Memory bench quirks

Status
Not open for further replies.
S

Skid

Joined
Aug 5, 2003
Messages
190
Reaction score
0
Doing a little head-scratching since I went from 512 megs to a gig. Wondering if anyone can shed some light on these results.

First the specs: Abit IC7, P4 2.4 @ 275x12=3ghz. Ram is 256-meg sticks of Buffalo Tech PC3700, running 2,3,2,6 at 5:4 ratio, 220 mhz.

With two sticks, Sandra Max memory gives me:
buffered: 5630+ int, 5820+ flt
unbuffered: 2420+ int, 2460+, flt

With four sticks, Sandra says:
buffered: 5000+ int, 5000+ flt
unbuffered: 2550+ int, 2600+ flt

In other words, the unbuffered scores went up noticeably, which I was expecting from reading the "Memory Holy Grail" article at www.anandtech.com

What I didn't expect was that the buffered scores went way down. The only clue I have is that now Sandra says PAT is not working with four sticks -- even though it should be there on my 875P chipset.

Any theories? Any suggestions? (Other than ignoring Sandra and getting on with my life.)
 
Skid wrote:
Doing a little head-scratching since I went from 512 megs to a gig. Wondering if anyone can shed some light on these results.

First the specs: Abit IC7, P4 2.4 @ 275x12=3ghz. Ram is 256-meg sticks of Buffalo Tech PC3700, running 2,3,2,6 at 5:4 ratio, 220 mhz.

With two sticks, Sandra Max memory gives me:
buffered: 5630+ int, 5820+ flt
unbuffered: 2420+ int, 2460+, flt

With four sticks, Sandra says:
buffered: 5000+ int, 5000+ flt
...
Click to expand...

....................................
Skid ,

A damn 'fair' <?> ...
dunno either ... "it" dont compute on my abacus either.

**IF** you get/find an answr from article_author=Prometheus ... get back to "us" , if you would please ...

About all I can ?manage/swag? is a timing issue ...
mis_matched ' pairs of "pairs" ' .... ??? assuming bought a diff times ... ie ... Pair #_1 ==> Module_1 & _2 Pair #_2 ==> " _3 & _4

Dunno ... dont even got a good 'swag', even.

............................................................ A 'suggestion' ... ask the article author "Why /// HowCome " , maybe ???

Go to : (a) & register to "ask" ... (b) otherwize just read. ==> http://www.abxzone.com/

To 'register' :
==> http://www.abxzone.com/forums/register.php?s=&action=signup ... register as an 'adult' ... :-)

Or , just 'read' , not 'ask' :
==> http://www.abxzone.com/forums/showthread/t-53568.html author = Prometheus

............................................................ If you "start" at the ABX top page ... look for :

.. Forums
... Components & Peripherals Zone
... Mamory
... Discussions, troubleshooting, tweaks and overclocking related to memory

==> http://www.abxzone.com/forums/forumdisplay/f-16.html gets about 13 pgs 'worth' of posts ... hit "2", in horizontal list of '13' pages ...

get 2nd page of posts ...
then mouse 'scan' left col of "titles' for art# = [53568] at the URL 'end' ... ==> http://www.abxzone.com/forums/forumdisplay/f-16-p-2.html ... pg2 of 13

finally get to [53568] :
==> http://www.abxzone.com/forums/showthread/t-53568.html author = Prometheus

Sorry for the 'rant' ...
its 'tough' to talk to an Eskimo about "snow" ...
when all you 'know' is [snow],
Qed.
 
"0_Qed" wrote:

Skid wrote:

Doing a little head-scratching since I went from 512 megs to a gig.
Wondering if anyone can shed some light on these results.

First the specs: Abit IC7, P4 2.4 @ 275x12=3ghz. Ram is 256-meg sticks
Click to expand...
of
Buffalo Tech PC3700, running 2,3,2,6 at 5:4 ratio, 220 mhz.

With two sticks, Sandra Max memory gives me:
buffered: 5630+ int, 5820+ flt
unbuffered: 2420+ int, 2460+, flt

With four sticks, Sandra says:
buffered: 5000+ int, 5000+ flt
unbuffered: 2550+ int, 2600+ flt

In other words, the unbuffered scores went up noticeably, which I was
expecting from reading the "Memory Holy Grail" article at
Click to expand...
www.anandtech.com
What I didn't expect was that the buffered scores went way down. The only clue I have is that now Sandra says PAT is not working with four sticks -- even though it should be there on my 875P chipset.

Any theories? Any suggestions? (Other than ignoring Sandra and getting on with my life.)

....................................
Skid ,

A damn 'fair' <?> ...
dunno either ... "it" dont compute on my abacus either.

**IF** you get/find an answr from article_author=Prometheus ... get back to "us" , if you would please ...

...
Click to expand...

Good idea. I e-mailed the author and will let you know if he replies. The sticks were bought at different times, but labels and specs match, all are Winbond BH-5.

What I would expect from the article is that the unbuffered scores would go up a little, while the buffered scores would stay about the same.

