RAM on ASUS mobo

Paul wrote:
Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
Andy Burns wrote:
Ed Cryer wrote:

The four are all the same; all 1600MHz, DDR3, 8GB.

Well, theoretically they are fine for that MB.

https://www.corsair.com/uk/en/Categories/Products/Memory/Mainstream-Memory/Corsair-Memory-%E2%80%94-8GB-DDR3-Memory/p/CMV8GX3M1A1600C11


Try a pair of the new ones in the blue slots at the tested speed
of 1333MHzÂ*Â*Â* 9-9-9-24

if that works then all four at that speed.

if that works, try two and finally all four of them
at the XMS speed of 1600MHz 11-11-11-30

I've not used an AMD CPU for years, so I'll take Paul's word for it
that a quad SIMM configuration is tricky.Â* I have no issues with
4x8GB on "workstation" class intel motherboards.

Thanks for this. This sounds practical.
I shall try them, as you say.
Please explain "9-9-9-24"; and tell me how to set the speed.

Ed

9-9-9-24

tCAS tRCD tRP tRC

In roughly the same order as your DRAM customization screen.

The numbers are counts of "cycles".

At 1600MHz, the inverse is 0.6ns. 9 cycles at 0.6ns is 5.4ns.

Internally, all the RAM understands is nanoseconds.
"Did you meet my 5.4ns minimum ???" it asks.

If the RAM is set to DDR1333, the inverse of that is 0.75ns.
To make a number close to 5.4ns, we need about 7.2 cycles.
Rounding up, 8 cycles would work, without causing a smaller
number than the 5.4ns minimum.

9-9-9-24 DDR3-1600 or 8-8-8-X DDR3-1333

The timing table you can see in CPUZ shows
example timings that are recorded in the SPD
for this. Other values can be done by the BIOS
code by interpolation and so on.

Â*Â*Â* Paul

Here are the SPD tabs of CPU-Z for the two working modules.
Any comments?
https://www.dropbox.com/s/1d8v4h5mji...ts2-4.png?dl=0

Ed

Assuming those two samples are DIMMs from different
sets, they're very close to one another at the top setting.

You could start by using the frequency setting, and
let the BIOS do the math on the individual timings.

*******

Â*Â* AI Overclock Tuner [Manual]
Â*Â* Memory FrequencyÂ*Â* [DDR3-1600]Â* === Start with 1600, switch to 1333 if
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* there is trouble, let the BIOS
do the
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* math for the rest.
Â*Â* DRAM Timing ControlÂ*Â* [example at 1600]
Â*Â*Â*Â*Â*Â* tCASÂ*Â* Auto (11)
Â*Â*Â*Â*Â*Â* tRCDÂ*Â* Auto (11)
Â*Â*Â*Â*Â*Â* tRPÂ*Â*Â* Auto (11)
Â*Â*Â*Â*Â*Â* tRCÂ*Â*Â* Auto (41)

Â*Â*Â*Â*Â*Â* DRAM Command RateÂ* 2TÂ*Â*Â*Â* === definitely want 2T
Â*Â* DRAM VoltageÂ* [1.50V]Â*Â*Â*Â* === You should be able to use between
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* Â*Â*Â*Â*Â*Â*Â*Â* 1.5 and 1.65V without issue.
Â*Â* NB VoltageÂ*Â*Â* [Auto]Â*Â*Â*Â*Â* === Leave this for now

If that doesn't work:Â* 11-12-11-41

Â*Â*Â*Â*Â*Â* tRCDÂ*Â* 12
Â*Â*Â*Â*Â*Â* DRAM Voltage [1.65V]

If that doesn't work

Â*Â*Â*Â*Â*Â* tCASÂ*Â* 12
Â*Â*Â*Â*Â*Â* tRCDÂ*Â* 12
Â*Â*Â*Â*Â*Â* tRPÂ*Â*Â* 12
Â*Â*Â*Â*Â*Â* tRCÂ*Â*Â* Auto === let BIOS work it out

Sometimes, the error rate and whether the errors are exactly
reproducible on each run, hints at whether the RAM chips
have an issue, or the bus timing is flaky, or the Vnb needs
to be adjusted up a notch or two. The Vnb regulators now
have fairly tiny step sizes and are capable of fine adjustment.

*******

While your board has an "auto overclock" kind of feature,
I don't like the voltages or extreme settings those tend to use.
Yes, if you jam the controls to max, you can practically
guarantee an overclock, and the methods they use on these
boards aren't exactly "subtle". Still, if you read an article
somewhere that says the method used is good, I won't argue
with that.

I try to sneak up on the settings if I can, and prefer
to not apply extreme voltages if I can avoid it. My previous
set of "hot DIMMs" had a problem, so I'm trying to not make
any more "hot DIMMs" on purpose by using extreme settings.

When the RAM on the machine started throwing errors a couple
of sets of RAM ago, I used Vnb as the only setting, to fix it.
Nothing else needed to be adjusted. I could do this though,
because I was working on a theory that the Northbridge was aging.
I knew the settings I'd used for everything else, were fine
at the time of the build.

Normally, the board will be selecting Command Rate 2T on its
own, whether because the DIMM clock is rather high, or
because the bus is loaded (four DIMMs). There's hardly a chance
to realistically use 1T (maybe a single set of single sided
4GB DIMMs).

Maybe you'll have more joy at 1333, with a full set.

Some older AMD setups used to work that way, DDR400
with two DIMMs, DDR333 with four DIMMs.

HTH,
Â*Â*Â* Paul

This looks brilliant to me, Paul, and I thank you heartily.
It scares me a bit, though. I've never done memory-module clocking and
setting before; and when I went into the BIOS of this new machine it
presented me with a screen that looked like something from a flight-deck.
UEFI; so I clicked on "exit advanced" and it exited into booting.

