Thirty Four PSUs Tested - Is Biggest Best?

SteveK writes "Hexus has been testing some 34 PC power supplies to see which is best. There are some interesting results. An Enermax 535 Watt PSU couldn't deliver much over 450W, while a cheap 250W PSU did exactly what it said on the box. There's also a video of a (very cheap) 650W PSU under 400W of load, requiring over 1kW of input power to sustain the load, before blowing up."

Hexus = good reviews, shitty servers.

[jgaynor]

Mod me down for slander, but I don't understand why we keep linking to Hexus reviews. Their content quality is high but their servers can't take a slashdotting for more than 3-4 minutes. 0 comments and it's taken over a minute to load as it is :(.

Where's Antec?

[tgbrittai]

It's a fairly popular high-end PSU brand. Seems like it should have been included in the review. Hmmm...

Re:Hexus = good reviews, shitty servers.

[unts]

Steve from HEXUS here. :)

We've got new kit going into place soon, but that's not my department. We've taken measures in the meantime to cope with any traffic surges, like Slashdottings, but with a massive article like this one, it's tricky.

Thanks for your patience, guys.

For PSUs, these days...

[Silverlancer]

Quality usually goes hand in hand with price. The best ones are usually the most expensive (PC Power and Cooling). The cheap ones do stupid crap like toss 400 watts onto the 5 volt rail and then call it a 650 watt power supply, when it might crash when you put in that 7800 GTX. Cheap supplies also often are very inefficient, dissipating huge amounts of perfectly good elecricity as heat. There are some exceptions to the rule, but in general I've found that the better ones tend to cost more.

Coral Cache link

[grimwell]

Maybe try a Coral Cache url instead of linking directly to Hexis http://www.hexus.net.nyud.net:8090/content/static/ psu_roundup.html [nyud.net]

Re:Clean input

[GeekDork]

Well, I can't RTFA due to some other people trying to do so, but a good test setup usually includes a "clean" primary power supply for fairness as was already suggested and then some fun add-ons to simulate controlled SNAFUs like bursts, surges and very short interruptions of up to, say, 100ms.

Question on slashdottings.

[Grendel Drago]

I assume you guys are server-limited, not bandwidth-limited, when the Slashdot beast comes around. So why don't you put up a static version of the page when the Slashdotting hits? Why don't you at least use some sort of caching to reduce the load on your servers? Is there something obvious that I, not being a mighty server admin, am missing?

Re:External Power Supply Macho

[Anonymous Coward]

Easy enough, look to the telcom world.

Get some nice -48V atx dcdc converters for your boxes, and a couple of really nice rectifiers.

Makes battery (and generator) backup so much easier, and it's much more efficient.

Re:For PSUs, these days...

[Eil]

Where I work, our rule of thumb is that heavier power supplies are higher quality than lighter ones. While I'm sure this isn't going to be true in every single case, it makes a certain amount of sense. A manufacturer of cheap power supplies is going to try to put the least amount of material and labor into their units as possible. Quality PSU manufacturers tend to put in better components and beefier heatsinks. (Hence the fan(s) can spin slower, resulting in a quieter PSU as well.)

Coral Cache Link!!! Use this!

[cbiltcliffe]

Coral Cache link, and at least page one is there.

Use this one instead of the submitter's link!

http://www.hexus.net.nyud.net:8090/content/static/ psu_roundup.html [nyud.net]

Re:Hexus = good reviews, shitty servers.

[tolan-b]

here's the mirrordot link

http://www.mirrordot.org/stories/2400897cd3c6f4ab7 791acee73a8c04f/index.html [mirrordot.org]

Re:For PSUs, these days...HEC/Sparkle

[slaker]

I'd just like to add that HEC power supplies are also surprisingly quiet and generally very reasonably priced. Sparkle PSUs are loud SOBs, but the parent here is absolutely right: Sparkle and HEC units are generally so reliable that they verge on boring. Which is very good thing to say about power supplies.

Tom's Hardware's PSU Strees Test, Aug. '05

[PoisonousPhat]

Only 19 PSUs tested, but you can still get to the site (for now).

http://www.tomshardware.com/howto/200507111/index. html [tomshardware.com]

To cut to the chase, TH recommended the Fortron FSP300-60GNF and the Seasonic S12 600.

Re: Antec

[LordKronos]

no Antec must mean they didn't get a "free" one

From TFA:
We were very careful to use retail power supplies for our testing, mindful of not falling into the trap of asking manufacturers for supplies only to have special units sent which stand up more than a retail unit would.

Re:Where's Antec?

[pmc]

Nope - don't go for Antec if you live in the UK. I bought one of their 430W power supplies, and after eight months it died. Fine, I thought, it is under warranty. So went to the web-site and after a bit of mucking about I managed to get an RMA. Or thought I did. I actually had filled in a form to request a form to request an RMA. Pointless bureaucracy gone mad. Still I got the form. Or rather excel spreadsheet. So now I need a) a working computer (erm, guys, the power supply's gone) and a copy of excel (probably an other speadsheet would have done) to tell them who I was, what I'd bought and when I'd bought it. Oh - they also wanted a scanned copy of the receipt sent back to them too. I did have an electronic copy that could have sent them, but it was on the computer that was dead. (They did suggest I could take a digital photo of the invoice and send that instead, but this was getting too Alice-in-Wonderlandish for me.)

But all this was just slightly stupid and annoying. What was very stupid and immensely annoying was that I had to send the power supply to them at my own expense to a different country. The power supply originally cost about 50GBP - to post it to the Netherlands (for that is where their warehouse is) from the UK cheaply (but insured) would cost about 25GBP. And they would not send me a new one until they had the old one back and checked out. I would end up out about half the cost of the power supply, and be without one for possible a couple of weeks. Suddenly, paying a premium price for a quality product did not seem to be such a good idea when faced with a avaricious and slow customer service department based in an entirely different country.

