SPCR

What kind of results are folks seeing with the PicoPSU in terms of reliability? I've been using one to run my Atom-based file server and have had two burn out in 9 months. I'm wondering if this is an anomaly or if the 24x7 duty cycle is just too much for it...

I think I have heard a mention or two of short life. I was wondering about that.

I was one of the early adopters when the original SPCR article came out. I've bought 3 of them in total, all 80w with the 80w EDAC brick. All are still going strong after 3-4 years of use and abuse. Easily the best PC part I have ever bought, as it never needs upgrading.

I have, however, made a point of sticking by my 80w limit despite upgrading various other components in search of more performance, perhaps the issues relate to overheating when people push the more poweful models?

Unfortunately I have to RMA the whole thing, so I don't know if it's the circuit or the brick that's failing. I could understand if there was a problem working it too hard, but my server is an Atom-based Ubuntu system with nothing but samba and sftp running (no GUI, etc.), integrated graphics, one stick of value RAM, one HDD, no optical, no USB, and one Nexus fan. If a that is too much for a PSU rated at 90W, then there are some serious issues with the design.

The main reason I'm asking is that I was thinking about rebuilding using the Antec ISK-300. I really like the look & design of the case, but given the issues I've had with the Pico I'm a little hesitant to go with a mini DC-DC PSU again.

OmegaZero wrote:Unfortunately I have to RMA the whole thing, so I don't know if it's the circuit or the brick that's failing.

Get the Winmate.

Both my main system and my disk server are powered by PicoPSUs. I've had no problem with them.

I used for a while pico 90 and pico 120 in two rigs and all was fine. Actually my main pc is powered by a pico 120-wi, and all is OK.

I had a problem only with a pico 80-wi because it could not power a 2.5" HDD to spin on boot.

Eunos, was that a typo? The 80W unit are newer, and don't have the 2x2 12V auxiliary ATX mobo connector. I wonder if there are design differences between the 80W and 90W units that lead to a difference in longevity.

Gee, I'm pretty hopeless for a Pico advocate.

I just referred to the original article, and mine is indeed a 120w Pico (the only option at the time) with the 80w EDAC brick. I tried the 120w EDAC but it belongs only at DeafeningPCReview.com.

Eunos wrote:I tried the 120w EDAC but it belongs only at DeafeningPCReview.com.

The fan kicks in at about 80W though. So you could see it as a fanless 80W brick with "overdrive".

I had a system (42 watts idle, 50-60% of the time on) with a pico-90 that ran fine for 3 years. Just build a system with pico-150 two months ago, worked fine so far.

What kind of power draw could an atom based file server really put on them? Lots of HDDs? maybe the 3.3/5v is being over-stressed? I've had a Pico powering my sig system of well over a year, and it was operating on a similarly power hungry system for over a year before that. Never had any problems. I only use 1 hdd, I imagine most of the loading in my system is on the 12v lines.

How do you guys feel now a few years later?

I just ordered the 80w DC-DC PSU with 60w ac adapter kit for a pfSense build I am working on, and I am wondering if this might have been a mistake?

Eunos wrote:I tried the 120w EDAC but it belongs only at DeafeningPCReview.com.

LOL, you obviously haven't heard the HP DL180G6 server I bought used on eBay a couple of years ago.

It's jokingly been called the "HP DL180 Dreamliner" by those home server hobbyists among us who have accidentally bought it :p

I've seen someone hold up a decibel meter in front of one of these and measure about 100db!

Even though mine was going in my basement, I wound up not being able to use it. I could hear it two floors up in my bed room even with all the doors shut!

mattlach wrote:How do you guys feel now a few years later?

My picoPSU 150XT + 150W brick still running fine after 3 years, no issues, now powering an i7 setup with 50%-100% load 24/7.

Abula wrote:

mattlach wrote:How do you guys feel now a few years later?

My picoPSU 150XT + 150W brick still running fine after 3 years, no issues, now powering an i7 setup with 50%-100% load 24/7.

Thank you!

That is good to know. I had heard some comments on another forum that the Pico PSU's were failure prone, and tended to take motherboards and other components with them when they went. I didn't know if the person was speaking from bias or from actual experience.

