe extension cables and connect your power supply directly to the ASICs. I found putting the PSU on the rig lid sitting next to the rig body allows everything to fit. If you get no trips connected directly and it continues to trip when using the extension cables then one of them likely has a loose pin.
|
|
|
|
DeathAndTaxes
Donator
Legendary
Offline
Activity: 1218
Merit: 1079
Gerald Davis
|
|
December 13, 2013, 04:40:07 PM Last edit: December 13, 2013, 04:52:18 PM by DeathAndTaxes |
|
I don't think it's possible for this to be correct. In most (all?) modern ATX PSUs, AC is converted to +12V, and the +12V rail provides the input power to the DC-DC converters that produce the 3.3V and all the other power rails, so it would not be possible for the 3.3V ramp to control the 12V ramp time. I have pointed this out and I agree it doesn't make sense. The days of rectifying AC into all voltages directly haven't existed for years (probably a decade). Modern PSU are essentially 12V "engines" with the rest added as legacy support. This is why if you look at a modern PSU say a 1250W SeaSonic it can delivery 1248W on the 12V rail alone. All the power is being converted into 12 VDC and then if needed a sub board (and on the SeaSonic it literally is a completely different board) converts some of that 12V to 3.3V, 5V, 5VSB, and -12V if needed.
|
|
|
|
DeathAndTaxes
Donator
Legendary
Offline
Activity: 1218
Merit: 1079
Gerald Davis
|
|
December 13, 2013, 04:49:44 PM |
|
I know what it is and how it works, but my router implementation is poor. There is no option to setup the IP address myself. All I can do is select: Always use the same IP address Then I would say get a new router. Honestly even a cheap $30 router by any of the major brands (Dlink, netgear, etc) will have solid DHCP reservation support these days. Just check the reviews on a networking site before you buy one to make sure it will do what you want it to do. For example: Grr.. bitcointalk's image proxy is broke. Direct link: http://i.minus.com/jZPw0b7bBMH99.png
|
|
|
|
|
MANofthePEOPLE
|
|
December 13, 2013, 05:55:56 PM |
|
I was just about to order a black arrow x-3 but it seems that batch 1 is sold out and batch 2 won't be shipped until april. Anyone have an idea if it's the same for Knc?
|
|
|
|
sickpig
Legendary
Offline
Activity: 1260
Merit: 1008
|
|
December 13, 2013, 05:59:10 PM |
|
Sure nice improvments, they come quite late, though. Batch1 BJ customer have paid 50 btc for one unit long time ago. Just saying...
|
Bitcoin is a participatory system which ought to respect the right of self determinism of all of its users - Gregory Maxwell.
|
|
|
faetos
|
|
December 13, 2013, 06:04:57 PM |
|
32 neptunes left - I hope knc deliver in q1 and rain on all these mfs parades.
|
|
|
|
ASIC-K
Sr. Member
Offline
Activity: 280
Merit: 250
Hell?
|
|
December 13, 2013, 06:05:27 PM |
|
|
|
|
|
vesperwillow
|
|
December 13, 2013, 06:11:43 PM |
|
28nm 750gh now is >>>> than 20nm 3-6 months from now.
|
|
|
|
The Avenger
|
|
December 13, 2013, 06:12:02 PM |
|
That is one of knc's key selling point for knobtune, but what difference does it make if you can hash exactly the same TH on a well designed 28nm? Whoever suggested a few weeks back they should have just worked on a better 28nm design rather than 20nm was probably right. Let's be honest - the only reason knc are doing 20nm is to justify the ridiculous price.
|
"I am not The Avenger" 1AthxGvreWbkmtTXed6EQfjXMXXdSG7dD6
|
|
|
ASIC-K
Sr. Member
Offline
Activity: 280
Merit: 250
Hell?
|
|
December 13, 2013, 06:16:08 PM |
|
That is one of knc's key selling point for knobtune, but what difference does it make if you can hash exactly the same TH on a well designed 28nm? Whoever suggested a few weeks back they should have just worked on a better 28nm design rather than 20nm was probably right. i do agree, id rather have a tweaked 28nm now....ah well....if i wanted every single thing the way i like it, i would have started my own company, clearly thats not gonna happen so we depend on knc!
