Bios settings for TWINX2048-3200C2 on Asus A8N32-SLI Deluxe ?

  • Thread starter Thread starter Skybuck Flying
  • Start date Start date
S

Skybuck Flying

Hello,

I bought four of these memory chips:

Corsair TWINX2048-3200C2 DDR SDRAM, 1 GB, PC3200, 400 MHz (Double Side,
Double Data Rate, Supported by motherboard).

I also received a green card with instructions on it which shows how to
install the memory chips into the motherboard's memory slots.

Today I ran my own (32 bit) memory benchmark (on windows xp 64 bit) which
reads as many integers from the memory as possible. The results were quite
shocking... only 2.4 million bytes per second. Even my 6 year old pentium
III 450 mhz overclocked to 600 mhz is able to achieve 3.1 million bytes per
second.

My first suspect for the poor memory performance/access time is the memory
chips themselfes and there settings.

The support on the website of Corsair is quite bad. There are no settings to
be found anywhere. Their specification is mostly a marketing ad. That's not
a specification !

I would like to know all the settings about these memory chips !

Is there a real/full specification of these memory chips and if so were can
I download/view it ?

Failing that... does anybody know the correct settings for these memory
chips in the bios of the motherboard asus a8n32-sli ?

Which of these settings have to be changed and what would be there values ?:

Memclock mode: [Auto]
MCT Timing mode: [Auto]
Cas latency (cl): [Auto]
TRAS: [Auto]
TRP: [Auto]
TRCD: [Auto]
TRRD: [Auto]
TRC: [Auto]
TRFC: [Auto]
TRWT: [Auto]
User Config Mode: [Auto]
Read Preamble: 9.5 ns (?)
Asyc Latency: 11.0 ns (?)
Bank interleaving: [Auto]
Burst Length: [4 Beats]
Hardware Memory Hole: [Enabled] <- I changed that for 4 GB support.

MCT Extra Timing Mode:
TREF: 7.8 us
TWCL: 1
R/W Queue Bypass Count: 8
ByPass Max: 4
Idle Cycle Limit: 16
Dynamic Idle Cycle Center: Enabled
DDR Driving Strength: Normal
Enable 32-byte granularity: Disabled (what is this ?)
Twr: 3 clk
DDR Input Strobe Skew: Disabled
DDR Data Driving Strength: Disabled

ECC Configuration
Dram ECC Enable: Disabled
L2 Cache BG Scrub: Disabled
Data Cache BG Scrub: Disabled

DDR Vcore: Auto
DDR Clock Skew: Auto

(Message also posted on Corsair support forum)

Bye,
Skybuck.
 
"Skybuck said:
Hello,

I bought four of these memory chips:

Corsair TWINX2048-3200C2 DDR SDRAM, 1 GB, PC3200, 400 MHz (Double Side,
Double Data Rate, Supported by motherboard).

I also received a green card with instructions on it which shows how to
install the memory chips into the motherboard's memory slots.

Today I ran my own (32 bit) memory benchmark (on windows xp 64 bit) which
reads as many integers from the memory as possible. The results were quite
shocking... only 2.4 million bytes per second. Even my 6 year old pentium
III 450 mhz overclocked to 600 mhz is able to achieve 3.1 million bytes per
second.

My first suspect for the poor memory performance/access time is the memory
chips themselfes and there settings.

The support on the website of Corsair is quite bad. There are no settings to
be found anywhere. Their specification is mostly a marketing ad. That's not
a specification !

I would like to know all the settings about these memory chips !

Is there a real/full specification of these memory chips and if so were can
I download/view it ?

Failing that... does anybody know the correct settings for these memory
chips in the bios of the motherboard asus a8n32-sli ?

Which of these settings have to be changed and what would be there values ?:

Memclock mode: [Auto]
MCT Timing mode: [Auto]
Cas latency (cl): [Auto]
TRAS: [Auto]
TRP: [Auto]
TRCD: [Auto]
TRRD: [Auto]
TRC: [Auto]
TRFC: [Auto]
TRWT: [Auto]
User Config Mode: [Auto]
Read Preamble: 9.5 ns (?)
Asyc Latency: 11.0 ns (?)
Bank interleaving: [Auto]
Burst Length: [4 Beats]
Hardware Memory Hole: [Enabled] <- I changed that for 4 GB support.

MCT Extra Timing Mode:
TREF: 7.8 us
TWCL: 1
R/W Queue Bypass Count: 8
ByPass Max: 4
Idle Cycle Limit: 16
Dynamic Idle Cycle Center: Enabled
DDR Driving Strength: Normal
Enable 32-byte granularity: Disabled (what is this ?)
Twr: 3 clk
DDR Input Strobe Skew: Disabled
DDR Data Driving Strength: Disabled

ECC Configuration
Dram ECC Enable: Disabled
L2 Cache BG Scrub: Disabled
Data Cache BG Scrub: Disabled

DDR Vcore: Auto
DDR Clock Skew: Auto

(Message also posted on Corsair support forum)

Bye,
Skybuck.
Click to expand...

The datasheets are on the XMS memory web page. They aren't hard to find.

http://www.corsairmicro.com/corsair/products/specs/twinx2048-3200c2.pdf

"Tested at aggressive latency settings of 2-3-3-6 for all platforms.
2-3-3-6 on Intel and AMD based systems
Test voltage: 2.75V"

Slide 24 here, tells you the order of parameters:

http://corsairmicro.com/corsair/products/tech/memory_basics/153707/index.html

"CAS Latency, tRCD, tRP, tRAS"

When using four sticks of memory, the expected settings would be
DDR400 Command Rate 2T, which is better than the alternative
setting of DDR333 Command Rate 1T. You cannot expect to run
DDR400 Command Rate 1T with four memory sticks.

