2003 Anandtech Power Supply Roundup Part II

PSUs: The source of DC power for all components in the PC & often a big noise source.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee, Devonavar

Post Reply
MGP
Posts: 519
Joined: Thu Dec 05, 2002 4:32 pm

2003 Anandtech Power Supply Roundup Part II

Post by MGP » Thu Jul 31, 2003 11:01 am


POLIST8
*Lifetime Patron*
Posts: 703
Joined: Mon Jul 14, 2003 12:59 pm
Location: Madison, WI USA
Contact:

Post by POLIST8 » Thu Jul 31, 2003 11:18 am

Has anyone really defined why PSUs are rated so high but yet always deliver less than their specifications?

Is it just a marketing scheme that they all got caught up in?

MikeC
Site Admin
Posts: 12283
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Thu Jul 31, 2003 1:07 pm

It's a pretty thorough roundup, at first look. Some good technically relevant details. Efficiency was not even mentioned, however.

Interesting about ThermalTake PurePower 480W being tops for noise -- mind you they did not have some of our fave quiet PSUs in the mix. Sparkle FSP350-60PN does come out as being louder than ThermalTake. Have to get a hold of one I guess...

POLIST8
*Lifetime Patron*
Posts: 703
Joined: Mon Jul 14, 2003 12:59 pm
Location: Madison, WI USA
Contact:

Post by POLIST8 » Thu Jul 31, 2003 1:22 pm

Mike, I don't trust any reviews (aside from Tom's) except the ones that come from here.

It's also been mentioned about Tom's being somewhat biased...

I guess I just go there to get the goods on new tech and do further research elsewhere...

jaeger66
Posts: 4
Joined: Wed Jul 30, 2003 5:38 pm
Location: Phoenix, AZ

Post by jaeger66 » Thu Jul 31, 2003 2:29 pm

POLIST8 wrote:Mike, I don't trust any reviews (aside from Tom's) except the ones that come from here.

It's also been mentioned about Tom's being somewhat biased...
Not so much biased as completely incompetent.

Necroz
Posts: 10
Joined: Wed Jul 23, 2003 3:50 am

Post by Necroz » Fri Aug 01, 2003 5:10 am

MikeC wrote:Interesting about ThermalTake PurePower 480W being tops for noise -- ..
I have a hard time believing it also. It would have been nice to see how the Seasonic matches up.

Riffer
Posts: 517
Joined: Mon Jan 06, 2003 4:14 pm
Location: Toronto, Canada

Post by Riffer » Fri Aug 01, 2003 5:52 am

No surprises in that roundup, at least to anyone from these forums:

Best PSU - PC Power & Cooling, but noisy.
Best "Bang for the Buck" - FSP/Sparkle/Aopen/Zalman etc. etc. etc.

They also allude to the quiet fan mod to PC Power & Cooling PSU's made by yours truly and others.

Still think that MikeC's PSU reviews are the best there is, and one of the real good things about this website. Unfortunately, it appears MikeC is a quality versus quantity guy, so I doubt we will ever see any big roundups on SilentPCReview.
Last edited by Riffer on Fri Aug 01, 2003 8:43 am, edited 1 time in total.

dukla2000
*Lifetime Patron*
Posts: 1465
Joined: Sun Mar 09, 2003 12:27 pm
Location: Reading.England.EU

Post by dukla2000 » Fri Aug 01, 2003 6:13 am

Riffer wrote:Still think that MikeC's PSU's are the best there is, ...
Hear Hear! (or at least his reviews!)

Intrigued by the test using MemTest. I thought DRAM refreshed after any read or write, as well as automatically periodically (like every few msec)! Their results indicate your choice of psu can inject memory errors which is hard for me to believe, at least in general. I can believe a very poor psu could do this, but to see that ALL 18 psus tested can inject memory errors (results here) makes me wonder what that test was really measuring? I mean, does this imply our crashes and instabilities should be blamed on our psus and not Micro$oft?

soundOFsilence
Posts: 40
Joined: Sat Jul 05, 2003 1:03 pm
Location: San Francisco, CA

Post by soundOFsilence » Fri Aug 01, 2003 10:01 am

Necroz wrote:
MikeC wrote:Interesting about ThermalTake PurePower 480W being tops for noise -- ..
I have a hard time believing it also. It would have been nice to see how the Seasonic matches up.
Look at the thermal performance roundup as well.

