RAM compatibility

One of my towers has two slabs of DDR3 RAM. They have the same part
number but a slightly different max bandwidth.
Are these running well together? Or should I make them more compatible?
Here are the details given by Speccy;

Slot #1
Type DDR3
Size 8192 MBytes
Manufacturer Corsair
Max Bandwidth PC3-10700 (667 MHz)
Part Number CMV8GX3M1A1600C11
SPD Ext. XMP
Timing table
Frequency CAS# Latency RAS# To CAS# RAS# Precharge tRAS tRC Voltage
JEDEC #1 457.1 MHz 6.0 6 6 17 22 1.500 V
JEDEC #2 533.3 MHz 7.0 7 7 20 26 1.500 V
JEDEC #3 666.7 MHz 9.0 9 9 24 33 1.500 V
XMP-1600 800 MHz 11.0 11 11 30 1.500 V
Slot #2
Type DDR3
Size 8192 MBytes
Manufacturer Corsair
Max Bandwidth PC3-12800K (800 MHz)
Part Number CMV8GX3M1A1600C11
Timing table
Frequency CAS# Latency RAS# To CAS# RAS# Precharge tRAS tRC Voltage
JEDEC #1 436.4 MHz 6.0 6 6 17 21 1.500 V
JEDEC #2 509.1 MHz 7.0 7 7 20 25 1.500 V
JEDEC #3 654.5 MHz 9.0 9 9 25 32 1.500 V
JEDEC #4 800.0 MHz 11.0 11 11 30 39 1.500 V

Ed Cryer wrote:
One of my towers has two slabs of DDR3 RAM. They have the same part
number but a slightly different max bandwidth.
Are these running well together? Or should I make them more compatible?
Here are the details given by Speccy;

Slot #1
Type DDR3
Size 8192 MBytes
Manufacturer Corsair
Max Bandwidth PC3-10700 (667 MHz)
Part Number CMV8GX3M1A1600C11
SPD Ext. XMP
Timing table
Frequency CAS# Latency RAS# To CAS# RAS#
Precharge tRAS tRC Voltage
JEDEC #1 457.1 MHz 6.0 6 6 17 22 1.500 V
JEDEC #2 533.3 MHz 7.0 7 7 20 26 1.500 V
JEDEC #3 666.7 MHz 9.0 9 9 24 33 1.500 V
XMP-1600 800 MHz 11.0 11 11 30 1.500 V
Slot #2
Type DDR3
Size 8192 MBytes
Manufacturer Corsair
Max Bandwidth PC3-12800K (800 MHz)
Part Number CMV8GX3M1A1600C11
Timing table
Frequency CAS# Latency RAS# To CAS# RAS#
Precharge tRAS tRC Voltage
JEDEC #1 436.4 MHz 6.0 6 6 17 21 1.500 V
JEDEC #2 509.1 MHz 7.0 7 7 20 25 1.500 V
JEDEC #3 654.5 MHz 9.0 9 9 25 32 1.500 V
JEDEC #4 800.0 MHz 11.0 11 11 30 39 1.500 V

The BIOS selects a set of conditions that "fit"
within the SPD tables. It examines all the sticks,
and in effect, sees what they have in common.

You can still change the clocking selector, within those
choices. Say for example, I dropped the clock to 533MHz.
Then the BIOS works out on each stick, what the equivalent
CAS is, and considers whether that would work for all the
sticks and so on. If the math for one stick is "CAS 5"
and the math for the other is "CAS 6", then both will be
run at CAS 6.

Using CPUZ (cpuid.com) portable version in the ZIP,
you would also have a page with the "current" settings
as measured at OS level. The BIOS has the SPD tables to
work with (from the EEPROM on each DIMM, connected to
the Serial Presence Detect bus). Once the BIOS has made
its choice, that is reflected in the "current" settings,
and those settings are used for all sticks.

No attempt is made to run all the sticks asynchronous from
one another, in a free-for-all battle-royale. That would
lead to conflicts and dumpster fires, if it was done
that way. So instead, all the rowers in the row boat,
row in synchrony, to best effect. Their oars never
bang together, because they all run at the same speed.