There really isn't much point in adjusting other parameters
on that memory page.

As for your little test program, try the following. Download
memtest86+ from www.memtest.org . The program will format a
testing floppy for you. Boot your new computer with the boot
floppy disk. When memtest86+ starts, there are three bandwidth
indicators in the upper left hand corner. The third one down,
monitors memory bandwidth (unbuffered). Specifically, what the
code does, is it flushes the cache before sequentially reading
a large chunk of memory.

Pop that test floppy into your old system, then into your
new system, and you should see an appropriate increase in
performance. The AMD memory controller does a pretty good
job with DDR.

The memtest86+ source code is available for download. I've
downloaded the source, and made a minor modification to the
program, so I could test memory bandwidth at different addresses
within system memory. I've had a look at the code and
their approach seems reasonable. Basically flush the cache,
and then read a block of memory larger than any cache, a
number of times, and average the results.

Are you certain that the time base measurement code in
your test program, is able to deal with the faster operation
of your new computer ? There are many ways to explain failures
of your test program.

The top post here, lists numbers like 368MB/sec for some PC133
SDRAM.

http://groups.google.ca/group/alt.comp.periphs.mainboard.asus/browse_frm/thread/249aeea20d9a440d

There is a picture of a memtest86+ screen here, for a 3500+
with 2x1GB memory. It lists 3908MB/sec, or roughly 10x
the memory bandwidth of the other system. This 3500+ is running
at 2700MHz and is overclocked. The memory is also overclocked
big time. I expect your system will be a bit lower
than that.

http://www.abxzone.com/forums/showthread.php?p=1238439

HTH,
Paul
 
Hello,

I bought four of these memory chips:

Corsair TWINX2048-3200C2 DDR SDRAM, 1 GB, PC3200, 400 MHz (Double Side,
Double Data Rate, Supported by motherboard).
Click to expand...

Double sided is a bad idea.

Supported by motherboard doesn't specify how they'll
perform. As you can see, they do in fact run, just slow as
shit.

Welcome to the world of 2T performance and remember, if
people weren't so stupid you could at least have had the
benefit of ECC, if not registered memory. Yep, there was a
reason for all that stuff.
 
Paul said:
"Skybuck said:
Hello,

I bought four of these memory chips:

Corsair TWINX2048-3200C2 DDR SDRAM, 1 GB, PC3200, 400 MHz (Double Side,
Double Data Rate, Supported by motherboard).

I also received a green card with instructions on it which shows how to
install the memory chips into the motherboard's memory slots.

Today I ran my own (32 bit) memory benchmark (on windows xp 64 bit) which
reads as many integers from the memory as possible. The results were
quite
shocking... only 2.4 million bytes per second. Even my 6 year old pentium
III 450 mhz overclocked to 600 mhz is able to achieve 3.1 million bytes
per
second.

My first suspect for the poor memory performance/access time is the
memory
chips themselfes and there settings.

The support on the website of Corsair is quite bad. There are no settings
to
be found anywhere. Their specification is mostly a marketing ad. That's
not
a specification !

I would like to know all the settings about these memory chips !

Is there a real/full specification of these memory chips and if so were
can
I download/view it ?

Failing that... does anybody know the correct settings for these memory
chips in the bios of the motherboard asus a8n32-sli ?

Which of these settings have to be changed and what would be there values
?:

Memclock mode: [Auto]
MCT Timing mode: [Auto]
Cas latency (cl): [Auto]
TRAS: [Auto]
TRP: [Auto]
TRCD: [Auto]
TRRD: [Auto]
TRC: [Auto]
TRFC: [Auto]
TRWT: [Auto]
User Config Mode: [Auto]
Read Preamble: 9.5 ns (?)
Asyc Latency: 11.0 ns (?)
Bank interleaving: [Auto]
Burst Length: [4 Beats]
Hardware Memory Hole: [Enabled] <- I changed that for 4 GB support.

MCT Extra Timing Mode:
TREF: 7.8 us
TWCL: 1
R/W Queue Bypass Count: 8
ByPass Max: 4
Idle Cycle Limit: 16
Dynamic Idle Cycle Center: Enabled
DDR Driving Strength: Normal
Enable 32-byte granularity: Disabled (what is this ?)
Twr: 3 clk
DDR Input Strobe Skew: Disabled
DDR Data Driving Strength: Disabled

ECC Configuration
Dram ECC Enable: Disabled
L2 Cache BG Scrub: Disabled
Data Cache BG Scrub: Disabled

DDR Vcore: Auto
DDR Clock Skew: Auto

(Message also posted on Corsair support forum)

Bye,
Skybuck.
Click to expand...

The datasheets are on the XMS memory web page. They aren't hard to find.

http://www.corsairmicro.com/corsair/products/specs/twinx2048-3200c2.pdf

"Tested at aggressive latency settings of 2-3-3-6 for all platforms.
2-3-3-6 on Intel and AMD based systems
Test voltage: 2.75V"

Slide 24 here, tells you the order of parameters:

http://corsairmicro.com/corsair/products/tech/memory_basics/153707/index.html

"CAS Latency, tRCD, tRP, tRAS"

When using four sticks of memory, the expected settings would be
DDR400 Command Rate 2T, which is better than the alternative
setting of DDR333 Command Rate 1T. You cannot expect to run
DDR400 Command Rate 1T with four memory sticks.