|
|
|
|
soy
Legendary
Offline
Activity: 1428
Merit: 1013
|
|
December 13, 2013, 06:20:59 PM |
|
Kaboom! http://s28.postimg.org/v71hnbm71/SAM_1801.jpg PSU: OCZ ZX 1250W. Found the Jupiter shut down. Tried to start it and nothing happened. Removed cables, put them back and poof fireworks. Waiting for my RMA. Hashing with 3 boards and not whining. Has anyone tested soy's theory about the 3.3v filter caps on the PSU being the issue causing max current to be delivered and blowing caps? I wanted to but I have no way of measuring more than 10A without frying my mulitimeter so its not a good test. By the by I received my RMA board back yesterday and everything works great! I did put a 2w 18Ohm resistor on a 3.3V line just in case but who knows really. Thanks again KnC! I would suggest if you shut down your PSU for anything, unplug it and give it 10 minutes before starting again, just in case. A bit of paranoia is better than straight up gambling I think. The ramping of current on the 12v line isn't theory but said by the tech rep from the VRM manufacturer. When one of two engineers designing a switching-servo-amplifier that I was breadboarding as a technician around 1977 designed the then novel switching scheme while our US company was a division of Kollmorgen, he having gotten his degree from Brooklyn Poly Tech, the triangle wave which to a comparitor input would see a lowered DC voltage on the other comparitor leg (that voltage the feedback) until the top of the triangle wave was encountered at which point the output would swing from one rail to the other for the duration of the interception. When the lower voltage (feedback) would reach the triangle wave and the output swing from say zero to full voltage, the controlled voltage would increase raising the feedback voltage level above the triangle. Typically to maintain an output voltage at some current the comparitor would have the feedback level at some stable level intercepting the triangle wave producing a square wave with the necessary duty cycle. The triangle wave was produced with a current source charging a capacitor then discharging the capacitor. Voltage charging a capacitor happens at a changing rate over five time constants. Current charging a capacitor, or current discharging a capacitor, will result in a non-curving slope; charge/discharge will give a triangle wave, not an approximation like if you only use the first time constant but a linear charge and discharge. That up-ramp of current on the 12v line would follow a straight line, a non-curving slope. That rise would be dictated by a voltage produced elsewhere by a current source charging a capacitor, that would not be a filter capacitor. There would be an internal filter capacitor on the 3.3v line. My theory is that the up ramp for the current would be perhaps between 0v and 1v produced using the 3.3v rail because the 3.3v rail would have established itself before the 5 volt rail or the 12 volt rail, and that with no load on the 3.3v rail, e.g. no motherboard in place and a jumper on the motherboard plug, the up-ramp charged capacitor might not be brought back to 0 volts but remain at 1 volt or the full current condition.
|
|
|
|
soy
Legendary
Offline
Activity: 1428
Merit: 1013
|
|
December 13, 2013, 06:30:45 PM |
|
Kaboom! http://s28.postimg.org/v71hnbm71/SAM_1801.jpg PSU: OCZ ZX 1250W. Found the Jupiter shut down. Tried to start it and nothing happened. Removed cables, put them back and poof fireworks. Waiting for my RMA. Hashing with 3 boards and not whining. Has anyone tested soy's theory about the 3.3v filter caps on the PSU being the issue causing max current to be delivered and blowing caps? I wanted to but I have no way of measuring more than 10A without frying my mulitimeter so its not a good test. By the by I received my RMA board back yesterday and everything works great! I did put a 2w 18Ohm resistor on a 3.3V line just in case but who knows really. Thanks again KnC! I would suggest if you shut down your PSU for anything, unplug it and give it 10 minutes before starting again, just in case. A bit of paranoia is better than straight up gambling I think. The ramping of current on the 12v line isn't theory but said by the tech rep from the VRM manufacturer. When one of two engineers designing a switching-servo-amplifier that I was breadboarding as a technician around 1977 designed the then novel switching scheme while our US company was a division of Kollmorgen, he having gotten his degree from Brooklyn Poly Tech, the triangle wave which to a comparitor input would see a lowered DC voltage on the other comparitor leg (that voltage the feedback) until the top of the triangle wave was encountered at which point the output would swing from one rail to the other for the duration of the interception. When the lower voltage (feedback) would reach the triangle wave and the output swing from say zero to full voltage, the controlled voltage would increase raising the feedback voltage level above the triangle. Typically to maintain an