A7N8X-E running XP-M2500+ @2300MHz

[Paul Busby]

I ordered 2 sticks of Crucial 512MB PC3200 RAM last Sunday & fitted them
Tuesday night. having removed 2 sticks of Crucial 256MB PC2100 which ran
with slack timings @172MHz, 1:1, 12.5x & 1.575V.

Entered the bios & set the FSB to 200MHz with a multiplier of 10x, upped the
CPU supply to ~1.75V then rebooted - XP blue screened. Upped the RAM to 2.7V
& again crashed. Set the FSB to 190MHz & XP booted but was very unstable &
remained so after changing other settings. The system became stable @
177MHz - only a little more than I was able to achieve with the PC2100!

I value XP's ACPI S3 Standby mode but if I don't set System Performance to
Optimal in the bios (modded 1013), I loose it. I ended up setting the FSB to
166MHz & thought through what further settings I could tweak to get 200MHz
FSB & retain S3 Standby.

Pulled out one stick of RAM then reset the FSB to 200 then booted straight
into Memtest86 @ 7.5x. Got through a single run then repeated with the RAM
back in dual channel mode but in different slots, again Memtest86 passed.
Rebooted into XP & blue screen. Set the 2 sticks in single channel & XP
booted fine. Set 11.5x, Prime95 & everything else runs fine with my Zalman
7000Cu set for 1700rpm for temps between 50C idle (Case 28) & 60 running
burnk7- a little high but the fan still has plenty of headroom. The CPU is
set to 1.75V & would need to raise it to get to 2400MHz with a consequent
increase in noise.

Anyone care to speculate why dual channel mode won't run at all @ 200MHz at
any multiplier? I'll try the unmodded 1013 bios just to check it's not that.
Not a huge loss but it would be nice to get it working without going to LL
RAM with its heavy price premium & small gain in performance for my usage.

 

[Paul]

I ordered 2 sticks of Crucial 512MB PC3200 RAM last Sunday & fitted them
Tuesday night. having removed 2 sticks of Crucial 256MB PC2100 which ran
with slack timings @172MHz, 1:1, 12.5x & 1.575V.

Entered the bios & set the FSB to 200MHz with a multiplier of 10x, upped the
CPU supply to ~1.75V then rebooted - XP blue screened. Upped the RAM to 2.7V
& again crashed. Set the FSB to 190MHz & XP booted but was very unstable &
remained so after changing other settings. The system became stable @
177MHz - only a little more than I was able to achieve with the PC2100!

I value XP's ACPI S3 Standby mode but if I don't set System Performance to
Optimal in the bios (modded 1013), I loose it. I ended up setting the FSB to
166MHz & thought through what further settings I could tweak to get 200MHz
FSB & retain S3 Standby.

Pulled out one stick of RAM then reset the FSB to 200 then booted straight
into Memtest86 @ 7.5x. Got through a single run then repeated with the RAM
back in dual channel mode but in different slots, again Memtest86 passed.
Rebooted into XP & blue screen. Set the 2 sticks in single channel & XP
booted fine. Set 11.5x, Prime95 & everything else runs fine with my Zalman
7000Cu set for 1700rpm for temps between 50C idle (Case 28) & 60 running
burnk7- a little high but the fan still has plenty of headroom. The CPU is
set to 1.75V & would need to raise it to get to 2400MHz with a consequent
increase in noise.

Anyone care to speculate why dual channel mode won't run at all @ 200MHz at
any multiplier? I'll try the unmodded 1013 bios just to check it's not that.
Not a huge loss but it would be nice to get it working without going to LL
RAM with its heavy price premium & small gain in performance for my usage.

It could be a chip design limitation. Imagine 160 or more output signals
switching simultaneously. That is a significant electrical disturbance,
and amps of current flow into or out of the chip during a transition.
(The normal solution to this problem, is to add more pins to a chip,
like what happened to the LGA775 processor socket. But adding more
power and ground signals, increases the cost of the chipset. In
the case of a LGA775 processor, the extra pins are there due to the
sheer number of DC amps flowing into the chip, as much as anything
else.)

With your mobile, you have many options for accomodating the RAM.
By keeping the CPU:Mem ratio 1:1, running the memory at whatever
speed it will take in dual channel, you can use the multiplier to
make the final adjustment. Using a hacked BIOS that supports
Command Rate 2T timing (CPC off), will allow 200MHz to be reached,
but at the expense of reduced memory bandwidth. Of all the techniques
I know of, all of the fixes result in reduced memory bandwidth,
so the LL RAM is the only brute force approach that works well.
I cannot provide a sound technical argument for why LL RAM is
making a difference, as the interface is synchronous, and CAS2
versus CAS3 should be a totally internal issue for the RAM, not
affecting interface timing.