Note that PSUs vary in their susceptibility to power transients. A PSU that dies after living on a circuit prone to power disruptions isn't necessarily unreliable ... just picky about the conditions in which it's run. My Picos run off a UPS with AVR and pure sine wave output, which might be related to their longevity.

FWIW personal experience is biased towards that person, i.e. it's only anecdotal evidence and not necessarily indicative of a larger trend. I've had two smaller units running for awhile now, on systems that only need to be rebooted for kernel updates.

(Wow ... i last posted on SPCR in 2010. Those were the days.)

kirbysdl wrote:Note that PSUs vary in their susceptibility to power transients. A PSU that dies after living on a circuit prone to power disruptions isn't necessarily unreliable ... just picky about the conditions in which it's run. My Picos run off a UPS with AVR and pure sine wave output, which might be related to their longevity.

FWIW personal experience is biased towards that person, i.e. it's only anecdotal evidence and not necessarily indicative of a larger trend. I've had two smaller units running for awhile now, on systems that only need to be rebooted for kernel updates.

(Wow ... i last posted on SPCR in 2010. Those were the days.)

That is a good point too. Mine will also be behind a true sine wave UPS. (Either APC SmartUPS. SUA750 or SUA1500 depending on which one I plug it into)

Yeah, my last post here was quite a few years back as well!

mattlach wrote:How do you guys feel now a few years later?

All of my picoPSUs (main system, htpc and server) are still going strong. No UPS. I have good bricks though, and the Swedish power grid is solid.

Vicotnik wrote:

mattlach wrote:How do you guys feel now a few years later?

All of my picoPSUs (main system, htpc and server) are still going strong. No UPS. I have good bricks though, and the Swedish power grid is solid.

+1 for this. I have a 120WI with a 60W brich from a high-end 2004 laptop, no UPS and a solid grid. And it still works like a charm. Granted, it runs a low-powered home server only (see sig), but it still runs 24/7. I can't remember when I bought the PicoPSU, but should be about 5-10 years ago. Before the current server, it ran a Athlon X3 system that also ran silent, and for years.

Thanks guys. This is great info!

Anyone know what kind of power use I'll see at the Wal with a 35W tdp dual core Haswell chip with one of these Pico PSU's?

I've got an i5-4570T coming next week, probably going to disable HT to save power, as I don't need it. Going barebones with 2x1GB 1.35v SODIMM's a small CPU fan controlled via PWM, and SATA disabled, booting from a USB stick.

Hoping to get sub 15w idle. Might lower voltage and underclock if I don't get the desired efficiency.

mattlach wrote:Hoping to get sub 15w idle. Might lower voltage and underclock if I don't get the desired efficiency.

Based on the little research I've done, you won't gain anything by underclocking, as it doesn't affect the idle states. It only limits the max.

rruff wrote:Based on the little research I've done, you won't gain anything by underclocking, as it doesn't affect the idle states. It only limits the max.

+1

Also, since the CPU idle at less than 5W it's mostly the choice of motherboard that determine how low the idle power consumption can be.

Side question, slightly OT: which are the merits of such a T-SKU, if run with HT disabled, over a G4500?

quest_for_silence wrote:Side question, slightly OT: which are the merits of such a T-SKU, if run with HT disabled, over a G4500?

Well, that depends really.


Turbocore is a good feature which the g4500 lacks, but then again the g4500 clocks higher in the first place so it may not make much of a difference.

The 35W TDP of the 4570T was a huge draw to me. The g4500 is higher. (Over 50?)

What it really came down to for me was the fact that I got a really good open box deal on a Supermicro LGA1150 mini server motherboard. Since skylake is socket 1151 it wouldn't have worked for me. I was originally looking for a 35W Haswell Pentium or Celeron, but they are difficult to find. Eventually I found the 4570T cheap on eBay and decided to go with it.

I haven't shopped for skylake chips like the G4500 yet. I recall that when they launched they were very difficult to find. Is that still the case?

Also, have companies like Supermicro and Tyan released any good socket 1151 server boards yet?

mattlach wrote:The 35W TDP of the 4570T was a huge draw to me. The g4500 is higher. (Over 50?)

I'd say rather unlikely.

mattlach wrote:What it really came down to for me was the fact that I got a really good open box deal on a Supermicro LGA1150 mini server motherboard.

