Epox 8RDA+:changing multiplier affects performace a lot

[Iolao]

This board behaves strange... with ram timings of cas2 2-2-5 and 195 fsb async if I set multiplier to 10 I have a grea bandwith (efficiency in sandra is 98%-93% which is great!!!) if I lower multiplier to 9,5 or 9 or whatever else bandiwth efficiency lowers to 93%-87% which is really bad IMHO. Checking with wpcredit woithout a pcr file I can only see a lot of register changing when playing with multipliers but I cannot even set them back to the values of 10x because they change in a different way than my sets... BTW I'm using latest beta bios (3 dec). Another problem I was having was tha corruption of data at 195+ fsb. Putting a fan blowing air on the northbridge seened to solve the problem and now I can boot into windows at 200 with no problems. My ram can hit 225 with those timings flawless as I tested my two sticks in asynch with a dos utility so I'm wondering if the problem is actually the northbridge...

This is my system spec:
AXP 1600+ @ 196x10
Epox 8RDA+
2xOCZ PC3500 256MB at cas 2 2-2-5 with cpu interface set to aggressive (optimal dosen't solve anything)
Radeon 8500
Western Digital 80GB 8Mb cache

What do you think about it?

[eliteone]

What fan did you put on your northbridge?

[Iolao]

I didnd't put any fan directly, there's a 12x12 fan 4-5cm far from the northbridge blowing a lot of fresh air on

[baller02]

same problem here

with multiplier under 10 i get 2960mb's with sandra membench
@200fsb with sync dual channel (2-4-2-2 timings).
with multiplier 10@ 200fsb i get over 3100 mb's!!!!!

i tried all three bios version which are available and all have the same anomaly

can anyone explain that

[baller02]

i tried it also @ 166fsb

with multiplier under 10 i get 24xxmb's
with 10 or higher i get 2600mb's

register e4 and e5 are:

multiplier under 10
e4=16
e5=30

multiplier 10 and above

e4=15
e5=16

register f3 changed always if another multiplier is selected

register e4 and e5 only changes between multplier 9,5 and 10

all other register doesn't change

register e4 and e5 seems to have influence of performance!!
but they can't changed with wpcredit

ET please give us a bios version without this failure!!!

[kowalski]

Originally posted by baller02
same problem here

with multiplier under 10 i get 2960mb's with sandra membench
@200fsb with sync dual channel (2-4-2-2 timings).
with multiplier 10@ 200fsb i get over 3100 mb's!!!!!

i tried all three bios version which are available and all have the same anomaly

can anyone explain that

if you're leaving the fsb at 200 and altering the multiplier then your sandra mem score will be affected cos it's heavily dependent on cpu speed.

[eliteone]

haha kowalski...my last name is Kowalewski . thats pronounced koveleski.

[Iolao]

Originally posted by kowalski


if you're leaving the fsb at 200 and altering the multiplier then your sandra mem score will be affected cos it's heavily dependent on cpu speed.

that's not correct. Sandra performance is indeed affected by cpu speed but only a little bit, performance change between 9.5 and 10 multiplier is a lot, believe me.

[franlofter]

I wonder if a bios fix can solve this? hope so

[baller02]

they must set register e4 and e5 @ multi under 10 like they are above 10!!!

that's it

[kowalski]

Originally posted by eliteone
haha kowalski...my last name is Kowalewski . thats pronounced koveleski.

hehe, it's nothing like my real name, i just like it...

[Galifrey]

dont forget bandwidth is cpu speed dependent as well...

[kowalski]

they don't believe us galifrey...

[Iolao]

maybe it's you that don't believe me.... if at 195x10 bandwith is 3050/2860, at 195x9.5 is 28xx/26xx, at 195x9 is 28xx/26xx, at 195x8.5 is 28xx/26xx, what do you think about cpu speed? This is an abnormal behaviour, at least for me...

[Swifty2002]

"if at 195x10 bandwith is 3050/2860, at 195x9.5 is 28xx/26xx, at 195x9 is 28xx/26xx, at 195x8.5 is 28xx/26xx, what do you think about cpu speed?"

at 195x10 you have 1950mhz... at 195x9.5 you have 1852mhz, at 195x9 you have 1755mhz, at 195x8.5 you have 1657mhz

now try to grasp that Sisoft Sandra's memory benchmark is CPU intensive... meaning, that given the same FSB speed and different CPU speeds, the higher CPU speeds will give better bandwith benchmarks...

That is why you are seeing that using the 10x multiplier has greater bandwith than the other lower multipliers...

[Iolao]

ok, so you can see cpu dependant benchs from 10x to 9.5x, but not from 9.5 to 9? And so on... Do you think it's normal?

[Iolao]

BTW I cannot try 195x10.5 because my cpu can't do that, but I bet bandwith will be better with 10x multiplier

[kowalski]

ok, so there's some non-linearity in the scaling relative to CPU speed. Sandra's not the be-all-and-end-all you know. in fact it's not all that at all...

try running unbuffered tests. settings like these:

http://members.madasafish.com/~pevis...unbuffered.jpg

[pic nicked (not hotlinked) from OcUK - cheers Gibbo ]

also try the PCMark memory test, and see what happens...

[Galifrey]

Yeah ditch the cpu dependancy a bit and things will level out...

Kowalski is spot on (sorry bud shoulda read your comment before I added mine)

[kowalski]

Originally posted by Galifrey
Yeah ditch the cpu dependancy a bit and things will level out...

Kowalski is spot on (sorry bud shoulda read your comment before I added mine)

np, you're backing me up mate