I have some learning to do here, including the pages from the mobo
manual that Andy has shown me.
I'll get round to it.

Ed

Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
Andy Burns wrote:
Ed Cryer wrote:

The four are all the same; all 1600MHz, DDR3, 8GB.

Well, theoretically they are fine for that MB.

https://www.corsair.com/uk/en/Categories/Products/Memory/Mainstream-Memory/Corsair-Memory-%E2%80%94-8GB-DDR3-Memory/p/CMV8GX3M1A1600C11


Try a pair of the new ones in the blue slots at the tested speed
of 1333MHz 9-9-9-24

if that works then all four at that speed.

if that works, try two and finally all four of them
at the XMS speed of 1600MHz 11-11-11-30

I've not used an AMD CPU for years, so I'll take Paul's word for
it that a quad SIMM configuration is tricky. I have no issues
with 4x8GB on "workstation" class intel motherboards.

Thanks for this. This sounds practical.
I shall try them, as you say.
Please explain "9-9-9-24"; and tell me how to set the speed.

Ed

9-9-9-24

tCAS tRCD tRP tRC

In roughly the same order as your DRAM customization screen.

The numbers are counts of "cycles".

At 1600MHz, the inverse is 0.6ns. 9 cycles at 0.6ns is 5.4ns.

Internally, all the RAM understands is nanoseconds.
"Did you meet my 5.4ns minimum ???" it asks.

If the RAM is set to DDR1333, the inverse of that is 0.75ns.
To make a number close to 5.4ns, we need about 7.2 cycles.
Rounding up, 8 cycles would work, without causing a smaller
number than the 5.4ns minimum.

9-9-9-24 DDR3-1600 or 8-8-8-X DDR3-1333

The timing table you can see in CPUZ shows
example timings that are recorded in the SPD
for this. Other values can be done by the BIOS
code by interpolation and so on.

Paul

Here are the SPD tabs of CPU-Z for the two working modules.
Any comments?
https://www.dropbox.com/s/1d8v4h5mji...ts2-4.png?dl=0

Ed

Assuming those two samples are DIMMs from different
sets, they're very close to one another at the top setting.

You could start by using the frequency setting, and
let the BIOS do the math on the individual timings.

*******

AI Overclock Tuner [Manual]
Memory Frequency [DDR3-1600] === Start with 1600, switch to
1333 if
there is trouble, let the
BIOS do the
math for the rest.
DRAM Timing Control [example at 1600]
tCAS Auto (11)
tRCD Auto (11)
tRP Auto (11)
tRC Auto (41)

DRAM Command Rate 2T === definitely want 2T
DRAM Voltage [1.50V] === You should be able to use between
1.5 and 1.65V without issue.
NB Voltage [Auto] === Leave this for now

If that doesn't work: 11-12-11-41

tRCD 12
DRAM Voltage [1.65V]

If that doesn't work

tCAS 12
tRCD 12
tRP 12
tRC Auto === let BIOS work it out

Sometimes, the error rate and whether the errors are exactly
reproducible on each run, hints at whether the RAM chips
have an issue, or the bus timing is flaky, or the Vnb needs
to be adjusted up a notch or two. The Vnb regulators now
have fairly tiny step sizes and are capable of fine adjustment.

*******

While your board has an "auto overclock" kind of feature,
I don't like the voltages or extreme settings those tend to use.
Yes, if you jam the controls to max, you can practically
guarantee an overclock, and the methods they use on these
boards aren't exactly "subtle". Still, if you read an article
somewhere that says the method used is good, I won't argue
with that.

I try to sneak up on the settings if I can, and prefer
to not apply extreme voltages if I can avoid it. My previous
set of "hot DIMMs" had a problem, so I'm trying to not make
any more "hot DIMMs" on purpose by using extreme settings.

When the RAM on the machine started throwing errors a couple
of sets of RAM ago, I used Vnb as the only setting, to fix it.
Nothing else needed to be adjusted. I could do this though,
because I was working on a theory that the Northbridge was aging.
I knew the settings I'd used for everything else, were fine
at the time of the build.

Normally, the board will be selecting Command Rate 2T on its
own, whether because the DIMM clock is rather high, or
because the bus is loaded (four DIMMs). There's hardly a chance
to realistically use 1T (maybe a single set of single sided
4GB DIMMs).

Maybe you'll have more joy at 1333, with a full set.

Some older AMD setups used to work that way, DDR400
with two DIMMs, DDR333 with four DIMMs.

HTH,
Paul

This looks brilliant to me, Paul, and I thank you heartily.
It scares me a bit, though. I've never done memory-module clocking and
setting before; and when I went into the BIOS of this new machine it
presented me with a screen that looked like something from a flight-deck.
UEFI; so I clicked on "exit advanced" and it exited into booting.

I have some learning to do here, including the pages from the mobo
manual that Andy has shown me.
I'll get round to it.

Ed


There are sufficient "Autos" in there, you can work it a
bit at a time. For example, if you're timid, you can
just dial the DDR3-1600 setting to DDR3-1333 and see
if that's enough to get it working. Then you don't
have to look at all those pesky Tthis and Tthat ones.

And memtest86+ can give you a good idea how close you're
getting, or whether the situation is "hopeless". On my floppy-less
computer, I use a USB floppy drive for "tradition" and load
the memtest86+ floppy that way. (This would be with *no*
Windows hard drives connected, for safety from corruption.)

