Re: Building computer - power supplies
A point I forgot to make. This is something everyone should know.
Subject: The marketing myth of multiple +12V rails
Switching regulator MOSFETs provide the 3.3, 5/5VSB, and 12VDC output
current of a PC PSU. Before the days of waffle iron CPUs and GPUs, PSUs
had a single 12V regulator providing a single 12V circuit, or "rail", at
about 16 amps, good for 16*12=192 watts. That was sufficient for some
drives and adding a little extra power to the CPU socket via that 4 pin
plug, and a little extra to a hot graphics card in an AGP slot.
Then CPUs came along eating 130W and PCIe GPUs arrived eating 100W.
230/12=19A. Oops, more than 16A, not quite enough amps for the CPU/GPU,
let alone my drives. The initial response to this by the PSU industry
was simply to use larger switching MOSFETs and support devices in the
output circuit to give us ~25A on the 12V rail. This worked well enough
for a while, even though the larger MOSFETs were a bit more expensive.
They simply increased the price a bit and passed it on to the
wholesaler, retailer, and the consumer. No big deal. Yet.
Then arrived the massive power hog GPUs and SLI/Xfire adding insult to
injury. Some of these GPU combos require up to 300 watts (or much more
for quad SLI). Add in an 8 core CPU pulling 75W from the 12V rail, and
now you need 31A, plus another 8A for spinning HDs, burners, etc. So we
now need at least 40A of 12V power. However, if you pull more than
85-90% of the maximum rail current, voltage will drop, and the system
will become unstable. We need a safety margin of ~20%, so we now need
at least 48 amps of 12V power. Let's just round that to 50, a nice even
number. And, not coincidentally, we have a plethora of PSUs on the
market with dual 25A +12V rails. Fancy that.
But why two at 25A instead of one at 50A? Simple: component cost. A
50A switching regulator and support devices costs significantly more
than using two sets of 25A devices. In real terms it's not a huge
difference, let's say $8 vs $3. So the price of your dual rail PSU
should have gone down by $5, right? Wrong. Here's why.
Some marketing genius at one of the PSU companies, years ago, figured
out that telling consumers "2 is better than 1" would work. And he was
right. People without the technical knowledge have been indoctrinated.
The rest of the industry followed. So not only are they saving $5 per
unit, they're charging the consumer $5 more per unit because "2 is
better than 1". That's an extra $10 per unit, in essence, in the
manufacturers' pockets. Now, if just one company produces and sells
100,000 units, that's an extra half $million in profit. Not chump
change. If they sell 6 models with dual rails, all at 100K units,
that's $3 million extra profit.
And what has the consumer gained from this? Nothing. It simply makes
wiring your PC more difficult, as you have to balance your 12V devices
across an arbitrary boundary placed across the 12V output current of
your power supply.
Look at any *real* server PSU. Only one 12V rail, always has been,
always will be. Some models go well over 100 amps on a single rail.
Here are two good SuperMicro examples.
Standard 865W ATX form factor PSU, single 70A 12V rail:
http://www.newegg.com/Product/Product.aspx?Item=N82E16817377004
Proprietary form factor 1620W PSU, single 135A 12V rail:
http://www.newegg.com/Product/Product.aspx?Item=N82E16817377052
And of course, now you should be asking yourself, given what I've
stated, and if 1 rail is better for servers, why are 2/3/4 rails
supposedly better for a consumer PC? And of course the answer is, more
rails is not better. It's worse.
It's exactly the same situation as digital camera marketeers convincing
consumers that a 12MP camera is intrinsically better than an 8MP camera.
The simple fact is that even 2MP is sufficient for 99.9999% of all
users, because they share their photos online, not in print. 2MP camera
format is typically 1600x1200. The average screen res, across desktops
and laptops, is far less. Higher numbers sell, in America anyway. The
big upshot for the industry? 8/12/16MP cameras need bigger flash cards
for a given number of snaps. That equals more flash sales, etc, etc.