There really isn't much point in adjusting other parameters
on that memory page.

As for your little test program, try the following. Download
memtest86+ from www.memtest.org . The program will format a
testing floppy for you. Boot your new computer with the boot
floppy disk. When memtest86+ starts, there are three bandwidth
indicators in the upper left hand corner. The third one down,
monitors memory bandwidth (unbuffered). Specifically, what the
code does, is it flushes the cache before sequentially reading
a large chunk of memory.

Pop that test floppy into your old system, then into your
new system, and you should see an appropriate increase in
performance. The AMD memory controller does a pretty good
job with DDR.

The memtest86+ source code is available for download. I've
downloaded the source, and made a minor modification to the
program, so I could test memory bandwidth at different addresses
within system memory. I've had a look at the code and
their approach seems reasonable. Basically flush the cache,
and then read a block of memory larger than any cache, a
number of times, and average the results.

Are you certain that the time base measurement code in
your test program, is able to deal with the faster operation
of your new computer ? There are many ways to explain failures
of your test program.

The top post here, lists numbers like 368MB/sec for some PC133
SDRAM.

http://groups.google.ca/group/alt.comp.periphs.mainboard.asus/browse_frm/thread/249aeea20d9a440d

There is a picture of a memtest86+ screen here, for a 3500+
with 2x1GB memory. It lists 3908MB/sec, or roughly 10x
the memory bandwidth of the other system. This 3500+ is running
at 2700MHz and is overclocked. The memory is also overclocked
big time. I expect your system will be a bit lower
than that.

http://www.abxzone.com/forums/showthread.php?p=1238439
Click to expand...

Hello,

Your story is consistent with what I tried.

I wasn't sure if command rate t1 would work so I tried it =D then I got a
little worried. The computer completely stopped responding, no screen, no
boot screen, no beep, no nothing, I couldn't even turn it off or reboot.
Finally I had to turn off/on the computer via the power switch.

Fortunately for me the asus motherboard and american trends bios is able to
detect this overclock failure and reports it as such.

Then I tried command rate T2 and this time it worked.

I also tested memtest 1.65. Since the rawrite.exe utility does not work in
windows xp 64 bit, I installed it as follows. Fortunately for me I still had
a very cool ms-dos bootdisk... First I thought about using the windows 98
startup disk since it was the first disk in the box and had to make a quick
decision... but I already knew that worked... fortunately for me the boot
time is 10 seconds so I could still switch the boot disk... I wondered if
ms-dos 6.22 boot disk/os would still work on my new dream pc... so I put it
in... and haven't used that for a few years or so. Fortunately for me
everything worked perfectly... It worked even better than I ever imagined...
when I made the bootdisk I was even smart enough to install a microsoft ram
drive =D So it has default 4 MB of ram drive space =D and even qedit ;) but
ok.

So I was able to simply copy rawrite.exe and other files to the ramdisk...
since I wasn't sure if starting from a floppy would work... maybe that would
work as well... anyway... so finally I made the new disk. First I wanted to
make it on my old pc... but it doesn't have a floppy drive anymore lol... so
the point of the story is... where there is a will there is a way ;)
Ofcourse I first tried the iso image.. but that didn't work... rar and
winimage showed an empty iso image.

Now back to the memory/benchmarking test etc.

First I ran the memory testing before changing any memory settings.
Everything worked ok.

The speed reported was:

16 GB/sec (I think it was called L1 cache)
3.5 GB/sec(I think it was called L2 cache)
2 GB/sec (I think this was main memory or controller or something)

Then I changed the default memory settings to command rate t2 and 2-3-3-6
and ran the test again.

Now it reported this:

16 GB/sec
4 GB/sec
2 GB/sec

Finally I also run my own 32 bit benchmark and it reported still a somewhat
disappointing: 2.5 million bytes / sec

It's still 100.000 bytes faster than with the old memory settings so the new
memory settings does seem to help a bit... but only slighty...

So all in all the new memory settings aren't really worth bothering with..
or taking any corruption risks... but for now I will simply keep these
settings for the coming weeks/month/years to see what happens etc... It will
probably run just fine... so no need to run with any slower settings... but
all in all it wouldn't matter much.

Also the difference in the L2 cache is kinda weird... where is the L2 cache
located ? Is it in the memory chips or is it on the cpu or somewhere in
between ? It could just be a random measurement glitch/curiosity ;)

Finally I also noticed a little bit more of freezing when running both
benchmarks and other programs... when trying to move the windows around...

And finally when I ran two benchmarks, one benchmark instance didn't run at
all.. .which is really weird... my benchmark is mostly about testing how
many integers can be read randomly from memory and there is maybe also some
writing involved... not sure how the code gets compiled... so it's more
about access time than bandwidth. I have seen some very strange readings...
once I saw a reading of 500 MB/sec... now I tested another application
actually running two instances and now it only achieved 5 MB/sec through
loopback adapter though/winsock etc.. it's kinda strange.. since the same
application reported 500 MB/sec when testing with two computers... maybe I
am being delirius ;) but I don't think so ;)

I have seen some weird bugs on windows xp 64 bit with 32 bit applications,
also dos programs won't work etc. I do look forward to writing better
applications for windows xp 64 bit though.

The bottom line is that for now my PIII 600 mhz is able to read more 32 bit
integers per second than the AMD X2 3800+ which is kinda weird.

Actually I also tested before changing memory settings and running two
benchmarks... and then every instance achieves 1.5 million so that's about
the same as the PIII which also achieves 3 million.

Then I tried running two instances on the PIII and one instances achieved
2.4 million and the other one 0.6 million. The one that has focus runs
faster.
I thought I set foreground boost option disabled... but I probably enabled
it again to play games faster... so the foreground boost option is probably
enabled and at worked explaining this difference.

I dont know if windows xp 64 bit has a foreground boost option and if it's
enabled... that's something I might check out in the future ;)

Also I just ran the test.. and in the beginning one test behaved like it
didn't want to run probably because the memory first needs to be locked
etc.. all in all quite a fuzzy test. Another explanation might be that the
high performance timer has problems but I don't think so because I tested
that with two other applications.. so why one benchmark doesn't want to run
is strange... I place my bet on the locking of the memory failing or
something. I could look into that later or maybe not all because in general
I don't use that feature in my other applications... I only used that
feature for the benchmark.

My next test will probably be testing the speed of the harddisks now that
will be very interesting.

All in all I can only recommend software developers to start developing for
windows xp 64 bit or windows vista as soon as possible...

If you still developing software on windows xp 32 bit... you might be faced
with your software not running on 64 bit software... which does seem to be
the feature...

Also be sure to have to bootable operating systems... windows xp 32 bit is
still necessary to use software which won't run on windows xp 64 bit... like
dos software... unless you wanna boot into ms-dos... etc...

Fortunatelly for me I have two development computers.. one xp 32 bit and one
xp 64 bit ;) I could even boot into ms-dos or windows 98 from a harddrive...
yeah my old pc has multi boot... my new pc doesn't though... I never used
multi boot on my old pc that's why... yeah I wanna keep this new pc 64 bit
completely I guess... just to be ready for the future ;) or maybe I will get
to me and then I decided to still install xp 32 bit... but then I need to
screw around with sata drivers.. I dont wanna do that... and I dont want to
download zillion-1 xp 32 bit patches if you know what I mean... so screw
that ;) yeaaaah =D
Besides I am not sure if system commander 7 or 8 will work with windows xp
64 bit... it probably should work since the file systems are still more or
less the same <- ntfs 64 bit.

All in all microsoft and the hardware world have made quite a mess of things
and breaking quite a lot of things... really funny =D

Though as other people have written.. using windows xp 64 bit has a nice
feel to it... it's just like xp 32 bit... with some minor improvements...
like a little bit better internet explorer...

Bye,
Skybuck ;)
 
Well as promised I also just tested the harddisk performance and also that
is throwing dirt in my eyes ;)

First I created two test images of 5 gigabyte with filedisk tool one on C:
and one D:

First image:

Then I started reading from it... the speed was 530 MB/sec.

Then I started writing to it.. the speed was 400 MB/sec.

Second image:

This time I reversed the order.

I first write to it... the speed was only 2 MB/sec to maybe 3 MB/sec

Then I read from it... the speed started slowly dont remember what... then
it proceeded to 10 MB/sec then 100 MB/sec then I stopped it... finally I
started it again it went on with 200 MB/sec and increased it's speed until
it finally again reached 500 MB/sec.

There is 4 GB of ram in the system... I also checked the performance
monitor... the kernel seems to be using 750 MB of ram or so...

So I can come really fast to the following conclusion:

Microsoft's windows xp 64 bit simply caches a whole lot of the file... maybe
like a 500 MB cache or so...

The read speed is probably coming directly from RAM buffers instead of the
harddisk.. which only has 16 MB of hd cache etc...

So the operating system manages to throw quite some dirt in my eyes ;)

Ofcourse the second test was more interesting... since it's speed was only 2
MB/sec to maybe 3 MB/sec or so...

Also after stopping the write test... when it reaches 500 MB/sec the
operating system simply continued writing to the harddisk... probably still
busy writing caches to the harddisk.. it took a few seconds before the hd
led shut down.

All in all testing like this isn't really a reliable way of testing the true
speed of harddisks.. ofcourse you and me probably already know this... still
it's interesting to see what happens of my new dream pc ;) Windows blunty
caches it 500 MB or so... no idea what algorithms are behind it or what
percentage etc... all in all seeing 500 MB/sec read/write speed is quite
statisfieing even if it isn't the true harddisk performance...

Things will get more interesting when transferring new data via the network
from another computer... it will still perform some write caching I guess...

Oh well so much for trying to establish a write disk performance true
maximum speed etc..

I'll have to find some other more advanced tools which can truely test the
speed of the harddisk.. hopefully without destroying my harddisk ;)

Bye,
Skybuck.
 
Double sided is a bad idea.

Supported by motherboard doesn't specify how they'll
perform. As you can see, they do in fact run, just slow as
shit.

Welcome to the world of 2T performance and remember, if
people weren't so stupid you could at least have had the
benefit of ECC, if not registered memory. Yep, there was a
reason for all that stuff.
Click to expand...

I hope you dont base your conclusion on my 2.5 MB/sec figure... because
that's not a bandwidth test... that's just my wacky memory access time
benchmark test ;) access time test is not a bandwidth test :)

Rest assured... even my harddisk benchmark programs achieve 530 MB/sec and
430 MB/sec which are probably reading and writing from windows xp 64 bit
ram/disk caches etc...

I wouldn't be surprised at all if my dream pc achieves gigabytes/sec with a
true memory bandwidth test..

However real world performance of my own application is what is more
interesting to me...

Finally your remark about slow as shit is probably the state of the
industry... <- not being able to make transistors switch any faster... sure
you can add more transistor to make more memory.. sure you can add more
lanes for more bandwidth... but when it comes to executing commands and
processing power... the transistors dont switch faster... hence the dual
core... multi core processors... <- industry hoping to get more speed by
doing this in parallism...

If parallism is the way to go then the whole pc architecture will have to be
redesigned... in other words take a look at cell processor... each core will
have to be connected to it's own memory... otherwise bottleneck at memory
possible... etc etc etc.

It's quite interesting what microsoft is going to do... I'll bet you 10
bucks microsoft might be bankrupt in 5, 10, 20 years down the road... I
surely hope not... but though times are ahead ohhhh yes sir bob ;)

Windows vista is probably still based on old pc architecture however it can
do somet multi threading as well..

It will be interesting to see if microsoft's operating system will be able
to adept to all this parallism stuff :D

Some applications probably won't even become faster because of parallism...
but those applications probably don't need to become any faster anyway..
hence no more reason to sell new ms word copies etc <- see dead milk cow etc
etc etc.

Quite frankly... most people probably don't need a new computer.. for
stuff... expect slowass flash media player in internet explorer.

But then again.. hardware interface change... so in the end people will
still have to buy new computers etc...

So the only thing the industry has to do to keep people buying new stuff..
is simply changing, changing, changing all the current stuff without
actually improving it or making it any faster... as long as you can keep
people believing that it's faster or even slighty faster... and people buy
in to it the industry is a little bit safe =D hehehehe. That's the sad part
of it... change only to keep afloat ;)

Bye,
Skybuck :)
 
"Skybuck said:
Paul said:
"Skybuck said:
Hello,

I bought four of these memory chips:

Corsair TWINX2048-3200C2 DDR SDRAM, 1 GB, PC3200, 400 MHz (Double Side,
Double Data Rate, Supported by motherboard).

I also received a green card with instructions on it which shows how to
install the memory chips into the motherboard's memory slots.

Today I ran my own (32 bit) memory benchmark (on windows xp 64 bit) which
reads as many integers from the memory as possible. The results were
quite
shocking... only 2.4 million bytes per second. Even my 6 year old pentium
III 450 mhz overclocked to 600 mhz is able to achieve 3.1 million bytes
per
second.

My first suspect for the poor memory performance/access time is the
memory
chips themselfes and there settings.

The support on the website of Corsair is quite bad. There are no settings
to
be found anywhere. Their specification is mostly a marketing ad. That's
not
a specification !

I would like to know all the settings about these memory chips !

Is there a real/full specification of these memory chips and if so were
can
I download/view it ?

Failing that... does anybody know the correct settings for these memory
chips in the bios of the motherboard asus a8n32-sli ?

Which of these settings have to be changed and what would be there values
?:

Memclock mode: [Auto]
MCT Timing mode: [Auto]
Cas latency (cl): [Auto]
TRAS: [Auto]
TRP: [Auto]
TRCD: [Auto]
TRRD: [Auto]
TRC: [Auto]
TRFC: [Auto]
TRWT: [Auto]
User Config Mode: [Auto]
Read Preamble: 9.5 ns (?)
Asyc Latency: 11.0 ns (?)
Bank interleaving: [Auto]
Burst Length: [4 Beats]
Hardware Memory Hole: [Enabled] <- I changed that for 4 GB support.

MCT Extra Timing Mode:
TREF: 7.8 us
TWCL: 1
R/W Queue Bypass Count: 8
ByPass Max: 4
Idle Cycle Limit: 16
Dynamic Idle Cycle Center: Enabled
DDR Driving Strength: Normal
Enable 32-byte granularity: Disabled (what is this ?)
Twr: 3 clk
DDR Input Strobe Skew: Disabled
DDR Data Driving Strength: Disabled

ECC Configuration
Dram ECC Enable: Disabled
L2 Cache BG Scrub: Disabled
Data Cache BG Scrub: Disabled

DDR Vcore: Auto
DDR Clock Skew: Auto

(Message also posted on Corsair support forum)

Bye,
Skybuck.
Click to expand...

The datasheets are on the XMS memory web page. They aren't hard to find.

http://www.corsairmicro.com/corsair/products/specs/twinx2048-3200c2.pdf

"Tested at aggressive latency settings of 2-3-3-6 for all platforms.
2-3-3-6 on Intel and AMD based systems
Test voltage: 2.75V"

Slide 24 here, tells you the order of parameters:

http://corsairmicro.com/corsair/products/tech/memory_basics/153707/index.html

"CAS Latency, tRCD, tRP, tRAS"

When using four sticks of memory, the expected settings would be
DDR400 Command Rate 2T, which is better than the alternative
setting of DDR333 Command Rate 1T. You cannot expect to run
DDR400 Command Rate 1T with four memory sticks.

There really isn't much point in adjusting other parameters
on that memory page.

As for your little test program, try the following. Download
memtest86+ from www.memtest.org . The program will format a
testing floppy for you. Boot your new computer with the boot
floppy disk. When memtest86+ starts, there are three bandwidth
indicators in the upper left hand corner. The third one down,
monitors memory bandwidth (unbuffered). Specifically, what the
code does, is it flushes the cache before sequentially reading
a large chunk of memory.

Pop that test floppy into your old system, then into your
new system, and you should see an appropriate increase in
performance. The AMD memory controller does a pretty good
job with DDR.

The memtest86+ source code is available for download. I've
downloaded the source, and made a minor modification to the
program, so I could test memory bandwidth at different addresses
within system memory. I've had a look at the code and
their approach seems reasonable. Basically flush the cache,
and then read a block of memory larger than any cache, a
number of times, and average the results.

Are you certain that the time base measurement code in
your test program, is able to deal with the faster operation
of your new computer ? There are many ways to explain failures
of your test program.

The top post here, lists numbers like 368MB/sec for some PC133
SDRAM.

http://groups.google.ca/group/alt.comp.periphs.mainboard.asus/browse_frm/thread/249aeea20d9a440d

There is a picture of a memtest86+ screen here, for a 3500+
with 2x1GB memory. It lists 3908MB/sec, or roughly 10x
the memory bandwidth of the other system. This 3500+ is running
at 2700MHz and is overclocked. The memory is also overclocked
big time. I expect your system will be a bit lower
than that.

http://www.abxzone.com/forums/showthread.php?p=1238439
Click to expand...

Hello,

Your story is consistent with what I tried.

I wasn't sure if command rate t1 would work so I tried it =D then I got a
little worried. The computer completely stopped responding, no screen, no
boot screen, no beep, no nothing, I couldn't even turn it off or reboot.
Finally I had to turn off/on the computer via the power switch.

Fortunately for me the asus motherboard and american trends bios is able to
detect this overclock failure and reports it as such.

Then I tried command rate T2 and this time it worked.

I also tested memtest 1.65. Since the rawrite.exe utility does not work in
windows xp 64 bit, I installed it as follows. Fortunately for me I still had
a very cool ms-dos bootdisk... First I thought about using the windows 98
startup disk since it was the first disk in the box and had to make a quick
decision... but I already knew that worked... fortunately for me the boot
time is 10 seconds so I could still switch the boot disk... I wondered if
ms-dos 6.22 boot disk/os would still work on my new dream pc... so I put it
in... and haven't used that for a few years or so. Fortunately for me
everything worked perfectly... It worked even better than I ever imagined...
when I made the bootdisk I was even smart enough to install a microsoft ram
drive =D So it has default 4 MB of ram drive space =D and even qedit ;) but
ok.

So I was able to simply copy rawrite.exe and other files to the ramdisk...
since I wasn't sure if starting from a floppy would work... maybe that would
work as well... anyway... so finally I made the new disk. First I wanted to
make it on my old pc... but it doesn't have a floppy drive anymore lol... so
the point of the story is... where there is a will there is a way ;)
Ofcourse I first tried the iso image.. but that didn't work... rar and
winimage showed an empty iso image.

Now back to the memory/benchmarking test etc.

First I ran the memory testing before changing any memory settings.
Everything worked ok.

The speed reported was:

16 GB/sec (I think it was called L1 cache)
3.5 GB/sec(I think it was called L2 cache)
2 GB/sec (I think this was main memory or controller or something)

Then I changed the default memory settings to command rate t2 and 2-3-3-6
and ran the test again.

Now it reported this:

16 GB/sec
4 GB/sec
2 GB/sec

Finally I also run my own 32 bit benchmark and it reported still a somewhat
disappointing: 2.5 million bytes / sec

It's still 100.000 bytes faster than with the old memory settings so the new
memory settings does seem to help a bit... but only slighty...

So all in all the new memory settings aren't really worth bothering with..
or taking any corruption risks... but for now I will simply keep these
settings for the coming weeks/month/years to see what happens etc... It will
probably run just fine... so no need to run with any slower settings... but
all in all it wouldn't matter much.

Also the difference in the L2 cache is kinda weird... where is the L2 cache
located ? Is it in the memory chips or is it on the cpu or somewhere in
between ? It could just be a random measurement glitch/curiosity ;)

Finally I also noticed a little bit more of freezing when running both
benchmarks and other programs... when trying to move the windows around...

And finally when I ran two benchmarks, one benchmark instance didn't run at
all.. .which is really weird... my benchmark is mostly about testing how
many integers can be read randomly from memory and there is maybe also some
writing involved... not sure how the code gets compiled... so it's more
about access time than bandwidth. I have seen some very strange readings...
once I saw a reading of 500 MB/sec... now I tested another application
actually running two instances and now it only achieved 5 MB/sec through
loopback adapter though/winsock etc.. it's kinda strange.. since the same
application reported 500 MB/sec when testing with two computers... maybe I
am being delirius ;) but I don't think so ;)

I have seen some weird bugs on windows xp 64 bit with 32 bit applications,
also dos programs won't work etc. I do look forward to writing better
applications for windows xp 64 bit though.

The bottom line is that for now my PIII 600 mhz is able to read more 32 bit
integers per second than the AMD X2 3800+ which is kinda weird.

Actually I also tested before changing memory settings and running two
benchmarks... and then every instance achieves 1.5 million so that's about
the same as the PIII which also achieves 3 million.

Then I tried running two instances on the PIII and one instances achieved
2.4 million and the other one 0.6 million. The one that has focus runs
faster.
I thought I set foreground boost option disabled... but I probably enabled
it again to play games faster... so the foreground boost option is probably
enabled and at worked explaining this difference.

I dont know if windows xp 64 bit has a foreground boost option and if it's
enabled... that's something I might check out in the future ;)

Also I just ran the test.. and in the beginning one test behaved like it
didn't want to run probably because the memory first needs to be locked
etc.. all in all quite a fuzzy test. Another explanation might be that the
high performance timer has problems but I don't think so because I tested
that with two other applications.. so why one benchmark doesn't want to run
is strange... I place my bet on the locking of the memory failing or
something. I could look into that later or maybe not all because in general
I don't use that feature in my other applications... I only used that
feature for the benchmark.

My next test will probably be testing the speed of the harddisks now that
will be very interesting.

All in all I can only recommend software developers to start developing for
windows xp 64 bit or windows vista as soon as possible...

If you still developing software on windows xp 32 bit... you might be faced
with your software not running on 64 bit software... which does seem to be
the feature...

Also be sure to have to bootable operating systems... windows xp 32 bit is
still necessary to use software which won't run on windows xp 64 bit... like
dos software... unless you wanna boot into ms-dos... etc...

Fortunatelly for me I have two development computers.. one xp 32 bit and one
xp 64 bit ;) I could even boot into ms-dos or windows 98 from a harddrive...
yeah my old pc has multi boot... my new pc doesn't though... I never used
multi boot on my old pc that's why... yeah I wanna keep this new pc 64 bit
completely I guess... just to be ready for the future ;) or maybe I will get
to me and then I decided to still install xp 32 bit... but then I need to
screw around with sata drivers.. I dont wanna do that... and I dont want to
download zillion-1 xp 32 bit patches if you know what I mean... so screw
that ;) yeaaaah =D
Besides I am not sure if system commander 7 or 8 will work with windows xp
64 bit... it probably should work since the file systems are still more or
less the same <- ntfs 64 bit.

All in all microsoft and the hardware world have made quite a mess of things
and breaking quite a lot of things... really funny =D

Though as other people have written.. using windows xp 64 bit has a nice
feel to it... it's just like xp 32 bit... with some minor improvements...
like a little bit better internet explorer...

Bye,
Skybuck ;)
Click to expand...

The memtest results listed here, is ~2200MB/sec for dual channel
on Athlon64, with a more normal memory speed.

http://www.anandtech.com/mb/showdoc.aspx?i=2337&p=5

The L2 cache is inside the processor, and if the L2 speed increases,
it means memtest86 is not testing the L2 properly.

Based on the description of your test program, I have two observations:

1) "only achieved 5 MB/sec through loopback adapter though/winsock"

If your code accesses a network stack in the OS, this is
hardly measuring a performance primitive on the hardware.
You are testing the efficiency of some network code, an
entirely different thing.

2) "my benchmark is mostly about testing how many integers can be
read randomly from memory"

The memory controller is optimized for fetching bursts of data
at a time, and reading 32 bit integers at random would be a
cache buster. What your benchmark would be testing, it the
transaction rate of the memory controller, rather than the
bandwidth. But that should still be faster than the
transaction rate of the PC133 SDRAM.

This GIF is for SDRAM, but the principles are the same for
DDR. The transaction time for a burst, is from ACTV to ACTV.

http://www.dewassoc.com/performance/memory/images/sdram_burst.gif

You are only getting one 32 bit integer, after your computer
has done all that work. The computer may choose to put that
data into cache, but since your code is a cache buster, the
effort is wasted. The inverse of the ACTV-ACTV time, gives
the transaction rate, which will be a lot lower than a program
that takes advantage of the burst access pattern when testing
memory access.

Memtest86+ does sequential access, so every byte of the data
burst is used, and is counted towards the memory bandwidth
value in megabytes/second.

Random access bandwidth is a different measure than
sequential access bandwidth. The measurement still has
value, since some algorithms have that kind of access pattern,
so some programs have no choice but to do it that way. Basically,
all you end up doing, is comparing the access time of the various
memory technologies. Memory access has not improved at
the same rate as CPU core speed, so random access
bandwidth measure would not be a very popular benchmark
for your average computer enthusiast.

Paul
 
Skybuck Flying said:
Hello,

I bought four of these memory chips:

Corsair TWINX2048-3200C2 DDR SDRAM, 1 GB, PC3200, 400 MHz (Double Side,
Double Data Rate, Supported by motherboard).

I also received a green card with instructions on it which shows how to
install the memory chips into the motherboard's memory slots.

Today I ran my own (32 bit) memory benchmark (on windows xp 64 bit) which
reads as many integers from the memory as possible. The results were quite
shocking... only 2.4 million bytes per second. Even my 6 year old pentium
III 450 mhz overclocked to 600 mhz is able to achieve 3.1 million bytes per
second.
Click to expand...

Are you still having problems?

FOR GOD SAKE take that box back to your dealer and get them to put it
together properly and stop messing around with it. You don't have a clue
what you are doing.
 
Craig Sutton said:
Are you still having problems?

FOR GOD SAKE take that box back to your dealer and get them to put it
together properly and stop messing around with it. You don't have a clue
what you are doing.
Click to expand...

Lol you too funny.

I am testing and taking my computer to the extreme to find any possible
failures before the warranty ends ;)

Also to learn the limits of my computer ;)

Bye,
Skybuck.
 
Skybuck Flying said:
Lol you too funny.

I am testing and taking my computer to the extreme to find any possible
failures before the warranty ends ;)
Click to expand...

Failure found - loose nut in front of the keyboard

Also to learn the limits of my computer ;)
Click to expand...

You have reached YOUR limits long before the rig did.
 
Well I just spent two days creating/inventing ;) a 64 bit random function.