For example, my old Asrock motherboard, the documentation
(even the VIA website that made the chipset) said only
1GB DIMMs worked. Well, I plugged in a 2GB DIMM and
it was fully recognized. But, what was the problem ?
It was throwing errors.

Since the BIOS designers believed the documentation they were
given, they never "tuned" operation of the board for a 2GB stick.
There's a certain parameter that needs tuning in a lab,
and it would require me to do 256 test cases to "nail" the
timing (by trial and error). Well, that simply wasn't going
to happen. I don't have the stamina for that.

Your situation isn't nearly as bad, as the hardware is
mostly dialed in, and it could take as little as one
parameter change to tip it upright. Relaxing a single
timing, or adjusting the voltage within safe limits,
could be enough to make a difference.

I've spent as long as a week working on this stuff,
but I don't do that any more. Minimal change, test
for stability, and move on... :-)

It's one of the reasons I really like XMP. It does
everything for you, but only with two sticks. It's a bit
too hard for the manufacturer to test and guarantee
that four sticks will work in XMP setup mode, and the table
in the DIMM may not have the info needed. XMP is a section
of the SPD table, that was not designed by JEDEC. Some
"computer people" dreamed it up. If only the
process was rigorous enough to standardize...

Paul

Paul wrote:
Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
Andy Burns wrote:
Ed Cryer wrote:

The four are all the same; all 1600MHz, DDR3, 8GB.

Well, theoretically they are fine for that MB.

https://www.corsair.com/uk/en/Categories/Products/Memory/Mainstream-Memory/Corsair-Memory-%E2%80%94-8GB-DDR3-Memory/p/CMV8GX3M1A1600C11


Try a pair of the new ones in the blue slots at the tested speed
of 1333MHzÂ*Â*Â* 9-9-9-24

if that works then all four at that speed.

if that works, try two and finally all four of them
at the XMS speed of 1600MHz 11-11-11-30

I've not used an AMD CPU for years, so I'll take Paul's word for
it that a quad SIMM configuration is tricky.Â* I have no issues
with 4x8GB on "workstation" class intel motherboards.

Thanks for this. This sounds practical.
I shall try them, as you say.
Please explain "9-9-9-24"; and tell me how to set the speed.

Ed

9-9-9-24

tCAS tRCD tRP tRC

In roughly the same order as your DRAM customization screen.

The numbers are counts of "cycles".

At 1600MHz, the inverse is 0.6ns. 9 cycles at 0.6ns is 5.4ns.

Internally, all the RAM understands is nanoseconds.
"Did you meet my 5.4ns minimum ???" it asks.

If the RAM is set to DDR1333, the inverse of that is 0.75ns.
To make a number close to 5.4ns, we need about 7.2 cycles.
Rounding up, 8 cycles would work, without causing a smaller
number than the 5.4ns minimum.

9-9-9-24 DDR3-1600 or 8-8-8-X DDR3-1333

The timing table you can see in CPUZ shows
example timings that are recorded in the SPD
for this. Other values can be done by the BIOS
code by interpolation and so on.

Â*Â*Â* Paul

Here are the SPD tabs of CPU-Z for the two working modules.
Any comments?
https://www.dropbox.com/s/1d8v4h5mji...ts2-4.png?dl=0

Ed

Assuming those two samples are DIMMs from different
sets, they're very close to one another at the top setting.

You could start by using the frequency setting, and
let the BIOS do the math on the individual timings.

*******

Â*Â*Â* AI Overclock Tuner [Manual]
Â*Â*Â* Memory FrequencyÂ*Â* [DDR3-1600]Â* === Start with 1600, switch to
1333 if
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* there is trouble, let the
BIOS do the
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* math for the rest.
Â*Â*Â* DRAM Timing ControlÂ*Â* [example at 1600]
Â*Â*Â*Â*Â*Â*Â* tCASÂ*Â* Auto (11)
Â*Â*Â*Â*Â*Â*Â* tRCDÂ*Â* Auto (11)
Â*Â*Â*Â*Â*Â*Â* tRPÂ*Â*Â* Auto (11)
Â*Â*Â*Â*Â*Â*Â* tRCÂ*Â*Â* Auto (41)

Â*Â*Â*Â*Â*Â*Â* DRAM Command RateÂ* 2TÂ*Â*Â*Â* === definitely want 2T
Â*Â*Â* DRAM VoltageÂ* [1.50V]Â*Â*Â*Â* === You should be able to use between
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* Â*Â*Â*Â*Â*Â*Â*Â*Â* 1.5 and 1.65V without issue.
Â*Â*Â* NB VoltageÂ*Â*Â* [Auto]Â*Â*Â*Â*Â* === Leave this for now

If that doesn't work:Â* 11-12-11-41

Â*Â*Â*Â*Â*Â*Â* tRCDÂ*Â* 12
Â*Â*Â*Â*Â*Â*Â* DRAM Voltage [1.65V]

If that doesn't work

Â*Â*Â*Â*Â*Â*Â* tCASÂ*Â* 12
Â*Â*Â*Â*Â*Â*Â* tRCDÂ*Â* 12
Â*Â*Â*Â*Â*Â*Â* tRPÂ*Â*Â* 12
Â*Â*Â*Â*Â*Â*Â* tRCÂ*Â*Â* Auto === let BIOS work it out

Sometimes, the error rate and whether the errors are exactly
reproducible on each run, hints at whether the RAM chips
have an issue, or the bus timing is flaky, or the Vnb needs
to be adjusted up a notch or two. The Vnb regulators now
have fairly tiny step sizes and are capable of fine adjustment.

*******

While your board has an "auto overclock" kind of feature,
I don't like the voltages or extreme settings those tend to use.
Yes, if you jam the controls to max, you can practically
guarantee an overclock, and the methods they use on these
boards aren't exactly "subtle". Still, if you read an article
somewhere that says the method used is good, I won't argue
with that.

I try to sneak up on the settings if I can, and prefer
to not apply extreme voltages if I can avoid it. My previous
set of "hot DIMMs" had a problem, so I'm trying to not make
any more "hot DIMMs" on purpose by using extreme settings.

When the RAM on the machine started throwing errors a couple
of sets of RAM ago, I used Vnb as the only setting, to fix it.
Nothing else needed to be adjusted. I could do this though,
because I was working on a theory that the Northbridge was aging.
I knew the settings I'd used for everything else, were fine
at the time of the build.

Normally, the board will be selecting Command Rate 2T on its
own, whether because the DIMM clock is rather high, or
because the bus is loaded (four DIMMs). There's hardly a chance
to realistically use 1T (maybe a single set of single sided
4GB DIMMs).

Maybe you'll have more joy at 1333, with a full set.

Some older AMD setups used to work that way, DDR400
with two DIMMs, DDR333 with four DIMMs.

HTH,
Â*Â*Â*Â* Paul

This looks brilliant to me, Paul, and I thank you heartily.
It scares me a bit, though. I've never done memory-module clocking and
setting before; and when I went into the BIOS of this new machine it
presented me with a screen that looked like something from a flight-deck.
UEFI; so I clicked on "exit advanced" and it exited into booting.

I have some learning to do here, including the pages from the mobo
manual that Andy has shown me.
I'll get round to it.

Ed


There are sufficient "Autos" in there, you can work it a
bit at a time. For example, if you're timid, you can
just dial the DDR3-1600 setting to DDR3-1333 and see
if that's enough to get it working. Then you don't
have to look at all those pesky Tthis and Tthat ones.

And memtest86+ can give you a good idea how close you're
getting, or whether the situation is "hopeless". On my floppy-less
computer, I use a USB floppy drive for "tradition" and load
the memtest86+ floppy that way. (This would be with *no*
Windows hard drives connected, for safety from corruption.)

For example, my old Asrock motherboard, the documentation
(even the VIA website that made the chipset) said only
1GB DIMMs worked. Well, I plugged in a 2GB DIMM and
it was fully recognized. But, what was the problem ?
It was throwing errors.

Since the BIOS designers believed the documentation they were
given, they never "tuned" operation of the board for a 2GB stick.
There's a certain parameter that needs tuning in a lab,
and it would require me to do 256 test cases to "nail" the
timing (by trial and error). Well, that simply wasn't going
to happen. I don't have the stamina for that.

Your situation isn't nearly as bad, as the hardware is
mostly dialed in, and it could take as little as one
parameter change to tip it upright. Relaxing a single
timing, or adjusting the voltage within safe limits,
could be enough to make a difference.

I've spent as long as a week working on this stuff,
but I don't do that any more. Minimal change, test
for stability, and move on... :-)

It's one of the reasons I really like XMP. It does
everything for you, but only with two sticks. It's a bit
too hard for the manufacturer to test and guarantee
that four sticks will work in XMP setup mode, and the table
in the DIMM may not have the info needed. XMP is a section
of the SPD table, that was not designed by JEDEC. Some
"computer people" dreamed it up. If only the
process was rigorous enough to standardize...

Â*Â* Paul

You're a delight to correspond with, Paul. You have something that I
think Ernest Shackleton must have had; perseverance and gumption way
beyond the average. My father was named "Ernest" after him, and if I'm
ever trapped in Antarctica Id like to have him with me. Do you know his
story?

You might be interested in this communication from the forum of the
manufacturers. I don't want him in the Antarctic with me; a quitter he
seems to be.


*******************

"I take it you only have those 2 slots populated because if you try to
populate the other 2 slots the system doesn't load Windows as you were
saying.

If you remove the pair you know to work fine and put the new pair of 8gb
sticks into those slots, do they work as a pair on their own? Since we
shouldn't discount out of hand the chance that the new kit is faulty.

If they're not faulty, and the system loads Windows , what does the SPD
tab say for those?


You could also try only 1 of your 4 sticks at a time and leave the PC
running Memtest overnight for each to see whether any 1 of the 4 sticks
is faulty. https://www.memtest86.com/



In the meantime, going by what you've already posted:

The RAM you've installed is a different frequency and has different
timings to each other:

-Always install DIMMs with the same CAS latency. For optimal
compatibility, we recommend that you install memory modules of the same
version or date code (D/C) from the same vendor. Check with the retailer
toget the correct memory modules.
This is in the manual as you say and also on the QVL:
http://dlcdnet.asus.com/pub/ASUS/mb/...748.1522537474

The QVL is the list of RAM that's guaranteed compatible with the mobo.

Lots of others will also work too, but if you mix and match RAM of
different timings and frequencies that aren't on the QVL and then it
doesn't work, it's unfortunately not entirely surprising. It's actually
not at all unheard of for apparently identical RAM (same model names,
same frequencies, same latencies - but often different manufacture
dates) to not be compatible if they are not purchased as a matched kit.
They can be made with different dies, or something.

If the system lets you into the BIOS with all 4 sticks in you could try
lowering the frequency and loosening the timings to the lowest values
they have in common and see if that helps. That's more effort than I'd
be bothered with though.

Another question is whether you need 32gb RAM? This is an extremely old
set of CPUs and motherboard and most tasks would be badly bottlenecked
by other parts of the system e.g. the CPU long before RAM usage was an
issue. e.g. gaming, video editing. Probably even something like VMs.

Your best option may be to return your two most recently purchased
sticks and stick to 16gb."

*****************

All the best,

Ed

P.S. That guy's estimate of how long a memtest would take is way off. I
ran it with two sticks in, and it took not much longer than 40 mins. for
the full two passes on the 16GB.

Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
Andy Burns wrote:
Ed Cryer wrote:

The four are all the same; all 1600MHz, DDR3, 8GB.

Well, theoretically they are fine for that MB.

https://www.corsair.com/uk/en/Categories/Products/Memory/Mainstream-Memory/Corsair-Memory-%E2%80%94-8GB-DDR3-Memory/p/CMV8GX3M1A1600C11


Try a pair of the new ones in the blue slots at the tested speed
of 1333MHzÂ*Â*Â* 9-9-9-24

if that works then all four at that speed.

if that works, try two and finally all four of them
at the XMS speed of 1600MHz 11-11-11-30

I've not used an AMD CPU for years, so I'll take Paul's word for
it that a quad SIMM configuration is tricky.Â* I have no issues
with 4x8GB on "workstation" class intel motherboards.

Thanks for this. This sounds practical.
I shall try them, as you say.
Please explain "9-9-9-24"; and tell me how to set the speed.

Ed

9-9-9-24

tCAS tRCD tRP tRC

In roughly the same order as your DRAM customization screen.

The numbers are counts of "cycles".

At 1600MHz, the inverse is 0.6ns. 9 cycles at 0.6ns is 5.4ns.

Internally, all the RAM understands is nanoseconds.
"Did you meet my 5.4ns minimum ???" it asks.

If the RAM is set to DDR1333, the inverse of that is 0.75ns.
To make a number close to 5.4ns, we need about 7.2 cycles.
Rounding up, 8 cycles would work, without causing a smaller
number than the 5.4ns minimum.

9-9-9-24 DDR3-1600 or 8-8-8-X DDR3-1333

The timing table you can see in CPUZ shows
example timings that are recorded in the SPD
for this. Other values can be done by the BIOS
code by interpolation and so on.

Â*Â*Â* Paul

Here are the SPD tabs of CPU-Z for the two working modules.
Any comments?
https://www.dropbox.com/s/1d8v4h5mji...ts2-4.png?dl=0

Ed

Assuming those two samples are DIMMs from different
sets, they're very close to one another at the top setting.

You could start by using the frequency setting, and
let the BIOS do the math on the individual timings.

*******

Â*Â*Â* AI Overclock Tuner [Manual]
Â*Â*Â* Memory FrequencyÂ*Â* [DDR3-1600]Â* === Start with 1600, switch to
1333 if
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* there is trouble, let the
BIOS do the
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* math for the rest.
Â*Â*Â* DRAM Timing ControlÂ*Â* [example at 1600]
Â*Â*Â*Â*Â*Â*Â* tCASÂ*Â* Auto (11)
Â*Â*Â*Â*Â*Â*Â* tRCDÂ*Â* Auto (11)
Â*Â*Â*Â*Â*Â*Â* tRPÂ*Â*Â* Auto (11)
Â*Â*Â*Â*Â*Â*Â* tRCÂ*Â*Â* Auto (41)

Â*Â*Â*Â*Â*Â*Â* DRAM Command RateÂ* 2TÂ*Â*Â*Â* === definitely want 2T
Â*Â*Â* DRAM VoltageÂ* [1.50V]Â*Â*Â*Â* === You should be able to use between
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* Â*Â*Â*Â*Â*Â*Â*Â*Â* 1.5 and 1.65V without issue.
Â*Â*Â* NB VoltageÂ*Â*Â* [Auto]Â*Â*Â*Â*Â* === Leave this for now

If that doesn't work:Â* 11-12-11-41

Â*Â*Â*Â*Â*Â*Â* tRCDÂ*Â* 12
Â*Â*Â*Â*Â*Â*Â* DRAM Voltage [1.65V]

If that doesn't work

Â*Â*Â*Â*Â*Â*Â* tCASÂ*Â* 12
Â*Â*Â*Â*Â*Â*Â* tRCDÂ*Â* 12
Â*Â*Â*Â*Â*Â*Â* tRPÂ*Â*Â* 12
Â*Â*Â*Â*Â*Â*Â* tRCÂ*Â*Â* Auto === let BIOS work it out

Sometimes, the error rate and whether the errors are exactly
reproducible on each run, hints at whether the RAM chips
have an issue, or the bus timing is flaky, or the Vnb needs
to be adjusted up a notch or two. The Vnb regulators now
have fairly tiny step sizes and are capable of fine adjustment.

*******

While your board has an "auto overclock" kind of feature,
I don't like the voltages or extreme settings those tend to use.
Yes, if you jam the controls to max, you can practically
guarantee an overclock, and the methods they use on these
boards aren't exactly "subtle". Still, if you read an article
somewhere that says the method used is good, I won't argue
with that.

I try to sneak up on the settings if I can, and prefer
to not apply extreme voltages if I can avoid it. My previous
set of "hot DIMMs" had a problem, so I'm trying to not make
any more "hot DIMMs" on purpose by using extreme settings.

When the RAM on the machine started throwing errors a couple
of sets of RAM ago, I used Vnb as the only setting, to fix it.
Nothing else needed to be adjusted. I could do this though,
because I was working on a theory that the Northbridge was aging.
I knew the settings I'd used for everything else, were fine
at the time of the build.

Normally, the board will be selecting Command Rate 2T on its
own, whether because the DIMM clock is rather high, or
because the bus is loaded (four DIMMs). There's hardly a chance
to realistically use 1T (maybe a single set of single sided
4GB DIMMs).

Maybe you'll have more joy at 1333, with a full set.

Some older AMD setups used to work that way, DDR400
with two DIMMs, DDR333 with four DIMMs.

HTH,
Â*Â*Â*Â* Paul

This looks brilliant to me, Paul, and I thank you heartily.
It scares me a bit, though. I've never done memory-module clocking
and setting before; and when I went into the BIOS of this new machine
it presented me with a screen that looked like something from a
flight-deck.
UEFI; so I clicked on "exit advanced" and it exited into booting.

I have some learning to do here, including the pages from the mobo
manual that Andy has shown me.
I'll get round to it.

Ed


There are sufficient "Autos" in there, you can work it a
bit at a time. For example, if you're timid, you can
just dial the DDR3-1600 setting to DDR3-1333 and see
if that's enough to get it working. Then you don't
have to look at all those pesky Tthis and Tthat ones.

And memtest86+ can give you a good idea how close you're
getting, or whether the situation is "hopeless". On my floppy-less
computer, I use a USB floppy drive for "tradition" and load
the memtest86+ floppy that way. (This would be with *no*
Windows hard drives connected, for safety from corruption.)

For example, my old Asrock motherboard, the documentation
(even the VIA website that made the chipset) said only
1GB DIMMs worked. Well, I plugged in a 2GB DIMM and
it was fully recognized. But, what was the problem ?
It was throwing errors.

Since the BIOS designers believed the documentation they were
given, they never "tuned" operation of the board for a 2GB stick.
There's a certain parameter that needs tuning in a lab,
and it would require me to do 256 test cases to "nail" the
timing (by trial and error). Well, that simply wasn't going
to happen. I don't have the stamina for that.

Your situation isn't nearly as bad, as the hardware is
mostly dialed in, and it could take as little as one
parameter change to tip it upright. Relaxing a single
timing, or adjusting the voltage within safe limits,
could be enough to make a difference.

I've spent as long as a week working on this stuff,
but I don't do that any more. Minimal change, test
for stability, and move on... :-)

It's one of the reasons I really like XMP. It does
everything for you, but only with two sticks. It's a bit
too hard for the manufacturer to test and guarantee
that four sticks will work in XMP setup mode, and the table
in the DIMM may not have the info needed. XMP is a section
of the SPD table, that was not designed by JEDEC. Some
"computer people" dreamed it up. If only the
process was rigorous enough to standardize...

Â*Â*Â* Paul

You're a delight to correspond with, Paul. You have something that I
think Ernest Shackleton must have had; perseverance and gumption way
beyond the average. My father was namedÂ* "Ernest" after him, and if I'm
ever trapped in Antarctica Id like to have him with me. Do you know his
story?

You might be interested in this communication from the forum of the
manufacturers. I don't want him in the Antarctic with me; a quitter he
seems to be.


*******************

"I take it you only have those 2 slots populated because if you try to
populate the other 2 slots the system doesn't load Windows as you were
saying.

If you remove the pair you know to work fine and put the new pair of 8gb
sticks into those slots, do they work as a pair on their own? Since we
shouldn't discount out of hand the chance that the new kit is faulty.

If they're not faulty, and the system loads Windows , what does the SPD
tab say for those?


You could also try only 1 of your 4 sticks at a time and leave the PC
running Memtest overnight for each to see whether any 1 of the 4 sticks
is faulty. https://www.memtest86.com/



In the meantime, going by what you've already posted:

The RAM you've installed is a different frequency and has different
timings to each other:

-Always install DIMMs with the same CAS latency. For optimal
compatibility, we recommend that you install memory modules of the same
version or date code (D/C) from the same vendor. Check with the retailer
toget the correct memory modules.
This is in the manual as you say and also on the QVL:
http://dlcdnet.asus.com/pub/ASUS/mb/...748.1522537474

The QVL is the list of RAM that's guaranteed compatible with the mobo.

Lots of others will also work too, but if you mix and match RAM of
different timings and frequencies that aren't on the QVL and then it
doesn't work, it's unfortunately not entirely surprising. It's actually
not at all unheard of for apparently identical RAM (same model names,
same frequencies, same latencies - but often different manufacture
dates) to not be compatible if they are not purchased as a matched kit.
They can be made with different dies, or something.

If the system lets you into the BIOS with all 4 sticks in you could try
lowering the frequency and loosening the timings to the lowest values
they have in common and see if that helps. That's more effort than I'd
be bothered with though.

Another question is whether you need 32gb RAM? This is an extremely old
set of CPUs and motherboard and most tasks would be badly bottlenecked
by other parts of the system e.g. the CPU long before RAM usage was an
issue. e.g. gaming, video editing. Probably even something like VMs.

Your best option may be to return your two most recently purchased
sticks and stick to 16gb."

*****************

All the best,

Ed

P.S. That guy's estimate of how long a memtest would take is way off. I
ran it with two sticks in, and it took not much longer than 40 mins. for
the full two passes on the 16GB.


I got it working with all 4 sticks, 32GB. But only by tweaking it down
to 1066MHz.
It seems ok, though. That's subjective, not an objective test. It seems
just as fast to me.

Thanks to you and Andy for guiding me into new territory, and sticking
by me. Your student's doing well; and I put that down to good teaching
from both of you.

I'm wondering if doing a trial of tweaking it up to 1333MHz will be
worth while. You never know; perhaps after settling down with all four
sticks, it might take it.

Ed

Ed Cryer wrote:

I got it working with all 4 sticks, 32GB. But only by tweaking it down
to 1066MHz.
It seems ok, though. That's subjective, not an objective test. It seems
just as fast to me.

If you're not actively running many task, or processing huge files, then
the processor caching will 'cover' for slower memory to a certain extent.