That's right.

mattlach wrote:I recall that when they launched they were very difficult to find. Is that still the case?

Also, have companies like Supermicro and Tyan released any good socket 1151 server boards yet?

No & yes.

quest_for_silence wrote:

mattlach wrote:The 35W TDP of the 4570T was a huge draw to me. The g4500 is higher. (Over 50?)

I'd say rather unlikely.

mattlach wrote:What it really came down to for me was the fact that I got a really good open box deal on a Supermicro LGA1150 mini server motherboard.

That's right.

mattlach wrote:I recall that when they launched they were very difficult to find. Is that still the case?

Also, have companies like Supermicro and Tyan released any good socket 1151 server boards yet?

No & yes.

Interesting. Intel's spec for the G4500 is a TDP of 51w.

Intel has clever marketing, and it works. We fight it, bitch about TDP in the forums and how it's intended to be read. Sometimes we win, sometimes we don't.

mattlach wrote:Interesting. Intel's spec for the G4500 is a TDP of 51w.

And the slower, less capable G4400 is specced as 54W... don't trust Intel, ever (or NVidia, Microsoft, AMD, IBM, Apple, Google, Amazon... ).

quest_for_silence wrote:

mattlach wrote:Interesting. Intel's spec for the G4500 is a TDP of 51w.

And the slower, less capable G4400 is specced as 54W... don't trust Intel, ever (or NVidia, Microsoft, AMD, IBM, Apple, Google, Amazon... ).

Well, as any good overclocker knows, performance/clock speed and heat output/power usage are not directly linked, even for the same design (in this case skylake) made on the same production line. There is chip to chip variability that is significant.

You could buy two identically labeled CPU's (Let's say i7-6700K). Both might overclock to 4.5Ghz, but one may require significantly more voltage in order to do so and be stable than the other, and the one that requires the higher voltage, is going to use more power.

It goes back to how chips are made. They are made at the border of our ability to make them, which means there is a huge amount of process variability, especially at the very beginning of manufacture of a new chip design.

There is no i7 line, or i5 line or Celeron line or anything like that. All current Skylake chips are made on the same 4 core production line. Before they are completed there is no difference between them. At the end of the line they are tested. Some may have a bad core or two. They likely become 2 core chips. Some may have a bad hyperthreading decoder, they become a product code that doesn't have hyperthreading. Some clock higher, they become high end models, some clock lower they become lower end models. At the very end after this testing process (which is usually referred to as binning) the lid of the chips is laser etched with a product code, and only then does it become an i7-6700k or a Pentium G4500 etc. etc.

Generally, chips towards the center of the wafer wind up being higher quality, and ones toward the edges wind up being lower, but it varies. Also as the product line matures, the process/manufacturing engineers usually tweak it and improve it, so yields improve over time. (this is why later chips of the same model tend to overclock better than earlier ones)

The binning process also sorts based on the voltage required to hit a certain clock speed. There is high demand for low power chips due to their thermal and power use characteristics, so as part of the binning process, if a chip can be stable at the required clock speeds at a much lower voltage than other chips, it becomes a low voltage model, gets marked with a T suffix and is sold as a 35W part. A chip that takes more voltage to reach the same clock, may be sold with the same clock speed,core count and everything, as a non T part and specified at 50W TDP.

Even within the same bin, there is going to be some significant variability, as they are sorted based on an "is at least a 35W part if not better" type criteria. This is why some people have success in lowering power use by undervolting their chips, and others don't.

My guess would be that the lower end G4400 chip is made from lower quality chips off the same line, and thus requires more voltage to be stable, and thus the specification of 3.3Ghz @ 54W applies, instead of the specification of 3.6Ghz @ 51W.


Don't get me wrong, I don't trust Intel in the slightest. They are a quite deceptive company, but in this particular regard, I understand how they operate.

You can buy a lower clocked part and maybe overclock it to the level of a higher clocked part. But for the underclocked parts there's no maybe. I have not seen a non-T part that would not be stable at T-settings. As far as I know, the T-parts don't go into the higher P-states and that's it. That and a nice turbo frequency.
The point of the T-part is mostly for special cases, like limited cooling (space) or power. At idle there's little difference between an i3 and an i7, let alone between T and non-T.