What is lost on most people is that *quality* 2/3MP cameras take better
quality pictures than cheap 16MP cameras. Just look at the current
smartphones. They ship with 8MP yes? Seen the photos from one? Yes,
they suck, horribly.
--
Stan
On 9/27/2013 5:00 AM, Stan Hoeppner wrote:
> On 9/26/2013 12:34 PM, Doug wrote:
>> On 09/26/2013 01:05 AM, Stan Hoeppner wrote:
>>> On 9/25/2013 12:52 PM, Catherine Gramze wrote:
>>>>
>>>> On Sep 25, 2013, at 12:52 AM, Stan Hoeppner <stan@hardwarefreak.com> wrote:
>>>>
>>
>> /very large snip/
>>
>> I just left this final section to illustrate the tome of the whole thing:
>>>
>>> The socket vs upgrade concern is not valid. It will be at least 10
>>> years before desktop applications, especially Linux apps, are
>>> sufficiently threaded to take advantage of today's 4 core CPUs, let
>>> alone 6/8 cores. New hardware doesn't make old hardware obsolete. New
>>> software does. There are many people, including myself, who would say
>>> this applies even to dual core CPUs. And in fact, at the rate of
>>> desktop software development today, WRT threads, one is better off
>>> purchasing a higher clocked dual core CPU with lots of cache than a quad
>>> core model at a lower clock, especially given power consumption.
>>>
>>> Worrying about which future CPU will fit in your mobo socket is a waste
>>> of brain cycles. It's irrelevant. By the time you *need* to upgrade,
>>> you'll want newer faster DRAM, etc. So you'll end up replacing the
>>> board, CPU, and memory, as people have been doing for some years
>>> already. Oh, add to that the fact that AMD and Intel keep changing
>>> sockets to accelerate this trend. The entire industry is better off
>>> when people keep buying new hardware, obviously, even if the consumers
>>> aren't.
>>>
>>
>> Stan, I think the advice you've given Catherine will be useful to a lot
>> of folks who read this list. Particularly in regard to ram usage and
>> power supply size.
>
> DRAM has trended very cheap for a while, though it's starting to go back
> up a little. Due to price many people have been buying more than they
> need, simply because they can, at least here in The States. And because
> it's been so cheap big box vendors have been using DRAM capacity as a
> selling point. So when people see most if not all retails desktop PCs
> shipping with 6GB or more RAM, they are conditioned to believe that's
> the "new minimum" that everyone should have. This is of course folly.
>
> The multi-core issue is of greater importance as CPUs typically cost
> much more than a couple of DIMMs. The overwhelming majority of people
> will never make use of 4 cores, let alone all the cycles of two cores.
> Ever time I see someone say they're going to buy a quad core CPU I
> cringe, and usually just bite my tongue. Many people simply don't want
> to know the truth. Buying 4 instead of 2 makes them "feel" better about
> the purchase, and that's what counts, I guess...
>
>> I have heard, however, that the _make_ of power supply may be important,
>> as some cheapies may not hold up. You might comment on that, and if
>> that's correct, possibly recommend a name or two.
>
> So much I can say WRT PSU purchasing. I'll try to be brief (and fail),
> but complete.
>
> When it comes to desktop PC power supplies brand has little to do with
> reliability/longevity, but it definitely has influence on price, and
> sometimes, but not always, a better warranty experience if you have to
> return one. The design of DC-DC switching power supplies hasn't changed
> much in 30+ years. It's a known science. The only thing that changes
> regularly is the cost, and sometimes quality, of components, due to the
> drive to reduce production cost and increase margins. And this is where
> you run into problems. These may include using low quality ICs, caps,
> assembly processes, etc, and/or may include skimping on testing during
> Q.C. A bad batch of components can bite any vendor in the butt.
> Sometimes good QC can catch this, sometimes it can't.
>
> My advice? Ignore brand. Go to your favorite retail/wholesale website.