Paul

 

[Paul Busby]

Thus spake Paul:

It could be a chip design limitation. Imagine 160 or more output
signals switching simultaneously. That is a significant electrical
disturbance, and amps of current flow into or out of the chip during
a transition. (The normal solution to this problem, is to add more
pins to a chip, like what happened to the LGA775 processor socket.
But adding more power and ground signals, increases the cost of the
chipset. In
the case of a LGA775 processor, the extra pins are there due to the
sheer number of DC amps flowing into the chip, as much as anything
else.)

With your mobile, you have many options for accomodating the RAM.
By keeping the CPU:Mem ratio 1:1, running the memory at whatever
speed it will take in dual channel, you can use the multiplier to
make the final adjustment. Using a hacked BIOS that supports
Command Rate 2T timing (CPC off), will allow 200MHz to be reached,
but at the expense of reduced memory bandwidth. Of all the techniques
I know of, all of the fixes result in reduced memory bandwidth,
so the LL RAM is the only brute force approach that works well.
I cannot provide a sound technical argument for why LL RAM is
making a difference, as the interface is synchronous, and CAS2
versus CAS3 should be a totally internal issue for the RAM, not
affecting interface timing.

Paul

Thanks Paul, I don't know what added demands dual channel RAM makes to a
system - I initially speculated it wouldn't make any at all but was
obviously wrong. I suspect running 2T & dual channel would be virtually the
same as I'm getting now. The hacked bios is the 1T command version, as is
the standard unmodded one. I am also very relieved that I can now reach
200MHz.

I can hardly complain with the current rock solid stability I'm getting with
the added grunt. Now I've gone with 1GB RAM @ 200MHz, the performance over &
above my previous XP1800+ has been worth while as a swansong to the 32bit
Athlon.

I had a further play by setting the FSB to 210 & upping the Vcore to 1.8V (3
4 4 9) but the added performance was virtually zero as measured with Sandra,
so have returned to 200MHz & 1.75 Vcore but with the Zalman fan turned up to
2220rpm (6000Cu not 7000Cu as I wrote originally) with ACPI S3 mode
restored. Quite why setting anything but optimal under System Performance
kills S3, I can't say but suspect a Bios bug unless this problem is unique
to me which I doubt. This upgrade is almost complete having recently bought
a XFX GF5200 which leaves a quieter PSU for the near future.

 

[Paul]

Thanks Paul, I don't know what added demands dual channel RAM makes to a
system - I initially speculated it wouldn't make any at all but was
obviously wrong. I suspect running 2T & dual channel would be virtually the
same as I'm getting now. The hacked bios is the 1T command version, as is
the standard unmodded one. I am also very relieved that I can now reach
200MHz.

I can hardly complain with the current rock solid stability I'm getting with
the added grunt. Now I've gone with 1GB RAM @ 200MHz, the performance over &
above my previous XP1800+ has been worth while as a swansong to the 32bit
Athlon.

I had a further play by setting the FSB to 210 & upping the Vcore to 1.8V (3
4 4 9) but the added performance was virtually zero as measured with Sandra,
so have returned to 200MHz & 1.75 Vcore but with the Zalman fan turned up to
2220rpm (6000Cu not 7000Cu as I wrote originally) with ACPI S3 mode
restored. Quite why setting anything but optimal under System Performance
kills S3, I can't say but suspect a Bios bug unless this problem is unique
to me which I doubt. This upgrade is almost complete having recently bought
a XFX GF5200 which leaves a quieter PSU for the near future.

I found my mobile didn't like too much voltage. I think I
ended up at 1.65, because I don't think I can POST if I
set mine to 1.75V. I run standard 3200+ settings now (200x11),
as I really didn't find adding a few hundred more MHz helped
significantly with benchmarks. I'm pretty happy with how it
turned out, but the solid week of testing to figure out
the RAM issues, I could have done without.

Paul

 

[Paul Busby]

Thus spake Paul:

I found my mobile didn't like too much voltage. I think I
ended up at 1.65, because I don't think I can POST if I
set mine to 1.75V. I run standard 3200+ settings now (200x11),
as I really didn't find adding a few hundred more MHz helped
significantly with benchmarks. I'm pretty happy with how it
turned out, but the solid week of testing to figure out
the RAM issues, I could have done without.

Paul

Ran Prime95 again for 7hrs overnight without errors. I'm rather surprised
that Memtest86 will pass a basic test but Windows itself will crash. Perhaps
a little patience by running it much longer would have shown errors. My next
tweak will be to lower Vcore if possible - it ran 12.5x 172MHz at the lowest
this board will go (1.575V IIRC). Getting the FSB to 200 definitely hit the
sweetspot performance wise. I'm very happy with the results, albeit without
dual channel mode. I was very impressed how far by old Crucial PC2100 could
be pushed.