http://www.anandtech.com/showdoc.html?i=1841&p=22

Do you really need another 42 Celcius component in your machine? With its low 30C temparature and its 12cm fan, the Fortron FSP350-60PN may be the winner. The Aurora version also has a manually adjustable fan speed knob (Anyone have/used/seen/heard one of these?). See newegg

Also, there is no qualitative sound testing like in Mike's reviews (other than the pitch of the Sparkle being mentioned). I'd like to hear about as much about the type of noise, high/low frequency content, buzzing, clicking etc. rather than just a dBA measurement.

nbac
Patron of SPCR
Posts: 142
Joined: Sat Jul 05, 2003 11:27 am
Location: Sweden

Post by nbac » Fri Aug 01, 2003 10:02 am

Did it really take months to complete this one? Time to consider another profession!

"The ambient temperature was 21.0C.".....(my guess: ambient=case
since the test method wasn't specified)

Hardware sites aren't what they used to be: this one is much better :)

marc999
*Lifetime Patron*
Posts: 230
Joined: Sat Oct 26, 2002 4:05 pm
Location: Calgary, Alberta, Canada

Post by marc999 » Fri Aug 01, 2003 12:41 pm

Do you really need another 42 Celcius component in your machine? With its low 30C temparature and its 12cm fan, the Fortron FSP350-60PN may be the winner.
Yeah, but it blows it's exhaust back into the computer case !! To me this totally negates the lower temp of the unit itself. Who cares how hot it gets as long as it exhausts the heat outside of your case ? It's for this reason I would never buy one of those style of PSUs even if they are quiet.

Harry Azol
Posts: 228
Joined: Fri May 02, 2003 9:21 am

Post by Harry Azol » Fri Aug 01, 2003 12:43 pm

marc999 wrote:
Do you really need another 42 Celcius component in your machine? With its low 30C temparature and its 12cm fan, the Fortron FSP350-60PN may be the winner.
Yeah, but it blows it's exhaust back into the computer case !!
no, it does not

marc999
*Lifetime Patron*
Posts: 230
Joined: Sat Oct 26, 2002 4:05 pm
Location: Calgary, Alberta, Canada

Post by marc999 » Fri Aug 01, 2003 12:49 pm

Guess I have to research these more. Sorry.

MikeC
Site Admin
Posts: 12283
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Fri Aug 01, 2003 12:52 pm

nbac wrote:"The ambient temperature was 21.0C.".....(my guess: ambient=case since the test method wasn't specified)
Umm, I doubt it. 21C is way too low for in-case temp. Much more likely to be room temp.

BaconTastesGood
Posts: 137
Joined: Mon Nov 04, 2002 3:55 pm

Post by BaconTastesGood » Fri Aug 01, 2003 5:00 pm

Overall I liked the article. The key is that it wasn't quite as rigorous as SPCR's stuff, but it did address things that aren't as focused on here such as stability and interference. I found it interesting that they were asserting a 330W Antec wasn't capable of reliably supplying a P4 of any type -- is this hyperbole, misinformation, or the scary truth?

Sledge
Posts: 156
Joined: Sat Nov 09, 2002 4:40 am

Post by Sledge » Fri Aug 01, 2003 5:19 pm

On Anand's temp page
Tests were conducted inside an industry standard Chenming case with no fans running
I'm guessing there was no processor, mobo, or even a incandescent light bulb in the case which is why the case ambient is so low.

lenny
Patron of SPCR
Posts: 1642
Joined: Wed May 28, 2003 10:50 am
Location: Somewhere out there

Post by lenny » Fri Aug 01, 2003 9:35 pm

Harry Azol wrote:
marc999 wrote:
Do you really need another 42 Celcius component in your machine? With its low 30C temparature and its 12cm fan, the Fortron FSP350-60PN may be the winner.
Yeah, but it blows it's exhaust back into the computer case !!
no, it does not
I think Harry may have been mistaken in its design.