I would say, without even looking, the BIOS will "make
them run well". Because that's the BIOS job. However,
you can examine the choices made, and decide whether
the sticks have more headroom. That's why one of the
sticks has an XMP table entry - it implies, for a two stick
dual channel setup, the DIMM could be "set to its
box label overclocker value" by using just one
tick box on the BIOS screen. That's the XMP box.

But the XMP choice is, naturally, gated by reality.
It works best if both DIMMs have the *exact same*
XMP value. XMP is intended to allow raising enthusiast
grade DIMMs, to the specified level. The idea is,
the "default" choice is not the "box label" value,
and the "default" choice is slightly slower, to ensure
(for a naive user), that the DIMM install leads to
"first time success". As the DIMM owner becomes
more familiar with the product, and the DIMM owner
is convinced it's going to be stable at 1600,
then you could tick the XMP thing.

If the computer crashes for some reason, then you'd
do a Clear CMOS (with all power removed) and
go back to the BIOS screen and prepare a less
aggressive setting. On some computers, if you
crash them three times in a row, the BIOS tables
are automatically reset to nominal, and you get
the effect of Clear CMOS, without taking the
door off the PC to get at the jumper.

On my own machines here, I don't always torture
the memory subsystem, by clocking it as high as
possible. If all slots are full, the bus loading
makes the XMP stop working, and manual tuning
is likely to be required to get the last ounce
of performance. And because Intel CPUs do a good
job on multi-level caching, the impact of getting
the last ounce of speed is minimal. (Mainly,
7ZIP compression runs faster, if you use
the highest possible settings. Many games
don't even notice.)

However, if your current settings were purposely
turned way way down, by a miscreant, then sure,
a visit to the BIOS and some adjustments, you might
just barely be able to detect the improvement.

As an example, using Intel graphics (which use system
memory), try with only one stick installed, try
with two sticks in dual channel, and you'll notice
the Intel graphics "are a tiny bit more snappy"
with two sticks in dual channel. It takes some
pretty significant mis-adjustments, to cause
the computer to be perceptibly altered like that.

Summary: We need to see the choice the BIOS made for
the "current" value, to know how good of a
job the BIOS did. Using CPUZ can get that
value for you. I'm sure speccy can too. Have
a look for the current value if you can find it.

XMP will only work if the "the machine can run
that fast", there are two DIMMs in dual channel,
and the XMP entries are identical. This helps
it to work "first time, minimal fuss". When I
installed half the memory I own, the XMP was
perfectly stable as soon as I ticked that box.
With all slots full, I can't use XMP (there's
no entry for "all slots" in their SPD table).

Paul

On 08/05/2020 12:48, Ed Cryer wrote:
One of my towers has two slabs of DDR3 RAM. They have the same part
number but a slightly different max bandwidth.
Are these running well together? Or should I make them more compatible?


If the machine is working and doing things that it's supposed to do then
why are you trying to be stupid to throw away good money on a very old
machine?Â* DDR3 is very old and and there is no point in spending any
more money on it just because some idiot tells you that you should buy
"more compatible" ram memory (whatever this crap means).

Just use the machine as it is because if you try to upgrade it, you're
likely to make it unstable and likely to cause more problems.Â* It's old
and you have to accept that everything has to doe one day and new young
ones will take their place!


--
With over 1.2 billion devices now running Windows 10, customer
satisfaction is higher than any previous version of windows.

Ed Cryer wrote:

One of my towers has two slabs of DDR3 RAM. They have the same part
number but a slightly different max bandwidth.
Are these running well together? Or should I make them more compatible?
Here are the details given by Speccy;

