Topic: [CLOSED] ASICMiner Prisma 1.4th/s - 1.47 BTC - page 8. (Read 49842 times)

1225Gh. I have a cooling duct aimed right at them.

What is your hash rate @ 210?

I haven't gone past 210. Might just keep it. That way.

Please move that prisma.  You are asking for trouble with it so close to wood.  Too many have caught on fire.

well my unit is in my garage.  on a woodbench near the garage door.

it is 40 f outside in NJ  door is open with a fan puling in air



note temps below  not sure it that sensor is in the rasp pi or on the miner. I will now shut the  garage door.  it is 11:11:11 am in NJ



door open fan blowing in.   

I think it is a good idea. This is what I have been doing.

What if you put a box fan infront of the miner?  Box fans are rated for ~2,000 CFM and that should help with cooling.  Maybe put some baffles to direct airflow around the miner?

Your calcs are correct if the Prisma's draw as advertised.
They don't.
@ 240 clock rate ours (2x) are drawing ~1250W each.
When OC'd to 250 they draw 1300W each.
When OC'd to 270 they draw ~1400W each.

We are powering one with a DPS1600BB (using your most excellent adapter board) and the other with 2x HP 1000's.
Both are being feed by 8 16ga PCie cables because in our initial test using 4 cables they got more than warm to the touch.

The thought of a PSU failing on the fan powering board has been scaring me for a while. I will be adding a secondary fan soon just in case but also rewiring the PSU's so a failure on the fan powering board shuts everything down. I can understand AM's viewpoint of the miners not being adequately cooled. They have little knowledge of my setup and I wasn't watching the miners as they failed so it's just best guess as to what happened. All I can say is there was a significant amount of airflow, the incident happened during the peak temperature of the day which was about 75f, there was likely some mixing of outside air with the nearby GPU miners but I think intake temperature would have been 80f at max for the prisma's.

80f isn't ideal for all datacenter conditions but even my in house servers with dedicated AC are only seeing 78. In reality 80f isn't' unheard of for servers. Even Intel has experimented with higher temps for their servers. Realistically, 80f isn't that bad when you consider that most miners are trying for lowest cost which often means cooling miners with outside air vs AC. Take everything as you will. If you are concerned with non conductive barriers, you can get paver stones from your local home improvement store for just a few dollars. It's basically a slab of cement which will absorb a fair amount of heat and has minimal conductivity. You wouldn't have to worry about an arc flash not that it's likely to happen anyways. Just to clear up another question on my setup, there were no signs of arching, there is only some slight surface rust on the scrap sheet metal which has been sitting for years after making a spacer and swapping a transmission into a vehicle it didn't belong in.

In the end, make your own opinions, this is just a engineering perspective from someone that works on 14nm and larger nodes.

If you were a participant of my group buy, please check your PM. Also, I have received contact from AM tonight with a few recommendations.

- Run in a cool environment.
- Use a PSU with overcurrent protection
- Do not run your unit near combustible materials
- Prevent fan failure.

ASICMINER did not mention a second fan, but if you are powering with two PSUs, it might not be a bad idea. Consider the case where two PSUS are powering a unit and the one that is powering the fan trips or fails. At that point, you may have multiple boards still running but with no fan. This actually happened to me, but because I forgot to plug the fan back into the board after messing with dipswitches. I had outside airflow, but the unit eventually just stopped hashing, no meltdown, and it came right back up when I plugged the fan back in. I'm not certain that would always be the case.

I also spoke with a number of engineers and one observation stood out. Most of the heat from each be chip goes downwards towards the heatsink, but there is a small amount that goes outwards towards the surface of the board. If you are running in a hot environment, this heat could build up and be absorbed by other components, capacitors, etc...

 try a side fan  let it blow air at the side of the units in the middle area.  I am keeping my unit at freq 230 not 240

very good numbers at freq 230   1396     3% errors  power is 1215 watts that is .87 watts at the plug, but I run the added delta fan on the unit the side fans and some other fans .  the extra fans use 15 watts or more.

I'm gonna be setting up some Prismas to host in the next day or two, probably on DPS2K PSUs. Definitely not in load-balanced, which shouldn't be necessary since these boards all have isolated 12V lines. A pair of Prismas shouldn't quite max out a DPS2K, maybe 110% rated. It wouldn't take much of a short to run up to OCP (about 120% rated). Hopefully we don't test that hypothesis with a "live-fire" experiment.

Thanks for the info on what's been going on.

 making lite of it since no one has been hurt.  

I think I see a problem with using your psu setup if 1 unit goes bad and sucks down a lot of power it is sucking on 4000 watts since it can access both 2000 watt psu's.  In my case I am attached to a 1600 watt psu   the evga 1600 p2  if my unit goes bad it only can access 1600 watts to suck in.