output voltage at some current the comparitor would have the feedback level at some stable level intercepting the triangle wave producing a square wave with the necessary duty cycle. The triangle wave was produced with a current source charging a capacitor then discharging the capacitor. Voltage charging a capacitor happens at a changing rate over five time constants. Current charging a capacitor, or current discharging a capacitor, will result in a non-curving slope; charge/discharge will give a triangle wave, not an approximation like if you only use the first time constant but a linear charge and discharge. That up-ramp of current on the 12v line would follow a straight line, a non-curving slope. That rise would be dictated by a voltage produced elsewhere by a current source charging a capacitor, that would not be a filter capacitor. There would be an internal filter capacitor on the 3.3v line. My theory is that the up ramp for the current would be perhaps between 0v and 1v produced using the 3.3v rail because the 3.3v rail would have established itself before the 5 volt rail or the 12 volt rail, and that with no load on the 3.3v rail, e.g. no motherboard in place and a jumper on the motherboard plug, the up-ramp charged capacitor might not be brought back to 0 volts but remain at 1 volt or the full current condition. One way to disprove that element of my theory, 3.3v rail getting stable before the 5v or 12 volt rails might be shown or not by someone with a nice fast sample and hold oscilloscope and a camera.
|
|
|
|
soy
Legendary
Offline
Activity: 1428
Merit: 1013
|
|
December 13, 2013, 06:31:34 PM |
|
Kaboom! http://s28.postimg.org/v71hnbm71/SAM_1801.jpg PSU: OCZ ZX 1250W. Found the Jupiter shut down. Tried to start it and nothing happened. Removed cables, put them back and poof fireworks. Waiting for my RMA. Hashing with 3 boards and not whining. Has anyone tested soy's theory about the 3.3v filter caps on the PSU being the issue causing max current to be delivered and blowing caps? I wanted to but I have no way of measuring more than 10A without frying my mulitimeter so its not a good test. By the by I received my RMA board back yesterday and everything works great! I did put a 2w 18Ohm resistor on a 3.3V line just in case but who knows really. Thanks again KnC! I would suggest if you shut down your PSU for anything, unplug it and give it 10 minutes before starting again, just in case. A bit of paranoia is better than straight up gambling I think. The ramping of current on the 12v line isn't theory but said by the tech rep from the VRM manufacturer. When one of two engineers designing a switching-servo-amplifier that I was breadboarding as a technician around 1977 designed the then novel switching scheme while our US company was a division of Kollmorgen, he having gotten his degree from Brooklyn Poly Tech, the triangle wave which to a comparitor input would see a lowered DC voltage on the other comparitor leg (that voltage the feedback) until the top of the triangle wave was encountered at which point the output would swing from one rail to the other for the duration of the interception. When the lower voltage (feedback) would reach the triangle wave and the output swing from say zero to full voltage, the controlled voltage would increase raising the feedback voltage level above the triangle. Typically to maintain an output voltage at some current the comparitor would have the feedback level at some stable level intercepting the triangle wave producing a square wave with the necessary duty cycle. The triangle wave was produced with a current source charging a capacitor then discharging the capacitor. Voltage charging a capacitor happens at a changing rate over five time constants. Current charging a capacitor, or current discharging a capacitor, will result in a non-curving slope; charge/discharge will give a triangle wave, not an approximation like if you only use the first time constant but a linear charge and discharge. That up-ramp of current on the 12v line would follow a straight line, a non-curving slope. That rise would be dictated by a voltage produced elsewhere by a current source charging a capacitor, that would not be a filter capacitor. There would be an internal filter capacitor on the 3.3v line. My theory is that the up ramp for the current would be perhaps between 0v and 1v produced using the 3.3v rail because the 3.3v rail would have established itself before the 5 volt rail or the 12 volt rail, and that with no load on the 3.3v rail, e.g. no motherboard in place and a jumper on the motherboard plug, the up-ramp charged capacitor might not be brought back to 0 volts but remain at 1 volt or the full current condition. One way to disprove that element of my theory, 3.3v rail getting stable before the 5v or 12 volt rails might be shown or not by someone with a nice fast sample and hold oscilloscope and a camera. But then that would only show the test supply on hand and there are many switching supplies out there.
|
|
|
|
|