Slot #1
Type DDR3
Size 8192 MBytes
Manufacturer Corsair
Max Bandwidth PC3-10700 (667 MHz)
Part Number CMV8GX3M1A1600C11
SPD Ext. XMP
Timing table
Frequency CAS# Latency RAS# To CAS# RAS# Precharge tRAS tRC Voltage
JEDEC #1 457.1 MHz 6.0 6 6 17 22 1.500 V
JEDEC #2 533.3 MHz 7.0 7 7 20 26 1.500 V
JEDEC #3 666.7 MHz 9.0 9 9 24 33 1.500 V
XMP-1600 800 MHz 11.0 11 11 30 1.500 V
Slot #2
Type DDR3
Size 8192 MBytes
Manufacturer Corsair
Max Bandwidth PC3-12800K (800 MHz)
Part Number CMV8GX3M1A1600C11
Timing table
Frequency CAS# Latency RAS# To CAS# RAS# Precharge tRAS tRC Voltage
JEDEC #1 436.4 MHz 6.0 6 6 17 21 1.500 V
JEDEC #2 509.1 MHz 7.0 7 7 20 25 1.500 V
JEDEC #3 654.5 MHz 9.0 9 9 25 32 1.500 V
JEDEC #4 800.0 MHz 11.0 11 11 30 39 1.500 V

The tool is getting the specs as reported by the sig data on the RAM
modules. The tool reports the SPD (Serial Presence Detect) reported by
the memory modules. See:

https://en.wikipedia.org/wiki/Serial_presence_detect

Speccy is reporting the same info it can get off the memory modules as
can WMIC in Windows. In a command shell with admin privileges, run:

wmic memorychip get devicelocator, manufacturer
wmic memorychip get devicelocator, partnumber
wmic memorychip get devicelocator, serialnumber
wmic memorychip get devicelocator, capacity
wmic memorychip get devicelocator, memorytype
wmic memorychip get devicelocator, speed
wmic memorychip get devicelocator, formfactor

and to see a report on a per-module basis, run:

wmic memorychip list full

The BIOS should be using the lowest of the reported SPD speeds. As I
recall, that's typically named the "optimal" setting. Presumably you
aren't overclocking to override the auto-detect in the BIOS. Some
BIOSes don't read or use the SPD from all memory modules. They may only
use the SPD from the modules in the first bank (and perhaps just the
first slot in the first bank). If you have a mix of memory modules
where some are slower than others, put the slowest in slot 1. Then all
modules will run at that slowest speed.

You should be able to go into the BIOS/UEFI settings to check Auto is
selected for RAM speed, and it might show the settings for each module.
You might have to go into overclock/tweak/advanced settings to see
those, but don't change anything unless you are extremely intimate with
the hardware.

I don't know if you'll remember this, Paul, but about a year and a half
ago I was looking for a "gaming PC" and you helped me with various
options I was considering. I ended up with this Speedy-Gonzales box of
tricks, and still love it. I particularly remember all the jazzy-jazzy
boxes we looked at, with names like Growls-like-a-Trex and Predator. We
agreed that most of them were built to impress teenagers, and overpriced.
I also had a lot of fun overclocking it, and you guided me through that;
and then how to handle cables neatly wrapped inside the box.
All good fun. I'm no gamer, but I did enjoy the learning curve.

Ed



On 08/05/2020 19:41, Paul wrote:
Ed Cryer wrote:
One of my towers has two slabs of DDR3 RAM. They have the same part
number but a slightly different max bandwidth.
Are these running well together? Or should I make them more compatible?
Here are the details given by Speccy;