I'm wondering if doing a trial of tweaking it up to 1333MHz will be
worth while.

I'd leave it as known working for a few days, until you're sure it's
running reliably, maybe leave memtest86+ or prime95 running overnight,
but then you've nothing to lose by trying 1333 ...

Ed Cryer wrote:
Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
Andy Burns wrote:
Ed Cryer wrote:

The four are all the same; all 1600MHz, DDR3, 8GB.

Well, theoretically they are fine for that MB.

https://www.corsair.com/uk/en/Categories/Products/Memory/Mainstream-Memory/Corsair-Memory-%E2%80%94-8GB-DDR3-Memory/p/CMV8GX3M1A1600C11


Try a pair of the new ones in the blue slots at the tested speed
of 1333MHz 9-9-9-24

if that works then all four at that speed.

if that works, try two and finally all four of them
at the XMS speed of 1600MHz 11-11-11-30

I've not used an AMD CPU for years, so I'll take Paul's word
for it that a quad SIMM configuration is tricky. I have no
issues with 4x8GB on "workstation" class intel motherboards.

Thanks for this. This sounds practical.
I shall try them, as you say.
Please explain "9-9-9-24"; and tell me how to set the speed.

Ed

9-9-9-24

tCAS tRCD tRP tRC

In roughly the same order as your DRAM customization screen.

The numbers are counts of "cycles".

At 1600MHz, the inverse is 0.6ns. 9 cycles at 0.6ns is 5.4ns.

Internally, all the RAM understands is nanoseconds.
"Did you meet my 5.4ns minimum ???" it asks.

If the RAM is set to DDR1333, the inverse of that is 0.75ns.
To make a number close to 5.4ns, we need about 7.2 cycles.
Rounding up, 8 cycles would work, without causing a smaller
number than the 5.4ns minimum.

9-9-9-24 DDR3-1600 or 8-8-8-X DDR3-1333

The timing table you can see in CPUZ shows
example timings that are recorded in the SPD
for this. Other values can be done by the BIOS
code by interpolation and so on.

Paul

Here are the SPD tabs of CPU-Z for the two working modules.
Any comments?
https://www.dropbox.com/s/1d8v4h5mji...ts2-4.png?dl=0

Ed

Assuming those two samples are DIMMs from different
sets, they're very close to one another at the top setting.

You could start by using the frequency setting, and
let the BIOS do the math on the individual timings.

*******

AI Overclock Tuner [Manual]
Memory Frequency [DDR3-1600] === Start with 1600, switch to
1333 if
there is trouble, let the
BIOS do the
math for the rest.
DRAM Timing Control [example at 1600]
tCAS Auto (11)
tRCD Auto (11)
tRP Auto (11)
tRC Auto (41)

DRAM Command Rate 2T === definitely want 2T
DRAM Voltage [1.50V] === You should be able to use between
1.5 and 1.65V without issue.
NB Voltage [Auto] === Leave this for now

If that doesn't work: 11-12-11-41

tRCD 12
DRAM Voltage [1.65V]

If that doesn't work

tCAS 12
tRCD 12
tRP 12
tRC Auto === let BIOS work it out

Sometimes, the error rate and whether the errors are exactly
reproducible on each run, hints at whether the RAM chips
have an issue, or the bus timing is flaky, or the Vnb needs
to be adjusted up a notch or two. The Vnb regulators now
have fairly tiny step sizes and are capable of fine adjustment.

*******

While your board has an "auto overclock" kind of feature,
I don't like the voltages or extreme settings those tend to use.
Yes, if you jam the controls to max, you can practically
guarantee an overclock, and the methods they use on these
boards aren't exactly "subtle". Still, if you read an article
somewhere that says the method used is good, I won't argue
with that.

I try to sneak up on the settings if I can, and prefer
to not apply extreme voltages if I can avoid it. My previous
set of "hot DIMMs" had a problem, so I'm trying to not make
any more "hot DIMMs" on purpose by using extreme settings.

When the RAM on the machine started throwing errors a couple
of sets of RAM ago, I used Vnb as the only setting, to fix it.
Nothing else needed to be adjusted. I could do this though,
because I was working on a theory that the Northbridge was aging.
I knew the settings I'd used for everything else, were fine
at the time of the build.

Normally, the board will be selecting Command Rate 2T on its
own, whether because the DIMM clock is rather high, or
because the bus is loaded (four DIMMs). There's hardly a chance
to realistically use 1T (maybe a single set of single sided
4GB DIMMs).

Maybe you'll have more joy at 1333, with a full set.

Some older AMD setups used to work that way, DDR400
with two DIMMs, DDR333 with four DIMMs.

HTH,
Paul

This looks brilliant to me, Paul, and I thank you heartily.
It scares me a bit, though. I've never done memory-module clocking
and setting before; and when I went into the BIOS of this new
machine it presented me with a screen that looked like something
from a flight-deck.
UEFI; so I clicked on "exit advanced" and it exited into booting.

I have some learning to do here, including the pages from the mobo
manual that Andy has shown me.
I'll get round to it.

Ed


There are sufficient "Autos" in there, you can work it a
bit at a time. For example, if you're timid, you can
just dial the DDR3-1600 setting to DDR3-1333 and see
if that's enough to get it working. Then you don't
have to look at all those pesky Tthis and Tthat ones.

And memtest86+ can give you a good idea how close you're
getting, or whether the situation is "hopeless". On my floppy-less
computer, I use a USB floppy drive for "tradition" and load
the memtest86+ floppy that way. (This would be with *no*
Windows hard drives connected, for safety from corruption.)