I have one sitting right in front of me now. There is one 120mm fan, where the second fan on a 2 fan PSU usually is, and it blows air from inside the case into the PSU (it doesn't exhaust into the case - that would not have been very smart). The back where the fan usually is for regular power supplies is a grill (stamped into the PSU case) of hexagonal holes. On the opposite end of that are four small slots. Conceivably, some air might escape back into the case from those slots. But I'd think those two slots are less than 5% of the surface area of the holes in the stamped grill.

I wonder what those four slots are for, and if it makes any difference if they are covered up.

Ralf Hutter
SPCR Reviewer
Posts: 8636
Joined: Sat Nov 23, 2002 6:33 am
Location: Sunny SoCal

Post by Ralf Hutter » Sat Aug 02, 2003 3:58 am

BaconTastesGood wrote:I found it interesting that they were asserting a 330W Antec wasn't capable of reliably supplying a P4 of any type -- is this hyperbole, misinformation, or the scary truth?
Can anyone comment on this?

Ralf Hutter
SPCR Reviewer
Posts: 8636
Joined: Sat Nov 23, 2002 6:33 am
Location: Sunny SoCal

Post by Ralf Hutter » Sat Aug 02, 2003 4:04 am

nbac wrote:Did it really take months to complete this one?
According to this:

(quote)
Unfortunately, it took 6 hours to run this test and we ran it three times for each supply to assure accuracy.
(enquote)

it would have taken them 13.5 days alone just to run that one Memory Interference Test.

nbac
Patron of SPCR
Posts: 142
Joined: Sat Jul 05, 2003 11:27 am
Location: Sweden

Post by nbac » Sat Aug 02, 2003 10:48 am

So, basically what they are saying is that current
EMC regulations aren't enough to guarantee
DRAM data retention?

MemTest is really a worst-case test - no computer will
never ever (talking about real world applications) have it's
memory filled with a single bit high and it's complement,
alternated on every address line. Like:

10000...
01111...
10000...

Why didn't they manipulate two PSUs to prove their theses? :
remove the cover on one PSU and locate it close to the RAM banks,
make the wires on the second PSU a bit longer and put it in an
empty case a few feet away.

" incomplete tests are worse than none at all"

thebiggreenone
Posts: 24
Joined: Sat Aug 02, 2003 9:23 am
Location: Seattle

Post by thebiggreenone » Sat Aug 02, 2003 11:31 am

Sledge wrote:On Anand's temp page
Tests were conducted inside an industry standard Chenming case with no fans running
I'm guessing there was no processor, mobo, or even a incandescent light bulb in the case which is why the case ambient is so low.
I am sure they are saying ambient temp is the air temp in the room.

nbac
Patron of SPCR
Posts: 142
Joined: Sat Jul 05, 2003 11:27 am
Location: Sweden

Post by nbac » Sat Aug 02, 2003 12:55 pm

Yes, that might be the case, but who cares about one temp
if you don't know the other?
Last edited by nbac on Sat Aug 02, 2003 1:20 pm, edited 1 time in total.

dukla2000
*Lifetime Patron*
Posts: 1465
Joined: Sun Mar 09, 2003 12:27 pm
Location: Reading.England.EU

Post by dukla2000 » Sat Aug 02, 2003 1:04 pm

Ralf Hutter wrote:
BaconTastesGood wrote:I found it interesting that they were asserting a 330W Antec wasn't capable of reliably supplying a P4 of any type -- is this hyperbole, misinformation, or the scary truth?
Can anyone comment on this?
I am P4 ignorant, but if 204W on the +12V is insufficient for any P4 then I'm from Mars. Complete bollocks would be my view of their statement: if I am wrong then I thank my fairy godmother I am in the AMD camp.
nbac wrote:Why didn't they manipulate two PSUs to prove their theses? :
remove the cover on one PSU and locate it close to the RAM banks,
make the wires on the second PSU a bit longer and put it in an
empty case a few feet away.
To me that table begs more questions than anything: is it EMC, wild fluctuations on 1 or more rails, sagging under load, noise from the mobo power circuits because of something in the psu, feedback loops because of psu design, moon phases or sun spots over 6 hours? They say they have been working on that test for a while - what are the results with 1 psu and varying DRAM? Varying CPU? Varying mobo? And in my heart I find it hard to believe memory errors can occur 'so easily' - why aren't we all noticing this phenomonen, or at least the consequences? I guess the % of PCs with ECC is <1%. Surely everyone running Prime, Folding (and other apps that are numerically sensitive) for months on end should notice these errors several times a week? Are all the distributed results erroneous - surely someone would have noticed by now? If you open a document in Word, leave your PC for 3 days and come back, how many characters have changed? Last time I looked it was none. (OK, I haven't done 3 days but I have certainly left stuff overnight, > 6 hours. And if this test made any sense I would expect to be a front row victim with my cheapo psus. :? )