Slot #1
Â*Â*Â* TypeÂ*Â*Â* DDR3
Â*Â*Â* SizeÂ*Â*Â* 8192 MBytes
Â*Â*Â* ManufacturerÂ*Â*Â* Corsair
Â*Â*Â* Max BandwidthÂ*Â*Â* PC3-10700 (667 MHz)
Â*Â*Â* Part NumberÂ*Â*Â* CMV8GX3M1A1600C11
Â*Â*Â* SPD Ext.Â*Â*Â* XMP
Â*Â*Â*Â*Â*Â*Â* Timing table
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* FrequencyÂ*Â*Â* CAS# LatencyÂ*Â*Â* RAS# To CAS#Â*Â*Â* RAS#
PrechargeÂ*Â*Â* tRASÂ*Â*Â* tRCÂ*Â*Â* Voltage
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* JEDEC #1Â*Â*Â* 457.1 MHzÂ*Â*Â* 6.0Â*Â*Â* 6Â*Â*Â* 6Â*Â*Â* 17Â*Â*Â* 22Â*Â*Â* 1.500 V
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* JEDEC #2Â*Â*Â* 533.3 MHzÂ*Â*Â* 7.0Â*Â*Â* 7Â*Â*Â* 7Â*Â*Â* 20Â*Â*Â* 26Â*Â*Â* 1.500 V
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* JEDEC #3Â*Â*Â* 666.7 MHzÂ*Â*Â* 9.0Â*Â*Â* 9Â*Â*Â* 9Â*Â*Â* 24Â*Â*Â* 33Â*Â*Â* 1.500 V
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* XMP-1600Â*Â*Â* 800 MHzÂ*Â*Â* 11.0Â*Â*Â* 11Â*Â*Â* 11Â*Â* 30Â*Â*Â*Â*Â*Â*Â*Â*Â* 1.500 V
Slot #2
Â*Â*Â* TypeÂ*Â*Â* DDR3
Â*Â*Â* SizeÂ*Â*Â* 8192 MBytes
Â*Â*Â* ManufacturerÂ*Â*Â* Corsair
Â*Â*Â* Max BandwidthÂ*Â*Â* PC3-12800K (800 MHz)
Â*Â*Â* Part NumberÂ*Â*Â* CMV8GX3M1A1600C11
Â*Â*Â*Â*Â*Â*Â* Timing table
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* FrequencyÂ*Â*Â* CAS# LatencyÂ*Â*Â* RAS# To CAS#Â*Â*Â* RAS#
PrechargeÂ*Â*Â* tRASÂ*Â*Â* tRCÂ*Â*Â* Voltage
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* JEDEC #1Â*Â*Â* 436.4 MHzÂ*Â*Â* 6.0Â*Â*Â* 6Â*Â*Â* 6Â*Â*Â* 17Â*Â*Â* 21Â*Â*Â* 1.500 V
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* JEDEC #2Â*Â*Â* 509.1 MHzÂ*Â*Â* 7.0Â*Â*Â* 7Â*Â*Â* 7Â*Â*Â* 20Â*Â*Â* 25Â*Â*Â* 1.500 V
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* JEDEC #3Â*Â*Â* 654.5 MHzÂ*Â*Â* 9.0Â*Â*Â* 9Â*Â*Â* 9Â*Â*Â* 25Â*Â*Â* 32Â*Â*Â* 1.500 V
Â*Â*Â*Â*Â*Â*Â*Â*Â*Â*Â* JEDEC #4Â*Â*Â* 800.0 MHzÂ*Â*Â* 11.0Â* 11Â*Â*Â* 11Â*Â* 30Â*Â*Â* 39Â*Â*Â* 1.500 V

The BIOS selects a set of conditions that "fit"
within the SPD tables. It examines all the sticks,
and in effect, sees what they have in common.

You can still change the clocking selector, within those
choices. Say for example, I dropped the clock to 533MHz.
Then the BIOS works out on each stick, what the equivalent
CAS is, and considers whether that would work for all the
sticks and so on. If the math for one stick is "CAS 5"
and the math for the other is "CAS 6", then both will be
run at CAS 6.

Using CPUZ (cpuid.com) portable version in the ZIP,
you would also have a page with the "current" settings
as measured at OS level. The BIOS has the SPD tables to
work with (from the EEPROM on each DIMM, connected to
the Serial Presence Detect bus). Once the BIOS has made
its choice, that is reflected in the "current" settings,
and those settings are used for all sticks.

No attempt is made to run all the sticks asynchronous from
one another, in a free-for-all battle-royale. That would
lead to conflicts and dumpster fires, if it was done
that way. So instead, all the rowers in the row boat,
row in synchrony, to best effect. Their oars never
bang together, because they all run at the same speed.

I would say, without even looking, the BIOS will "make
them run well". Because that's the BIOS job. However,
you can examine the choices made, and decide whether
the sticks have more headroom. That's why one of the
sticks has an XMP table entry - it implies, for a two stick
dual channel setup, the DIMM could be "set to its
box label overclocker value" by using just one
tick box on the BIOS screen. That's the XMP box.

