The moment of truth has arrived. Now, we're going to see what the Everest is like in relation to the older Epsilons. As is the norm for me, I'm going to use a SunMoon SM-268 programmable electronic load to test the Everest and see whether it turns out to be a mountain of merde or a towering monolith of stability.

The SM-268 is able to load test in up to five programmable loads, and also manually override those loads if needed. What I'm going to do first is program it for five progressive balanced loads going from 20% to 100% of full power. These will be done within the constraints of the specified maximum loads in the specs of the unit. Then, I'll crossload the unit with a high 3.3V/5V load and then a high 12V load - this will show us how well the unit can keep stable voltages, or regulate, when faced with unusual loading patterns.

When that's all done, I'll roast the unit in my hot box using the same tests and see what effect heat has on the whole thing. We'll also see whether that de-rating curve holds water.

While the Everest doesn't do too badly in the above table, there are a number of things up there that underwhelm me. First, the regulation (voltage stability) isn't too great. 12V sees the biggest drop, heading down to a mediocre 11.72V by test five. This, combined with the stunningly bad voltage readings in tests CL1 and CL2 point to a cheaper group regulated design. Just like the old... ugh... Epsilon.

However, these results are at a contrast to relatively high efficiency, which was always a good point about the Epsilon. They might have turned in less than stellar performance, but at least they were efficient. But, we're still barely hitting 80% at test five, which tells me that perhaps the Everest is straining a bit at that level. This does not bode well for the hot box, I'm afraid.

Overall, the unit does appear to be able to meet its 900W at 25 degree rating. But, most computers don't run at 25 degrees. My hot box simulates a running computer by venting the heat output of the SunMoon back into the unit being tested. So, let's get that started and see what happens.

You will recall that on page one, this unit's de-rating curve allows it to hit 850W at 40 degrees. See test five's temps? We're right there, baby. And the Everest, to its credit, is still running, with exhaust temps hitting a toasty 66 degrees.

But, all is not well here. While the voltage readings aren't too different from the cold tests, and the efficiency doesn't suffer too much, the unit actually failed the hot box testing. How can it have failed, you ask, when all the numbers in the table are there and it was still running? Well, before I answer that I need to show you some scope shots.

Why yes, Virginia, the 3.3V and 5V are indeed exceeding the ATX spec. At up to 70mV of ripple, those two rails are handing in amazingly out of spec (50mV) results. But, to my shock, 12V ripple and noise is under 50mV at all times. It was at 12V the original Epsilon design had its noise issues. Now, it would seem the buck has been passed to the 3.3V and 5V rails. Crappy.

I did mention failure, didn't I? Up until the end of test five in the hot box, old Everest was doing pretty well. I was getting some mediocre performance out of it, but nothing unexpected. And then, I started test CL1. Immediately, the unit began to squeal really loud, like a ten year old girl being given a free pony. I looked over at the scope display, and just about jumped out of my skin when I flicked over to the 5V and 3.3V rails. Case in point:

Yes, people, that there is the 5V rail in hot box test CL1. I turned the cursors on to show you just how sucktastic this waveform is. That's over 400mV of noise, folks. This is NOT good. Think about it this way - the 5V output is fluctuating by about 0.4V several hundred times a second. Your expensive hardware will not like this. This will produce excess heat in any component pulling from 5V, like that 1.5 TB hard drive you just bought, leading to possible premature failure. And guess what runs off the 3.3V rail? Right - your 8 GB of brand new RAM. It won't like noise like this either.

And what really ticked me off was, the unit did not shut down. It just kept going and going until I moved on to test CL2, where the 3.3V and 5V rails dropped back down to normal. That is, if you can still call 70mV of noise normal. Less terrifying, I guess I should put it.

Clearly, something in the unit failed to produce that jaw dropping 400mV+ result, however, as this did not happen in the cold tests. Let's take it apart and find out what.