Question The recommended Dual Channel config places the DIMM physically farther from CPU

[dor_13]

The MB recommends to place the DIMM physically farther from the CPU.
I don't understand why - it is commonly considered better to be closer.

Here is the recommendation from the motherboard specs:

https://imgur.com/a/irv6T4i

View: https://imgur.com/a/irv6T4i


Here is the schema of the DIMM sockets on the motherboard (from specs):

https://imgur.com/a/teYi1Vo

View: https://imgur.com/a/teYi1Vo

 

[Gam3r01]

Why would it be commonly considered better to be closer? Theres zero reason that would be the case.
Your motherboard manufacturer made the board, and they know how its wired, so follow what the manual recommends.
Every motherboard Ive used in my systems has been slots 2 and 4.

 

[USAFRet]

I can understand the basic theory behind "closer". Shorter traces/signal path between the RAM and CPU.
But the manufacturer has taken that into account. And is recommending A2 and B2.

Go with what they recommend.

-----------------------
In theory, there is no difference between practice and theory.
In practice, however...

 

[dor_13]

Why would it be commonly considered better to be closer? Theres zero reason that would be the case.
Your motherboard manufacturer made the board, and they know how its wired, so follow what the manual recommends.
Every motherboard Ive used in my systems has been slots 2 and 4.

I can understand the basic theory behind "closer". Shorter traces/signal path between the RAM and CPU.
But the manufacturer has taken that into account. And is recommending A2 and B2.

Go with what they recommend.

-----------------------
In theory, there is no difference between practice and theory.
In practice, however...

Sure I understand what you're saying, but I was hoping for a wise explanation.

 

[USAFRet]

Sure I understand what you're saying, but I was hoping for a wise explanation.

Possibly to alleviate CPU cooler clearance.

Which RAM Slots Should You Use?

Which RAM slots should you use for your RAM Sticks? Does it make a performance difference or can you place the Modules wherever you want to?

www.cgdirector.com

 

[BFG-9000]

DDR4 has the row of termination resistors on the DIMMs themselves, not the motherboard. So if you did not put a DIMM at the very end of the line, there would be an unterminated stub causing internal reflections to degrade the signal.

DDR5 goes even further and puts the terminator circuits on-die in the chips, and best practice there is also to put a DIMM in the furthest slot first.

In the very olden days of 50-pin SCSI it was required to insert DIP resistors into the very last device on both ends of the chain or the whole thing just plain wouldn't work. As SCSI devices shipped with these preinstalled, it was very common for people to neglect to remove them from devices placed in the middle of the chain, only to find everything downstream from it disappeared. These memory controllers are obviously doing something to automatically ignore the terminators in the middle of the chain, but still need the last one.

 

[drea.drechsler]

@dor_13 : This kind of completes what @BFG-9000 started with explaining the function of the termination resistors. But that alone doesn't really explain why the 2nd and not the 1st.

So yes, it can certainly help with cooler clearance as @USAFRet suggests but there is actually a very good electrical reason why. It's called an unterminated stub on a transmission line. So recall that each DIMM provides termination resistors but whichever socket you use there will be left an unterminated stub (length of transmission line, or PWB trace, to a DIMM socket) of unequal length on each channel of a 4-socket motherboard using daisy chain topology, the most common.

If you use the closer socket, the unterminated stub reaches all the way to the 2nd socket on the transmission line. It's much longer which causes it to generate a higher amplitude reflected signal or absorb more of the energy of the transmitted signal. During memory training the IMC tries to compensate for that but it's hard and approaches impossible at higher clocks for longer stub lengths at higher clock speeds.

If you use the further socket the remaining unterminated stub is quite short and those problems are mostly avoided and memory training is more likely to be successful at higher clock speeds.

This is also why two socket boards (one socket for each channel) are considered the best for maximum overclocking potential. The layout can be fully optimized for one termination with no unterminated stubs that need to be compensated for.

A t-topology provides stub paths of equal length to both DIMM's on a channel so it doesn't matter which socket you use (except when clearance is an issue) as they're both equally compromized (vs. the 2nd socket of a daisy chain) because there will always be one long unterminated stub left. However, because they are equal they're also more reliable for 4-DIMM setups at very high clocks.

