Quote from: jtoomim on May 06, 2015, 09:30:52 AM

Quote from: JaredR26 on May 06, 2015, 07:54:28 AM

Temperature rise is relatively linear even if it doesn't feel like it and the damage caused is not linear. You noticed around ~95F in the cold aisle @ ~61F outside. On a 81F day, that would be ~115F(~46C). Maybe a bit more, but the damage to these miners(and why other miners were not similarly damaged) was primarily because of the airflow around them in relation to the S4's you mentioned, not primarily because of the intake temperatures.

My math might not be right, of course. It's just an educated guess.

I don't mean to offend you on this point; most of your work is very thorough. But your temperature estimations are pure guesswork. Even approximating temperature in the midst of high airflow/turbulence/heat generation situations is extremely difficult. Temperature/thermal airflow software is extremely expensive exactly because it is so hard to approximate. Most of those formulas and estimations regarding that aren't going to hold up to any critical analysis at all. Delta T estimations for a point in the past with multiple unknowns, without actual measurements from a velometer or thermometer are obviously going to be terrible at best. And since this guesswork isn't needed to draw the conclusions... why?

It was hot. Too hot. Definitely less than 100C. Less than 70C. More than 40C. We can agree there for sure. Even if we could calculate them, we don't need more accurate numbers based on the conclusions you covered or the ones I reached...

Quote

Quote from: JaredR26 on May 06, 2015, 07:54:28 AM

Temperature rise is relatively linear even if it doesn't feel like it and the damage caused is not linear.

The damage we observed was not linear.

Which would be why I said it is not linear...

Quote

Quote

Polyethylene has a glass transition temperature of -80 to -120C. Polyethylene is malleable at room temperature and cheap to mass produce, just like the S5 plastic, which makes it a reasonable assumption unless Bitmain can confirm. Regardless, the temperatures that caused the deformation & damage would have largely been due to the "stuck" stagnant air that you described, not a direct result of the hot aisle temperatures. (i.e., primary cause = pressure) Other minor information I now know that I can't share agrees with that.

I disagree. I do not see how air pressure alone could have caused this deformation.

I wasn't saying that air pressure did the bend, air pressure only influenced the direction of bend at best.

I posit that air pressure, primarily, caused the runaway temperatures localized to a certain point on the boards.

Think of it this way; take two S5 miners in a chill, 15C room. Point the exhausts at eachother directly and start moving them towards eachother. Eventually at a certain point, maybe around an inch apart or less, the restricted airflow between them will cause the boards to overheat and fry(add time); The air temperature of the room can exacerbate the problem and make it happen faster, but it is otherwise irrelevant to the mechanism. That is what I think happened from your description, but rather than an S5 inches away it was dozens of S4's blasting towards S5's from a few feet.

The increased local temperature that resulted from this, likely within 3-10cm from the exhaust(as you mentioned) is sufficient to explain the plastic warping and board failure without any other input. To explain the mechanism I'm envisioning I'll have to take a big leap- Ever seen perler bead art? Looks like 8-bit graphics. Anyway, the beads are put on a plastic pegboard and then ironed. They are a flat board with flat beads on them that are ironed on a flat surface. The result, if the user isn't careful, is a warped and useless pegboard. The heated plastic undergoes thermal expansion, but it is not uniform, and the plastic deforms. I couldn't find any pictures to show this, unfortunately. But the concept is exactly the same reason you can't mix aluminum and copper wiring connections; over time and through heat cycling, the aluminum will disconnect no matter how much torque you apply. Same mechanism.

Specifically to this case... I looked up the coefficient of thermal expansion for PE and found from 80F to 135F you get ~0.5% expansion. Putting that in a right triangle gives you a bowing outwards of about 2.5cm from a ~17.5cm plastic strip. You said you measured 2cm at worst, so that's about right.

None of that even requires excessive intake or exhaust temperatures(whether present or not). That, combined with other things I know and the difficulty of calculating or approximating exhaust/intake temperatures, leave me stating 'The cause was localized cooling system failure on the miner caused by high incoming exhaust pressure, exacerbated by the fact that it was "too hot".'

Quote

The air pressure differences typically seen at "reasonable" air velocities for an HVAC system (i.e. < 20 m/s) are much lower than you can apply with a finger. Typical HVAC system pressure differences are on the order of 100 Pa. Axial fans typically have static pressure capabilities around 300 Pa. If we are very pessimistic and say that the hot aisle was 600 Pa higher pressure than the cold aisle due to the two S4 fans (i.e. the S4s had zero airflow, 100% static pressure), then that would mean about 0.06 newtons per square centimeter, or 0.08 PSI. I just went over and tested, and using my finger, it takes about 20 pounds of force to permanently deform a S5 side panel at room temperature by about 3 mm. That would mean you'd need nearly 1 atmosphere of pressure difference across the panel to deform it from pressure alone. I think temperature must have been a large factor.

Polylactic acid has a glass transition temperature of 60°C to 65°C. What are the panels actually made of? I don't know. Also, even below the glass transition temperature, plastics will become softer and more plastic as temperature increases.