Googling around leads me to believe DRAM refresh is related to the RAS precharge time which is a number of cycles. We are talking MHz for cycles, so why does it take 6 hours to show a failure? Are they measuring defects in the memory controller based on the power feeding the memory controller? At best 'non sense' is my view - did I miss the point :?:

aristide1
*Lifetime Patron*
Posts: 4284
Joined: Fri Apr 04, 2003 6:21 pm
Location: Undisclosed but sober in US

Post by aristide1 » Sat Aug 02, 2003 4:23 pm

If you open a document in Word, leave your PC for 3 days and come back, how many characters have changed?
In this example the memory values aren't changing at all by the pc, what happens when I run FAH for 72 hours or Prime 95? They're telling me I'd get at least 6 errors? And Standford U would find that acceptable?

OK, who's switching back to ECC memory?

Maybe we should cover every individual wire coming out of the psu with a wire sheath and ground it as well.

We wouldn't have these problems if we were still using vacuum tubes. :shock:

nbac
Patron of SPCR
Posts: 142
Joined: Sat Jul 05, 2003 11:27 am
Location: Sweden

Post by nbac » Sat Aug 02, 2003 11:41 pm

I'm glad to see not everyone bought Kristophers hypotheses,but there is
a thread about that (where it should be) at Anandtech Forum ( JPSJPS posts)
enought said!
Last edited by nbac on Sun Aug 03, 2003 8:28 am, edited 1 time in total.

dukla2000
*Lifetime Patron*
Posts: 1465
Joined: Sun Mar 09, 2003 12:27 pm
Location: Reading.England.EU

Post by dukla2000 » Sun Aug 03, 2003 12:43 am

Seems (at least for psus) Anandtech has joined that other site in the completely incompetent category.

I went to search Anandtech for more on this MemTest test - surprisingly nothing in a DRAM test a few days ago. But did find that thread as well. I love the following from jpsjps @ Anandtech:
I have never seen a review that is as pathetic as this one. I am shocked that Anandtech would post this review and make a complete fool of themselves.
Kristopher - Thanks for the PM! I see you got stuck doing a job out of your field of expertise and that explains the whole story.
RIP

aristide1
*Lifetime Patron*
Posts: 4284
Joined: Fri Apr 04, 2003 6:21 pm
Location: Undisclosed but sober in US

Post by aristide1 » Sun Aug 03, 2003 6:13 am

But the guy probably put an honest effort into it, so I found the remarks far more nasty than they should have been.

larrymoencurly
Posts: 170
Joined: Tue Dec 03, 2002 11:29 pm

Post by larrymoencurly » Mon Aug 04, 2003 8:34 am

What difference would it make to have MemTest86 wait six hours before reading back them memory rather than read it back right away? How would that be a better test of RF interference causing memory errors when the bits are refreshed every few milliseconds anyway? Is that much charge lost in the bit cell near the end of the refresh cycle, and how much more vulnerable does it make the memory to RFI? Has anybody tried goofing up a computer by holding a cell phone right next to the mobo? A couple of years ago that German PC magazine did that with a hard drive and caused it to crash (only one brand; others not affecte).

nbac
Patron of SPCR
Posts: 142
Joined: Sat Jul 05, 2003 11:27 am
Location: Sweden

Post by nbac » Mon Aug 04, 2003 11:00 am

I promised myself to leave this review behind me, but
obviously I couldn't. New things just keep popping up:

*Since when do Athlon MB's use the 3.3V rail to produce Vcore?
(it's either 5V or (not so often) 12V)

*The heatsinks of PC Power & Cooling TurboCool 475 reached
25.8'C under load, which means the test temperature must
have been even lower. Probably the stated temp : 21'C.

Combining a low test temperature with an unknown load
should make all test results pretty much useless.

Now, RIP

Post Reply