But the XMP choice is, naturally, gated by reality.
It works best if both DIMMs have the *exact same*
XMP value. XMP is intended to allow raising enthusiast
grade DIMMs, to the specified level. The idea is,
the "default" choice is not the "box label" value,
and the "default" choice is slightly slower, to ensure
(for a naive user), that the DIMM install leads to
"first time success". As the DIMM owner becomes
more familiar with the product, and the DIMM owner
is convinced it's going to be stable at 1600,
then you could tick the XMP thing.

If the computer crashes for some reason, then you'd
do a Clear CMOS (with all power removed) and
go back to the BIOS screen and prepare a less
aggressive setting. On some computers, if you
crash them three times in a row, the BIOS tables
are automatically reset to nominal, and you get
the effect of Clear CMOS, without taking the
door off the PC to get at the jumper.

On my own machines here, I don't always torture
the memory subsystem, by clocking it as high as
possible. If all slots are full, the bus loading
makes the XMP stop working, and manual tuning
is likely to be required to get the last ounce
of performance. And because Intel CPUs do a good
job on multi-level caching, the impact of getting
the last ounce of speed is minimal. (Mainly,
7ZIP compression runs faster, if you use
the highest possible settings. Many games
don't even notice.)

However, if your current settings were purposely
turned way way down, by a miscreant, then sure,
a visit to the BIOS and some adjustments, you might
just barely be able to detect the improvement.

As an example, using Intel graphics (which use system
memory), try with only one stick installed, try
with two sticks in dual channel, and you'll notice
the Intel graphics "are a tiny bit more snappy"
with two sticks in dual channel. It takes some
pretty significant mis-adjustments, to cause
the computer to be perceptibly altered like that.

Summary: We need to see the choice the BIOS made for
Â*Â*Â*Â*Â*Â*Â*Â* the "current" value, to know how good of a
Â*Â*Â*Â*Â*Â*Â*Â* job the BIOS did. Using CPUZ can get that
Â*Â*Â*Â*Â*Â*Â*Â* value for you. I'm sure speccy can too. Have
Â*Â*Â*Â*Â*Â*Â*Â* a look for the current value if you can find it.

Â*Â*Â*Â*Â*Â*Â*Â* XMP will only work if the "the machine can run
Â*Â*Â*Â*Â*Â*Â*Â* that fast", there are two DIMMs in dual channel,
Â*Â*Â*Â*Â*Â*Â*Â* and the XMP entries are identical. This helps
Â*Â*Â*Â*Â*Â*Â*Â* it to work "first time, minimal fuss". When I
Â*Â*Â*Â*Â*Â*Â*Â* installed half the memory I own, the XMP was
Â*Â*Â*Â*Â*Â*Â*Â* perfectly stable as soon as I ticked that box.
Â*Â*Â*Â*Â*Â*Â*Â* With all slots full, I can't use XMP (there's
Â*Â*Â*Â*Â*Â*Â*Â* no entry for "all slots" in their SPD table).

Â*Â* Paul

On 08/05/2020 20:00, 😉 Good Guy 😉 wrote:
On 08/05/2020 12:48, Ed Cryer wrote:
One of my towers has two slabs of DDR3 RAM. They have the same part
number but a slightly different max bandwidth.
Are these running well together? Or should I make them more compatible?


If the machine is working and doing things that it's supposed to do then
why are you trying to be stupid to throw away good money on a very old
machine?Â* DDR3 is very old and and there is no point in spending any
more money on it just because some idiot tells you that you should buy
"more compatible" ram memory (whatever this crap means).

Just use the machine as it is because if you try to upgrade it, you're
likely to make it unstable and likely to cause more problems.Â* It's old
and you have to accept that everything has to doe one day and new young
ones will take their place!


--
With over 1.2 billion devices now running Windows 10, customer
satisfaction is higher than any previous version of windows.

Don't let your spleen outrun your knowledge and reason, old feller.
You'll find that DDR3 is still far more widely used than DDR4. And as
for updating PCs, in this box alone I've installed a SSD, new
Bluray-DVDR combo, extra memory, and a few other odds and ends.
I enjoy doing it; I have a very high success rate; and the learning
acquired helps me help others.

ED