> Plug in the power rating and price range you're interested in. Read
> the reviews of the units. Yes, read them. Don't rely strictly on the
> number of stars, eggs, etc. One bad review by a novice builder who
> burns up his system, or one who gives 1/5 because the cables are too
> short for his case, or just doesn't know what he's doing, can ruin a
> rating, and cause you to skip over a perfectly good supply. Here's an
> example:
>
> http://www.newegg.com/Product/Product.aspx?Item=N82E16817103514
>
> The guy who gave 1 egg didn't buy the unit from Newegg but from an Expo.
> His PSU may not even be the SPI unit he's writing the review for. He
> can't figure out how to plug in the Molex connector, which is a standard
> all mobos and PSUs adhere to. He's an idiot. So you discard this
> review and his 1/5 rating. That makes this unit a 5/5 instead of a 3/5.
> Not the best example as there are only 2 reviews, and you should never
> rely on a single positive review. But you get the idea what to look out
> for in this regard.
>
> Following, however, is the review that makes you walk away from a
> particular unit:
>
> http://www.newegg.com/Product/Product.aspx?Item=N82E16817103089
>
> SPI is one of the top 5 quality PSU manufacturers and has been for over
> 15 years. They contract manufacture for some of the big box PC names as
> well as sell under their own. Their current top model of the latter
> category is available at Newegg for $210 (yikes, for a desktop PSU?!).
> It failed on two experienced builders, both verified purchasers, who
> posted the only two reviews, giving it a true 1/5 rating. Both failed
> at 3 months during normal operation. This tells us a common component
> failed in both units. SPI obviously received a batch of marginal
> components from a supplier and the deficiency wasn't detected during QC.
> The design is not faulty, nor the QC.
>
> Their other units all rate a true 4-5/5. They likely simply received a
> bad batch of components used in this model. Note what I said up above
> about even top manufacturers getting burned by faulty components.
>
> This is exactly why you should not rely on a brand name alone when
> making a PSU purchase, or any purchase really. No one "in the know"
> expects a $200 SPI unit to fail, ever, or any $200 PSU for that matter.
> In this price range quality/longevity *should* be an absolute.
>
> Read the reviews. Ignore brand.
>
>
> The last standalone PSU I purchased was this $20 480W (peak) generic
> unit branded as "Sunbeam":
>
> http://www.newegg.com/Product/Product.aspx?Item=N82E16817709010
>
> I rebuilt my workstation after a mobo failure and replaced the PSU for
> good measure. This Sunbeam was cheap. The ~100 reviews I read of the
> 200+ were mostly positive, and the failures reported coincided with
> standard failure rates.
>
> It has been running flawlessly for 2 years 1 month. The old PSU had
> about 8 years on it. It was a 350W generic unit branded "Qmax", made in
> China. Still works. It came in the mid tower case I purchased in 2003.
> $35 for case with PSU, so it was a very cheap PSU. However, so people
> don't get the idea that any arbitrary cheap PSU will run for 8+ years,
> the following needs to be noted.
>
> I buy the cheapest desktop PSUs I can find that have decent reviews.
> However, before installation I open 'em up and remove the junk stock
> 80mm low RPM "quiet" sleeve bearing fan and install a faster spinning
> yet still pretty quiet NMB ball bearing unit. I bought some in bulk on
> a closeout sale a decade ago. NMB is one of the top 4 quality DC axial
> fan manufacturers, along with Sanyo Denki, Nidec, and Delta. I've
> probably mentioned this on list before.
>
> Replacing the crappy stock fan with a quality long life model
> contributed the the 8+ year life of the last cheap PSU. These NMBs last
> forever. Instead of grabbing a new one from my stock I removed the unit
> from the old PSU, cleaned it with a toothbrush and compressed air (from
> my shop compressor, not the canned junk), and dropped it in. Before
> it's done this NMB may end up with 15 or more years on it. It already
> has 10, only 5 to go. I have other NMBs and Nidecs with over 10 years
> on them. Keep em clean and they're like the Energizer bunny.
>
>> An excellent presentation. I have put it in my "saved" file.
>
> Thanks. Maybe this one will be as informative.
>
Reply to: