Need quick help - Upgraded A7V8X-X with Sempron 2400+!

[ohaya]

Hi,

We decided to upgrade my daughter's PC from an Athlon 1.2 to a Sempron
2400+ this weekend.

I thought it'd be an easy upgrade, but it's been a bad move so far ...

She has an Asus A7V8X-X motherboard, Windows XP Home on the machine,
with a single 512MB stick of PC2700 (SimpleTech) memory, and before
attempting the upgrade, I checked the specs at Asus' website, etc. and
confirmed that the motherboard could support the Sempron if I upgraded
the BIOS to 1013.

I upgraded the BIOS before replacing the chip, then replaced the CPU and
heatsink fan. Then, I reset the BIOS defaults, and rebooted the system
and everything was ok. CPU was detected as a Sempron 2400+.

Then, I tried to boot to XP, and it wouldn't finish the boot, but kept
rebooting.

So, I booted to a floppy with DocMem (a memory tester), and ran that,
and almost immediately got errors.

I downclocked the system in BIOS to 1.337 GHz and set the FSB to 266,
and am re-running the memory test, and that seems to be going ok now.

Can anyone suggest what the problem might be?

I kind of have the feeling that the CPU/motherboard should be able to
run at "normal" specs (1.66GHz/333 FSB), but that the memory might not
be "up to speed", even though it's suppose to be PC2700.

She needs to get back to her campus this weekend, so I need to get this
system back up and running, and I guess I'll keep it downclocked if that
makes it work, for now.

Any ideas?

Thanks,
Jim

 

[Ben Pope]

Hi,

We decided to upgrade my daughter's PC from an Athlon 1.2 to a Sempron
2400+ this weekend.

I thought it'd be an easy upgrade, but it's been a bad move so far ...

She has an Asus A7V8X-X motherboard, Windows XP Home on the machine,
with a single 512MB stick of PC2700 (SimpleTech) memory, and before
attempting the upgrade, I checked the specs at Asus' website, etc. and
confirmed that the motherboard could support the Sempron if I upgraded
the BIOS to 1013.

I upgraded the BIOS before replacing the chip, then replaced the CPU and
heatsink fan. Then, I reset the BIOS defaults, and rebooted the system
and everything was ok. CPU was detected as a Sempron 2400+.

Then, I tried to boot to XP, and it wouldn't finish the boot, but kept
rebooting.

So, I booted to a floppy with DocMem (a memory tester), and ran that,
and almost immediately got errors.

I downclocked the system in BIOS to 1.337 GHz and set the FSB to 266,
and am re-running the memory test, and that seems to be going ok now.

Can anyone suggest what the problem might be?

It's probably a memory setting. See if you can find out what the
timings should be, and then manually to those timings.

A single stick should run at 1T command rate, but it depends on the
configuration of the DIMM. Set it to 2T for testing.

I kind of have the feeling that the CPU/motherboard should be able to
run at "normal" specs (1.66GHz/333 FSB), but that the memory might not
be "up to speed", even though it's suppose to be PC2700.

Shove it some more voltage if it's still not playing nice.

She needs to get back to her campus this weekend, so I need to get this
system back up and running, and I guess I'll keep it downclocked if that
makes it work, for now.

Indeed.

But if you can relax the memory timings a bit, and get away with the
correct FSB, you'd be much better off.

Ben

 

[ohaya]

Ben,

Thanks for the reply.

I was thinking about that (timing), but if the BIOS is set to use the
SPD, shouldn't that work, by itself?

BTW, I think that this is the memory:

SVM-DDR2700/512 :

http://www.simpletech.com/upgNav_ne...VM-DDR2700/512&std=1&showFrame=&ProfileID=330

And, I don't see any specs for timing. Is there some way to find out
what the memory says by reading the SPD?

Thanks,
Jim

 

[ohaya]

Ben,

Thanks for the reply.

I was thinking about that (timing), but if the BIOS is set to use the
SPD, shouldn't that work, by itself?

BTW, I think that this is the memory:

SVM-DDR2700/512 :

http://www.simpletech.com/upgNav_ne...VM-DDR2700/512&std=1&showFrame=&ProfileID=330

And, I don't see any specs for timing. Is there some way to find out
what the memory says by reading the SPD?

Thanks,
Jim

 

[Ben Pope]

Ben,

Thanks for the reply.

I was thinking about that (timing), but if the BIOS is set to use the
SPD, shouldn't that work, by itself?

Yes it *should*. But something appears to be going wrong...

BTW, I think that this is the memory:

SVM-DDR2700/512 :

http://www.simpletech.com/upgNav_ne...VM-DDR2700/512&std=1&showFrame=&ProfileID=330

And, I don't see any specs for timing. Is there some way to find out
what the memory says by reading the SPD?

Well that was my point... perhaps the SPD is programmed incorrectly, or
perhaps it is being read incorrectly. It has been known to happen, even
on well known brands.

Perhaps the RAM chips are visible? Is it possible to get a model off of
the chips themselves?

Ben

 

[ohaya]

Well that was my point... perhaps the SPD is programmed incorrectly, or
perhaps it is being read incorrectly. It has been known to happen, even
on well known brands.

Perhaps the RAM chips are visible? Is it possible to get a model off of
the chips themselves?

Ben,

Ok, thanks understood re. the SPD.

Well, it looks like I'm making some progress here.

First here are the options available in the BIOS for RAM timing and the
defaults I had originally (Set to Defaults):

SDRAM Configuration: by SPD
SDRAM CAS Latency: 2.5T (DDR)
SDRAM RAS to CAS Delay: 3T [options: 2T, 3T]
SDRAM RAS Precharge Delay: 3T [options: 2T, 3T]
SDRAM Active Precharge Delay 7T [options: 6T, 7T]
SDRAM 1T Command Control: Auto [options: Auto, Enabled, Disabled]

Memory is ok with the above settings if I set speed to 1.337GHz with FSB
266.

Memory gives error with the above settings if I set speed to 1.667GHz
with FSB 333.


I set the CPU speed settings back to 1.667 GHz with FSB 333, then tried
just disabling the "SDRAM 1T Command Control":

SDRAM Configuration: by SPD
SDRAM CAS Latency: 2.5T (DDR)
SDRAM RAS to CAS Delay: 3T
SDRAM RAS Precharge Delay: 3T
SDRAM Active Precharge Delay 7T
SDRAM 1T Command Control: Disabled

Memory test again failed.


I then tried setting the SDRAM CAS Latency to 3T:

SDRAM Configuration: User define
SDRAM CAS Latency: 3.0T
SDRAM RAS to CAS Delay: 3T
SDRAM RAS Precharge Delay: 3T
SDRAM Active Precharge Delay 7T
SDRAM 1T Command Control: Disabled

Memory test ran with NO errors!!


Finally, I then tried setting the SDRAM 1T Command Control back to Auto
and left the SDRAM CAS Latency at 3T:

SDRAM Configuration: User define
SDRAM CAS Latency: 3.0T
SDRAM RAS to CAS Delay: 3T
SDRAM RAS Precharge Delay: 3T
SDRAM Active Precharge Delay 7T
SDRAM 1T Command Control: Auto

Memory test ran with NO errors!!


So, at least between the CAS Latency and the 1T Comand Control, it looks
like changing the CAS Latency from 2.5T to 3.0T works.

At this point, I'm guessing that the last set of settings is about as
good as it is going to get.

Does that sound about right, or are there any other suggestions you, or
anyone else, might suggest?

Thanks,
Jim

 

[Paul]

Well that was my point... perhaps the SPD is programmed incorrectly, or
perhaps it is being read incorrectly. It has been known to happen, even
on well known brands.

Perhaps the RAM chips are visible? Is it possible to get a model off of
the chips themselves?

Ben,

Ok, thanks understood re. the SPD.

Well, it looks like I'm making some progress here.

First here are the options available in the BIOS for RAM timing and the
defaults I had originally (Set to Defaults):

SDRAM Configuration: by SPD
SDRAM CAS Latency: 2.5T (DDR)
SDRAM RAS to CAS Delay: 3T [options: 2T, 3T]
SDRAM RAS Precharge Delay: 3T [options: 2T, 3T]
SDRAM Active Precharge Delay 7T [options: 6T, 7T]
SDRAM 1T Command Control: Auto [options: Auto, Enabled, Disabled]

Memory is ok with the above settings if I set speed to 1.337GHz with FSB
266.

Memory gives error with the above settings if I set speed to 1.667GHz
with FSB 333.


I set the CPU speed settings back to 1.667 GHz with FSB 333, then tried
just disabling the "SDRAM 1T Command Control":

SDRAM Configuration: by SPD
SDRAM CAS Latency: 2.5T (DDR)
SDRAM RAS to CAS Delay: 3T
SDRAM RAS Precharge Delay: 3T
SDRAM Active Precharge Delay 7T
SDRAM 1T Command Control: Disabled

Memory test again failed.


I then tried setting the SDRAM CAS Latency to 3T:

SDRAM Configuration: User define
SDRAM CAS Latency: 3.0T
SDRAM RAS to CAS Delay: 3T
SDRAM RAS Precharge Delay: 3T
SDRAM Active Precharge Delay 7T
SDRAM 1T Command Control: Disabled

Memory test ran with NO errors!!


Finally, I then tried setting the SDRAM 1T Command Control back to Auto
and left the SDRAM CAS Latency at 3T:

SDRAM Configuration: User define
SDRAM CAS Latency: 3.0T
SDRAM RAS to CAS Delay: 3T
SDRAM RAS Precharge Delay: 3T
SDRAM Active Precharge Delay 7T
SDRAM 1T Command Control: Auto

Memory test ran with NO errors!!


So, at least between the CAS Latency and the 1T Comand Control, it looks
like changing the CAS Latency from 2.5T to 3.0T works.

At this point, I'm guessing that the last set of settings is about as
good as it is going to get.

Does that sound about right, or are there any other suggestions you, or
anyone else, might suggest?

Thanks,
Jim

That is good research work. It means the memory is really rated
at CAS3, at the current Vdimm voltage you are using.

It is probably too late now, but when you think you are finished
setting up the computer, get a copy of Prime95 (mersenne.org).
Prime95 runs the computer at 100%, and does a good job of
exercising the memory, Northbridge, and processor. It is a
better test than memtest86+, in the sense that it will uncover
faults that memtest86+ won't find. Prime95 should not report
any "roundoff errors", as that indicates the computer is still
not completely solid. Select "Torture Test" from the menu, to
test your system. If the computer passes Prime95, then it is
likely to be safe enough to do work on.

Your daughter could even run this while she is taking classes,
and then report back the results to you.

HTH,
Paul

 

[ohaya]

That is good research work. It means the memory is really rated
at CAS3, at the current Vdimm voltage you are using.

It is probably too late now, but when you think you are finished
setting up the computer, get a copy of Prime95 (mersenne.org).
Prime95 runs the computer at 100%, and does a good job of
exercising the memory, Northbridge, and processor. It is a
better test than memtest86+, in the sense that it will uncover
faults that memtest86+ won't find. Prime95 should not report
any "roundoff errors", as that indicates the computer is still
not completely solid. Select "Torture Test" from the menu, to
test your system. If the computer passes Prime95, then it is
likely to be safe enough to do work on.

Your daughter could even run this while she is taking classes,
and then report back the results to you.

Paul,

Yes, too late, for now at least. The system is back at school, but I'll
try to get her to bring it back next time and run Prime95.

I did run a several hours long burn-in test of DocMem, and no errors,
though.

I'm still kind of curious why I didn't have a problem before the CPU
switch. The BIOS was set to "by SPD" back then. I'm thinking maybe
this memory was just "marginally CL2.5"? And, going to the 333 MHz FSB
took it over the "marginal" edge?

Jim

 

[Ben Pope]

Paul,

Yes, too late, for now at least. The system is back at school, but I'll
try to get her to bring it back next time and run Prime95.

I did run a several hours long burn-in test of DocMem, and no errors,
though.

I'm still kind of curious why I didn't have a problem before the CPU
switch. The BIOS was set to "by SPD" back then. I'm thinking maybe
this memory was just "marginally CL2.5"? And, going to the 333 MHz FSB
took it over the "marginal" edge?

What CL means is Cas Latency, in clock ticks.

So CL2 at 133MHz is the same "delay" as CL3 at 200MHz.

Thus, increasing the speed from 133MHz to 166MHz shortens the real delay
from 2.5/133M to 2.5/166M.

Obviously 2.5/166M is too short a time, at the given voltage, but 3/166M
is fine. With 1 DIMM, a command rate of 1T is usually fine, but often 1
cheap DIMM can present the load of 2 better DIMMS, and require more
delay (or more Voltage).

You might find that by increasing voltage you can use CL2.5 at 166MHz,
but in all honesty, if it's stable where it is now, it's probably not
worth it. These setttings offer fairly marginal performance gains for
most tasks. So unless your daughter really needs the memory bandwidth,
I'd be quite content with it working at 98-100% of full speed.

Ben

 

[Paul]

What CL means is Cas Latency, in clock ticks.

So CL2 at 133MHz is the same "delay" as CL3 at 200MHz.

Thus, increasing the speed from 133MHz to 166MHz shortens the real delay
from 2.5/133M to 2.5/166M.

Obviously 2.5/166M is too short a time, at the given voltage, but 3/166M
is fine. With 1 DIMM, a command rate of 1T is usually fine, but often 1
cheap DIMM can present the load of 2 better DIMMS, and require more
delay (or more Voltage).

You might find that by increasing voltage you can use CL2.5 at 166MHz,
but in all honesty, if it's stable where it is now, it's probably not
worth it. These setttings offer fairly marginal performance gains for
most tasks. So unless your daughter really needs the memory bandwidth,
I'd be quite content with it working at 98-100% of full speed.

Ben

Absolutely. If CAS3 is working for you, leave it.

Memory basics - slide 21 - explains what CAS is.
http://corsairmicro.com/corsair/products/tech/memory_basics

If the clock is 200MHz, a single clock cycle lasts 5 nanoseconds.
If CAS is 2.5, you multiply the CAS number by the clock period,
to determine how much time the memory is getting to access the
column containing the data. 2.5 @ 133MHz = 18.75ns while
3.0 @ 166MHz = 18.0ns. Your memory is rated at 2.5 @ 166MHz,
which is 2.5 * 6ns or 15.0ns.

This just means your memory is not currently meeting its
stated rating. The memory should be able to dig up the data
in 15ns, but it looks more like 18 or so is needed. The simple
workaround, of increasing the CAS setting, allows the memory
a few more nanoseconds to find the data.

The command rate is a slightly different concept. It changes
the setup time, for information placed on the address/command
signals.

2T CMD Rate Command + Address signals X-----------------X
|<------>| Tsetup
___ ___ ___
Memory Clock _| |____| |____|
___________ ___
Strobe (wasted |________|
cycle)

1T CMD Rate Command + Address signals X--------X
|<---->| Tsetup
___ ___ ___
Memory Clock _| |____| |____|
___________ ___
Strobe usable |________|

When the Command Rate is 2T, the address is placed on the bus
for two clock cycles. The strobe is asserted during the second
cycle, which means the memory chips get a whole clock cycle to
consider the address or command placed on the address/command bus.

If the Command Rate is 1T, and the memory bus is heavily loaded,
it makes the address valid period shift to the right. This
causes the effective setup time Tsetup, to be less than a full
clock period. The memory must have a certain amount of setup time,
to be able to see the data.

That is why, on Athlon64 for example, a Command Rate of 2T is
required when two double sided sticks are installed per channel.
The address signal is arriving later, due to bus loading, and
the Command Rate 2T setting, uses two address cycles per interaction
with the memory. This kills the memory bandwidth performance, but
improves stability.

Command Rate 2T should not be required with a single stick of
RAM. But playing with the CAS setting, is valid for any number
of sticks of memory, as it is an issue internal to the operation
of the memory.

Paul

 

[ohaya]

Absolutely. If CAS3 is working for you, leave it.

Memory basics - slide 21 - explains what CAS is.
http://corsairmicro.com/corsair/products/tech/memory_basics

If the clock is 200MHz, a single clock cycle lasts 5 nanoseconds.
If CAS is 2.5, you multiply the CAS number by the clock period,
to determine how much time the memory is getting to access the
column containing the data. 2.5 @ 133MHz = 18.75ns while
3.0 @ 166MHz = 18.0ns. Your memory is rated at 2.5 @ 166MHz,
which is 2.5 * 6ns or 15.0ns.

This just means your memory is not currently meeting its
stated rating. The memory should be able to dig up the data
in 15ns, but it looks more like 18 or so is needed. The simple
workaround, of increasing the CAS setting, allows the memory
a few more nanoseconds to find the data.

The command rate is a slightly different concept. It changes
the setup time, for information placed on the address/command
signals.

2T CMD Rate Command + Address signals X-----------------X
|<------>| Tsetup
___ ___ ___
Memory Clock _| |____| |____|
___________ ___
Strobe (wasted |________|
cycle)

1T CMD Rate Command + Address signals X--------X
|<---->| Tsetup
___ ___ ___
Memory Clock _| |____| |____|
___________ ___
Strobe usable |________|

When the Command Rate is 2T, the address is placed on the bus
for two clock cycles. The strobe is asserted during the second
cycle, which means the memory chips get a whole clock cycle to
consider the address or command placed on the address/command bus.

If the Command Rate is 1T, and the memory bus is heavily loaded,
it makes the address valid period shift to the right. This
causes the effective setup time Tsetup, to be less than a full
clock period. The memory must have a certain amount of setup time,
to be able to see the data.

That is why, on Athlon64 for example, a Command Rate of 2T is
required when two double sided sticks are installed per channel.
The address signal is arriving later, due to bus loading, and
the Command Rate 2T setting, uses two address cycles per interaction
with the memory. This kills the memory bandwidth performance, but
improves stability.

Command Rate 2T should not be required with a single stick of
RAM. But playing with the CAS setting, is valid for any number
of sticks of memory, as it is an issue internal to the operation
of the memory.

Paul

Ben and Paul,

Thanks for all the explanation... Since she has the system at school
for the moment, and it seems to be running stably, I'll probably leave
it as is, with the CL3. I'm sure that she'll let me know if it starts
misbehaving. Typically this will be between 3:00AM-4:00AM !!

Jim

 

[Ben Pope]

I'm sure that she'll let me know if it starts
misbehaving. Typically this will be between 3:00AM-4:00AM !!

Yes... about 8 hours before the deadline... sounds about right

Ben

 

[ohaya]

Yes... about 8 hours before the deadline... sounds about right

Ben,

Hehe... Well, let's see.

On this one (the CPU upgrade), she got home for the weekend LAST
Thursday. Then, she decided she wanted to do the CPU upgrade at about
11:00 PM on SATURDAY NIGHT, but she had to be back at school early
Monday morning, at the latest.

Typical !!

BTW, in case I haven't mentioned your (and Paul's) help was invaluable!!

Jim

 

[Ben Pope]

Ben,

Hehe... Well, let's see.

On this one (the CPU upgrade), she got home for the weekend LAST
Thursday. Then, she decided she wanted to do the CPU upgrade at about
11:00 PM on SATURDAY NIGHT, but she had to be back at school early
Monday morning, at the latest.

Typical !!

I graduated from Uni coming up 2 years ago, and I think I left
everything until the last night - dunno how many times I pulled an
all-nighter to finish it. All part of the fun!

BTW, in case I haven't mentioned your (and Paul's) help was invaluable!!

No worries, nice when it comes together.

Ben