 So I am thinking  no one should use 1 pair of 1000 watt psu's or larger with these.

 ie  if you use 2 psu's  use a pair of 750 watters  both on the same power circuit    thus if the unit fucks up it can only access 1 power circuit and 2x  750 watt psu's

It may be the flaw in the design is if 1 chip weakens    the units will pull more  watts then it needs if it is there and available.

 thus  having 2 x 2000 watt psu's attached  the unit tossed thousands of extra watts at the bad chip to make it hash and popped a cap.

Thanks  to you for  letting us know you had 2 x 2000 psu's .  

Here are some more bright ideas. If you go to home depot you can buy very inexpensive wire shelves and put them under your miner (this one is tan, but they have metal color ones). I may actually elevate this more with a few bricks... You can also bend out the metal on the outsides to raise up the Prisma to get more air underneath, especially if you have fans on the side of the miner. I put my miner on concrete, there is some plastic next to the miner in this photo and also some Blue Max waterproofing (which works great by the way) But I will put a server case door (thin metal) on that side(not too close of course!)

Perhaps with the metal bent out like this, we could make a screen frame. I also bought aluminum L shaped framing material from home depot.

Hmm, that's really not a bad idea.  (Assuming a hot cap couldn't just melt its way through the mesh -- screen doors are pretty thin stuff, right?  

But that's the easiest and cheapest idea I've heard so far...a lot better than buying $50 of sheet metal and having to return it to Home Depot, with what I assume my excuse will be: "I'm sorry I bought this but I needed nonmetallic...sheet metal.  So this won't work.  Because it's metal."



Two IBM/Delta DPS-2000BB server PSUs, operating in current-share mode (so two 2000W PSUs acting as one big 4000W PSU, which should have been more than plenty for the 3 Prismas, IMO).   I should point out that in my case, no circuits blew -- until I threw the breaker and killed the subpanel that feeds my mining stuff.  It was just happily mining and burning along...and as I told CrazyGuy, I noticed (just because I was near my laptop at the time) that cgminer didn't show any errors or unusual activity when the alarm went off.  (No, I don't expect cgminer to somehow say "look out the board is on fire!" -- but I would have expected melted burning charred exploding chips to ...you know...throw the occasional HW error, lol.)

And I take offense at the term "Cap Popper."  I demand you call us "People who through no fault of their own became victims of cap popping phenomena" ... nah just call it whatever, I'm kidding of course.

anyone ?

Minera up and running on a RPI B+ 32 GB memory card with no problems with the image from Crazyguy. I have to thank you Crazyguy for not only making this a great GB for us but for also going that extra mile and then some for me especially. If anyone was ever on the fence about doing any deals or participating in any GB with Crazyguy you really would be doing yourself a disservice if you hesitated in any way whatsoever. Thank you Crazyguy for the refund that you pulled out of your own pocket for all of is because of delayed shipping. Now has anyone been able to get  http://www.mobileminerapp.com up and running yet with this image? That's the only thing I'm hung up on right now. Thanks again Crazyguy

Thank you for taking the time to explain the concept -- I was familiar with "thermal runaway" and "runaway brides" but not "current runaway."  So this was very educational for me, so thanks for posting it. 

That said, the explanation (above) strikes me so far as the most likely cause...in both fires. I'm basing this not just on "looks" but also on the fact that I recall Asicminer informing shareholders (last week maybe?) of a "Chip Popping" problem that had plagued the BE200 chip, causing failures for various OEMs --  so please correct me if  I'm way off base, but is it possible this "chip popping" could trigger the current runaway you described?

And if that *is* true, then (not trying to cause a panic, just asking the question) -- if there's no known way to identify chips with this defect, and they can run for several days/weeks before they "pop" -- is there any way a regular user (without access to absurdly advanced tools etc) could detect or otherwise know if their unit is at risk?  Wouldn't that mean (to some degree) that all Prismas are at risk for this kind of behavior?


BTW: Please note, I am not trying to push an agenda -- my 2 RMA Prismas are arriving today, and I still have one running (the one that wasn't damaged by the fire) so I'm not trying to worry people or discourage new sales or otherwise hurt ASICMiner.  I honestly just don't know the answer to the questions above, and they seem significant, so I thought I'd ask.  Not a fanboy, not a hater.  Just a miner.

EDIT:  One other question:  I had all my miners connected via the USB/UART connectors.  But I realize I could have "daisy chained" them together...  This is probably a dumb question but -- when you daisy chain boards, is it kind of like "extending the one string" ... meaning, could a chip pop on one board and then just follow that "string" back from one board to another and another, damaging more than one board (or more than one Prisma)?  That's my main question -- because if daisy chaining makes it possible for ONE popped chip to potentially damage everything that's connected to it, it seems to me like it might be safer to just connect each prisma to a USB hub (rather than chain them together and the connect the last one via USB).  Or is that just such an unlikely scenario that it's not worth worrying about?

how many PRIMAS can be daisy chained per controller ?
how many can be chained per wart port on controller ?
thank you !