So my advice is avoid Antec if you live in the UK - you effectively pay about half the cost of the power supply if you need warranty repairs/replacement.

The story does have a happy ending - I bought the supply thought Amazon originally, and so phoned them up. After a bit of reminding them of their duty under Sale of Goods act (basically a quality brand should last longer than eight months) they agreed to replace it. They dropped the ball on the first attempt, so I actually ended up with a better spec'ed supply. Still an Antec, so if it dies it hits the bin rather than muck about with any ludicrious postal demands.

Re:Hexus = good reviews, shitty servers.

[grazzy]

Two reasons;

1) Templates. A template for a large article wouldn't be usable for shorter (1-page) articles.
2) Pageviews. Equals money in pocket.

No Seasonic, Antec or PC Power & Cooling?

[Anonymous Coward]

That was a bad on their part, but they do have alot of the other "major" Power Supply venders. Q-Technology, Hiper and Fortron are respected brands. Antec is our company's main PSU and we have never had a problem with them. Too bad they weren't there.

Let me get this straight

[Psionicist]

The review tested all these products in FSG Groups facility, an employee of FSG Group is said to be "sexy" in the review, and a product from FSG Group won? Yeah, right...

Finally

[HunterZ]

I'd like to express my appreciation to these guys for performing a much-needed analysis and publishing the results for all to see. It's about time someone called PSU manufacturers' bluffs and published testing results for multiple brands and models. They even made sure to test mostly retail models to prevent the possibility of manufacturers supplying souped-up units.

Looks like the moral of the story is to look carefully to see whether the rating on the box is for peak or sustainable power output. I just had a cheap "550 Watt" PSU blow out on me a few weeks ago after about 8 months of moderate to heavy use (lots of hard drives, ATI Radeon 9800Pro, Athlon XP 3200+, but no neons or anything fancy like that), and bought a more expensive replacement in hopes that it will be of a higher quality. Several friends and family members have also had PSU troubles in the past few months. This article would have been nice to refer to when shopping for replacements.

Like many of the other posters here and in the HEXUS forum, I'm looking forward to future roundups that cover other brands. It was mentioned on the HEXUS forum that Antec and other brands will probably make the next round: http://forums.hexus.net/showpost.php?p=584160&post count=15 [hexus.net]

Re:Coral Cache link

[gid]

Use the print article [nyud.net] link instead. That just loaded up for me after several attempts. The images are still linked to the main server so it'll take awhile.

Re:External Power Supply Macho

[InvalidError]

In some installations, a 400V DC bus is used for simpler, more efficient and more reliable power distribution. By centralizing line inputs, it is easier to do power factor corrections, power conditioning and since it is all DC, a battery bank directly across the DC bus can replace the UPS. Each (sub-)system afterwards has it own DC-DC converter.

Why is power distribution done at high voltages? Simple: to reduce conduction losses in wires, semiconductors and other devices. Old systems were based on 5V until ATX came along, at which point 3.3V systems became the norm for a short while - stepping down from 3.3V involved too high currents to be practical, this is how we ended up with ATX-12V and today's 12V-based systems where most/all MoBo regulators are fed off the 12V rail.

In the near future, every device will have some form of on-board power regulation - many chips already do have built-in linear regulators for their more sensitive circuits. The trend to ever more localized power regulation is required by high-performance/high-speed circuitry which needs very fast transient response. Look at Intel's recent mobile CPU announcement: the next Pentium-M generation will be multi-chip modules containing, in addition to the CPU, both the north bridge and Vcore regulator.

If you go from line to load, power goes through 3-6 voltage regulators before reaching the target:
1- active PFC (relatively few PSUs where not required by law)
2- bulk transformer
3- magnetic regulator (for most PSUs that actually regulate auxiliary rails)
4- MoBo regulators
5- on-card/device regulators
6- on-chip regulators

Modern CPUs and GPUs require microsecond-scale response time from the bulk PSU when going from idle to 100% load. Failure to deliver will cause crashes. Integrated regulator can have sub-microsecond response times, local board-level regulators are at the microsecond scale and response times quickly climb in the tens of microseconds afterwards due to wiring impedance.

So, the DC-bus is pretty much as close to centralized power regulation that would be practical with high-speed loads.

Re:More PSU heat on 220/240 vs. 110 power?

[Anonymous Coward]

Actually, no... having looked inside a PSU, it is odd what they do. If you are running off 110, they voltage-double to about 300VDC and chop that; if you are running off 230, they rectify to about 300VDC and chop that. There is actually slightly more heat loss off the voltage doubler, so a PSU is slightly less efficient running on 110V. Why do they do this? Well, I can't be certain, but I think it's because that way they can switch between 110 and 230VAC with a single-pole, single-throw switch, which is cheap; and the circuitry that either doubles or rectifies is equally cheap (two diodes, two capacitors).

Comment removed

[account_deleted]

Comment removed based on user account deletion

Re:Hexus = good reviews, shitty servers.

[dajak]

Funnily enough, this actually encourages the Average User (a mythical beast, only whose footprints have ever been found) to read the whole article. Usability reports I remember reading a few months ago indicated that on an interactive medium like the web, users get "bored" if they don't have to interact with a page for too long. If you don't provide regular user-interaction (eg, by making them click for the next page) they get fractious and are more likely to drop out of reading the article.

Because they can scan through the article faster, and use ctrl-f on the whole article, they will find out quicker that the article was a false positive on their search. It has nothing to do with preventing boredom: Most pageviews on the Internet are more or less accidental, and bad design is rewarded with more clicks. The reader is still bored.