 

[dor_13]

@dor_13 : This kind of completes what @BFG-9000 started with explaining the function of the termination resistors. But that alone doesn't really explain why the 2nd and not the 1st.

So yes, it can certainly help with cooler clearance as @USAFRet suggests but there is actually a very good electrical reason why. It's called an unterminated stub on a transmission line. So recall that each DIMM provides termination resistors but whichever socket you use there will be left an unterminated stub (length of transmission line, or PWB trace, to a DIMM socket) of unequal length on each channel of a 4-socket motherboard using daisy chain topology, the most common.

If you use the closer socket, the unterminated stub reaches all the way to the 2nd socket on the transmission line. It's much longer which causes it to generate a higher amplitude reflected signal or absorb more of the energy of the transmitted signal. During memory training the IMC tries to compensate for that but it's hard and approaches impossible at higher clocks for longer stub lengths at higher clock speeds.

If you use the further socket the remaining unterminated stub is quite short and those problems are mostly avoided and memory training is more likely to be successful at higher clock speeds.

This is also why two socket boards (one socket for each channel) are considered the best for maximum overclocking potential. The layout can be fully optimized for one termination with no unterminated stubs that need to be compensated for.

A t-topology provides stub paths of equal length to both DIMM's on a channel so it doesn't matter which socket you use, they're both equally compromized (vs. the 2nd socket of a daisy chain) because there will always be one long unterminated stub left. However, because they are equal they're also more reliable for 4-DIMM setups at very high clocks.

So, for a motherboard which has 4 DIMM sockets - it would be better to install 4 DIMMs instead of 2 DIMMs, correct?

 

[drea.drechsler]

So, for a motherboard which has 4 DIMM sockets - it would be better to install 4 DIMMs instead of 2 DIMMs, correct?

Only if you know it uses a T-topology and you're shooting for very high overclocks.

Not many motherboards do, apparently. And those that do are in the very top-tier because it's costly running all the extra trace paths it requires in a region of the board already very dense. It's usually seen on boards with 6 or more PWB layers.

 

[dor_13]

Only if you know it uses a T-topology and you're shooting for very high overclocks.

Not many motherboards do, apparently. And those that do are in the very top-tier because it's costly running all the extra trace paths it requires in a region of the board already very dense. It's usually seen on boards with 6 or more PWB layers.

From the video (start 07:27):

View: https://youtu.be/3vQwGGbW1AE?t=447


The t-topolgy seems to have a lower performance than the daisy chain.

With these kinds of things, no wonder RAM errors could happen.

 

[drea.drechsler]

...
The t-topolgy seems to have a lower performance than the daisy chain.

With these kinds of things, no wonder RAM errors could happen.

He's DEFINITELY the guy to watch for these things

You can get 64GB DIMM's now, 2 DIMM's making a 128GB memory pool with no compromises a 4 DIMM setup would impose. The average user (a gamer) still only needs 16GB (2x8GB), 32GB total is probably unnecessary but what people do. I don't see a need for worrying about 4 DIMM's.

Personally, I think people want 4 DIMM's because it looks so cool all in a row, lit up in glowy RGB colors. Performance compromised for aesthetics, but whatever.

Using all 4 DIMM's is useful though when you need a LOT of memory. When you run out of memory in a process it starts hitting virtual memory which is several orders of magnitude slower than DDR4 even at lower clock speeds. THAT's the performance comparison that is most relevant.

 

[hotaru.hino]

The t-topolgy seems to have a lower performance than the daisy chain.

With these kinds of things, no wonder RAM errors could happen.

Taking all of this however, any performance deltas you may be concerned about only matters if you're trying to shoot for the top-end of performance. For instance, in the video mentioned above, the speeds mentioned where it starts mattering are DDR4-3600 or higher. For something like a Ryzen processor, going beyond DDR4-3600 isn't worthwhile because it has to operate a memory command ratio lower than 1:1, resulting in increased latency. While you could probably push the memory controller to go further so its speed matches the memory, you're playing the silicon lottery at that point.