Now that it's complete it turns out the CPU is a ****ing bottleneck beyong
my wildest dreams ;)

Or maybe it's windows xp 64 bit, 32 bit to wow to 64 bit translation... and
delphi's code generation.

Using my own 64 bit random funtion only creates 2.8 MB/sec of bandwidth
between CPU and Memory... etc.

Something is definetly bottlenecking the CPU and/or the memory.

The weird thing is... as soon as I switch to another random function the
speed increases to 90 MB/sec.

The random function is still the bottleneck here.

The harddisk might be able to achieve 500 MB/sec read and write speed.. .it
might be that the CPU is the real bottleneck.

If it has to process the data... you can forget about that ! Speeds will
drop !

Bye,
Skybuck.
 
Something is definetly bottlenecking the CPU and/or the memory.

The weird thing is... as soon as I switch to another random function the
speed increases to 90 MB/sec.

The random function is still the bottleneck here.

The harddisk might be able to achieve 500 MB/sec read and write speed.. ..it
might be that the CPU is the real bottleneck.
Click to expand...

LOL you still have no idea at all , gee I would love that harddisk that does
500 MB/Sec......
 
Lol you too funny.

I am testing and taking my computer to the extreme to find any possible
failures before the warranty ends ;)

Also to learn the limits of my computer ;)
Click to expand...

The only failure is situated between a pair of ears very near your
computer. Order a replacement immediately.

OR, more worthwhile ......

( You are running a dual-core processor aren't you ? If only
single-core ignore the Set Affinity exercises below, but leave the
other settings as recommended )

Run 4 instances of Prime 95 <<SIMULTANEOUSLY>>,
as follows:-
1. Set Affinity CPU0, default torture test #1 ( min memory .. actually
the only memory that will be used will be within the CPU0 cache)
2. Set Affinity CPU1, default torture test #1
3. Set Affinity CPU0+1, default torture test #3 ( max memory)
4. Set Affinity CPU0+1, default torture test #3 ( max. memory)
The last 2 instances should saturate all of your memory. If it does
not just add another instance of Prime 95.

(A simple exercise for the student to configure Prime95 for the
appropriate Set Affinity and for running 4 instances simultaneously. )

Run the above test for at least 10 hours continuously with your final
BIOS settings for CPU and memory. You must end up with ZERO errors. If
not, diagnose from the test that failed and make the appropriate CPU
clock or memory-timing adjustments. BTW, with 4 sticks of memory, if
you do not set the memory command-rate to 2T, expect a bunch of errors
in Prime95 instances 3 and/or 4.

John Lewis
 
John Lewis said:
The only failure is situated between a pair of ears very near your
computer. Order a replacement immediately.

OR, more worthwhile ......

( You are running a dual-core processor aren't you ? If only
single-core ignore the Set Affinity exercises below, but leave the
other settings as recommended )

Run 4 instances of Prime 95 <<SIMULTANEOUSLY>>,
as follows:-
1. Set Affinity CPU0, default torture test #1 ( min memory .. actually
the only memory that will be used will be within the CPU0 cache)
2. Set Affinity CPU1, default torture test #1
3. Set Affinity CPU0+1, default torture test #3 ( max memory)
4. Set Affinity CPU0+1, default torture test #3 ( max. memory)
The last 2 instances should saturate all of your memory. If it does
not just add another instance of Prime 95.

(A simple exercise for the student to configure Prime95 for the
appropriate Set Affinity and for running 4 instances simultaneously. )

Run the above test for at least 10 hours continuously with your final
BIOS settings for CPU and memory. You must end up with ZERO errors. If
not, diagnose from the test that failed and make the appropriate CPU
clock or memory-timing adjustments. BTW, with 4 sticks of memory, if
you do not set the memory command-rate to 2T, expect a bunch of errors
in Prime95 instances 3 and/or 4.

John Lewis
Click to expand...

and if you get ANY errors, don't come back to tell us until you fix them.
 
and if you get ANY errors, don't come back to tell us until you fix them.
Click to expand...

Yep, we don't need to have a guided walkthrough. Life is too short,
and net-bandwidth is still not a free commodity.

John Lewis
 
Craig Sutton said:
LOL you still have no idea at all , gee I would love that harddisk that
does
500 MB/Sec......
Click to expand...

Then go buy it.

Google for Hitachi's 500 GB harddisk and read the fricking specs or Google
for Skybuck's dream pc and proceed from there ;)

Bye,
Skybuck.
 
Skybuck Flying said:
Then go buy it.

Google for Hitachi's 500 GB harddisk and read the fricking specs or Google
for Skybuck's dream pc and proceed from there ;)
Click to expand...

Man you are hilarious, have a look at some figures here
http://www.storagereview.com/articles/200511/HDS725050KLA360_2.html

Even 2 Raptor 10000 rpm HD's in raid 0 will only average a read speed of
around 118 MB/s with a burst speed of around 230.

You also might like to tell us how you are bypassing the SATA as it has a
limit of 150MB/s /300MB/s

How about you use HDTACH and post some proper figures?
 
How about you use HDTACH and post some proper figures?
Click to expand...

Because he obviously does not have a clue about building or testing
PCs. Pompous (and probably pampered by mama) ass.

John Lewis
 
Because he obviously does not have a clue about building or testing
PCs. Pompous (and probably pampered by mama) ass.

John Lewis
Click to expand...

What is so sad about this is he loses no opportunity to let everyone here know
what a nitwit he is - including his cross-poasting to ...pc-homebuilt - which
only a tyro would think means assembling a system from store-bought chunks...
 
Back
Top