For example, my old Asrock motherboard, the documentation
(even the VIA website that made the chipset) said only
1GB DIMMs worked. Well, I plugged in a 2GB DIMM and
it was fully recognized. But, what was the problem ?
It was throwing errors.

Since the BIOS designers believed the documentation they were
given, they never "tuned" operation of the board for a 2GB stick.
There's a certain parameter that needs tuning in a lab,
and it would require me to do 256 test cases to "nail" the
timing (by trial and error). Well, that simply wasn't going
to happen. I don't have the stamina for that.

Your situation isn't nearly as bad, as the hardware is
mostly dialed in, and it could take as little as one
parameter change to tip it upright. Relaxing a single
timing, or adjusting the voltage within safe limits,
could be enough to make a difference.

I've spent as long as a week working on this stuff,
but I don't do that any more. Minimal change, test
for stability, and move on... :-)

It's one of the reasons I really like XMP. It does
everything for you, but only with two sticks. It's a bit
too hard for the manufacturer to test and guarantee
that four sticks will work in XMP setup mode, and the table
in the DIMM may not have the info needed. XMP is a section
of the SPD table, that was not designed by JEDEC. Some
"computer people" dreamed it up. If only the
process was rigorous enough to standardize...

Paul

You're a delight to correspond with, Paul. You have something that I
think Ernest Shackleton must have had; perseverance and gumption way
beyond the average. My father was named "Ernest" after him, and if
I'm ever trapped in Antarctica Id like to have him with me. Do you
know his story?

You might be interested in this communication from the forum of the
manufacturers. I don't want him in the Antarctic with me; a quitter he
seems to be.


*******************

"I take it you only have those 2 slots populated because if you try to
populate the other 2 slots the system doesn't load Windows as you were
saying.

If you remove the pair you know to work fine and put the new pair of
8gb sticks into those slots, do they work as a pair on their own?
Since we shouldn't discount out of hand the chance that the new kit is
faulty.

If they're not faulty, and the system loads Windows , what does the
SPD tab say for those?


You could also try only 1 of your 4 sticks at a time and leave the PC
running Memtest overnight for each to see whether any 1 of the 4
sticks is faulty. https://www.memtest86.com/



In the meantime, going by what you've already posted:

The RAM you've installed is a different frequency and has different
timings to each other:

-Always install DIMMs with the same CAS latency. For optimal
compatibility, we recommend that you install memory modules of the
same version or date code (D/C) from the same vendor. Check with the
retailer toget the correct memory modules.
This is in the manual as you say and also on the QVL:
http://dlcdnet.asus.com/pub/ASUS/mb/...748.1522537474

The QVL is the list of RAM that's guaranteed compatible with the mobo.

Lots of others will also work too, but if you mix and match RAM of
different timings and frequencies that aren't on the QVL and then it
doesn't work, it's unfortunately not entirely surprising. It's
actually not at all unheard of for apparently identical RAM (same
model names, same frequencies, same latencies - but often different
manufacture dates) to not be compatible if they are not purchased as a
matched kit. They can be made with different dies, or something.

If the system lets you into the BIOS with all 4 sticks in you could
try lowering the frequency and loosening the timings to the lowest
values they have in common and see if that helps. That's more effort
than I'd be bothered with though.

Another question is whether you need 32gb RAM? This is an extremely
old set of CPUs and motherboard and most tasks would be badly
bottlenecked by other parts of the system e.g. the CPU long before RAM
usage was an issue. e.g. gaming, video editing. Probably even
something like VMs.

Your best option may be to return your two most recently purchased
sticks and stick to 16gb."

*****************

All the best,

Ed

P.S. That guy's estimate of how long a memtest would take is way off.
I ran it with two sticks in, and it took not much longer than 40 mins.
for the full two passes on the 16GB.


I got it working with all 4 sticks, 32GB. But only by tweaking it down
to 1066MHz.
It seems ok, though. That's subjective, not an objective test. It seems
just as fast to me.

Thanks to you and Andy for guiding me into new territory, and sticking
by me. Your student's doing well; and I put that down to good teaching
from both of you.

I'm wondering if doing a trial of tweaking it up to 1333MHz will be
worth while. You never know; perhaps after settling down with all four
sticks, it might take it.

Ed


Well, it's purely a time trade-off.

How much time do you want to put into it ?

I used to "shoot for the value on the tin",
just to get my moneys worth.

The test time on my newest system is around 2 hours
when fully loaded, and I'd have to resort to testing
with just Test #5 while doing initial tuning.

If you've never tried these things before, I think
it's worthwhile to have a play with it, just to see
what you can achieve.

Since I have two computers in the room here, I could
take the second "offline" for some of this kind of
fun. But then I'd need to keep a notepad and pen
handy to write down what I'd tried and so on.

When you are finished your tuning and everything
is stable, on some motherboards you can "store"
a profile at the BIOS level. For example, my new
machine, I have two profiles stored. One is a
profile for "half-filled with sticks", which is
the old profile. And when I added the new sticks,
I have a "full of DIMMs" profile. If the computer
crashes or the power goes off at some unfortunate
moment, I have to go into the BIOS and "re-load"
the profile, to get all the custom settings back.

The machine I'm typing on, doesn't have profiles,
so if it becomes "forgetful" (CMOS battery dead,
power cord unplugged), then I'd better have my
notepad or digital camera pictures of the screen
to use as a reference :-) The "profile" idea on
newer motherboards, is great for eliminating
the drudgery of putting back settings after
some CMOS screwup.

Since the motherboard also has "Load Setup Defaults",
you can also accidentally blow out the custom setting,
just by being careless in the BIOS exit screen.
When the BIOS has a "profile" capability for re-loading,
hitting that by accident is no longer a problem.

Paul