The unbuffered went up more than I thought, and the buffered scores dropped like a rock. The PAT disabled thing is another head-scratcher, though someone in the Abit forum suggested that could be a bios glitch that misreports when all four slots are full.

Nothing else seems noticeably different, 3DMark '03 went up a piddling 20 points, probably withing the margin of error.
 
Skid wrote:

Good idea. I e-mailed the author and will let you know if he replies. The sticks were bought at different times, but labels and specs match, all are Winbond BH-5.
Click to expand...
...snip...

:-)
In the mean time ...
a bunch of NON_pono lecture slides ... re: mamory ==> http://www.ece.umd.edu/courses/enee759h.S2003/lectures/

Now ... I'm curious too ...
:-)

Qed.
 
"0_Qed" wrote:

Skid wrote:

Good idea. I e-mailed the author and will let you know if he replies. The sticks were bought at different times, but labels and specs match, all
Click to expand...
are
Winbond BH-5.
...snip...

:-)
In the mean time ...
a bunch of NON_pono lecture slides ... re: mamory
==> http://www.ece.umd.edu/courses/enee759h.S2003/lectures/

Now ... I'm curious too ...
:-)

Qed.
Click to expand...

Aw, I didn't know you were going to make me learn stuff!

Where's the Cliff notes?
 
I have read that there is additional overhead with more sticks... don't know if that is what is doing it. having PAT disabled could certainly have a negative impact.
 
"TomG" wrote:

I have read that there is additional overhead with more sticks... don't know if that is what is doing it. having PAT disabled could certainly have a negative impact.
Click to expand...

Yeah, I just wish there was some way to tell if PAT is really off, or just the bios passing out false info, as someone in the Abit forum suggested.

Sandra seems to be the only benchmark showing a big difference, and it wobbles around so much and is wrong so often I'm not sure what to think.

The bandwidth in MemTest goes up with fsb and doesn't seem to be any different from two sticks to four.

It just drives me crazy that unbuffered Sandra bandwidth -- which is supposed to be the real deal -- is going up while the buffered scores are headed south.
 
Skid wrote:
Aw, I didn't know you were going to make me learn stuff!
Click to expand...

`Smatter ???
Ruptured yer neuron(snglr).
:-)

Where's the Cliff notes?
Click to expand...

Dunno ...

Elderly "Audits" dont get no nothing ...
but, neither
. do they 'sweat' it,
or
. make/take notes,
or
. 'pay'.
<VBFG>

Qed.
 
Skid wrote:
Yeah, I just wish there was some way to tell if PAT is really off, or just the bios passing out false info, as someone in the Abit forum suggested.
Click to expand...

:-)
Yet another fair <?>
Whens "it" on/off ???
FIIK.
It aint got 'no' Big Beautiful Button, apparently.

==> ftp://download.intel.com/design/chipsets/applnots/

252730.htm ... the 'blip'
25273001.pdf ... wht ppr

Intel � 875P Chipset Memory
Configuration Guide
White Paper
April 2003

Me?
I think I'll worry more `bout understanding "String Theory" then poor old odd PAT ... 'poop'. Yah rite ... I'll get to "that" rsn. "It" almost sounds like more Marketing Magic Mumbo

:-)
Qed.
 
CPU-Z will tell you if PAT is on or off. version 1.19a or newer... on it's memory screen, it tells whether you are in dual channel mode and whether performance mode is enabled or not. I have a copy of it on my server under /utilities or you can get it at http://www.cpuid.com/
 
Thanks, Tom. I already had CPU-Z, but hadn't checked it. It also shows PAT disabled with four sticks.

That's a bummer, since I paid an extra $20 for an IC7 over an IS7 just to make sure I got PAT. There is a new bios due out any day, and I can only keep my fingers crossed that Abit will address this problem.

There is a great new article at www.lostcircuits.com that goes into detail explaining this whole subject. My conclusions from reading it are:

1.) Some OEMs chose to disable PAT in bios when all four DIMM slots are full to minimize addressing hassles, not the best choice for performance but the lazy way to avoid fine tuning.

2.) The difference between two sticks and four sticks is otherwise barely perceptible. Sandra buffered scores are apparently one of the most exaggerated indicators.

I can live with both these revelations. But it pisses me off that in typical fashion Abit doesn't mention any of this in any of its documentation, which means the only way to find out how a particular hardware config or bios setting works is trial and error.

That's more than a little frustrating, especially in a case where I blew a week's beer budget to make my rig run better, only to find out I'd slowed it down a little.
 
was the week's beer budget blown by the extra ram purchase (!!) or by the $20 extra for the board. if the first, then you need help and if the second, then you need help. just a different kind of help.
 
"TomG" wrote:

was the week's beer budget blown by the extra ram purchase (!!) or by the $20 extra for the board. if the first, then you need help and if the second, then you need help. just a different kind of help.
Click to expand...

The former. I have a habit of setting 'em up for all the sheep in the corral.
 
Status
Not open for further replies.

Latest threads

Back
Top