Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Status
The first post of this thread is a WikiPost and can be edited by anyone with the appropiate permissions. Your edits will be public.

wilberforce

macrumors 68030
Aug 15, 2020
2,930
3,207
SF Bay Area
And I think you're underthinking this.

Irrespective of whether what you said about memory kits vs singles is correct, the "it either meets the spec or it doesn't" is an oversimplification.

According to Crucial, "Slight variations in the assembly process, not to mention the differences in production among the manufacturers, might affect the performance of the modules." The phrase "differences is production among manufacturers" refers to different manufacturers using different production techniques. But the phrase "Slight variations in the assembly process" is referring to variations within a specific part no. by a specific mfr. And these are for modules that are sold commercially, and thus meet QC. So Crucial is saying it's not as simple as whether it meets the spec or it doesn't.

Besides, if all that mattered is if you're in-spec or not, then mixing and matching in-spec modules from different manufacturers, so long as those modules had the same spec, would never be an issue. Yet we know that is not the case.

This article is a mixture of facts and self-serving FUD; to be best informed I suggest separate the two. (Pretty obvious which is which, so not going to debate the article point by point.)
In any case, nothing in this article denies what I said: that if all part numbers are the same, then they are matched.

To address your original question: whatever you decide are "matched" DIMMs, then the DIMMs in Channel A (slots 1 and 2) should match the DIMMs in Channel B (slots 3 and 4), for full dual channel operation.

The 2020 iMac (only) has an additional unexplained requirement/quirk, if using 4 DIMMs: the part numbers in slots 1 and 2 must match, and the part numbers in slots 3 and 4 must match, else speed drops to 2133 MT/s.
 
Last edited:

theorist9

macrumors 68040
May 28, 2015
3,880
3,059
This article is a mixture of facts and self-serving FUD; to be best informed I suggest separate the two. (Pretty obvious which is which, so not going to debate the article point by point.)
In any case, nothing in this article denies what I said: that if all part numbers are the same, then they are matched.
OK, fair enough about the Crucial article. But I want to understand this at more than merely a superficial level. So let's recall this quote from the OP, which warns of the problem of mixing otherwise identical sticks from different vendors. And it's not just the OP that warns agains this—it's a general recommendation.
Please note that many people on this forum experienced problems when mixing genuine Apple RAM with third party RAM. It might not be possible to mix your genuine Apple RAM with the modules listed above without impacting stability and performance. Memory configuration with 2 or 4 memory modules coming from the same third party vendor is strongly recommended.

For troubleshooting RAM, please refer to this thread Memory upgrade troubleshooting in 2020 27-inch iMacs

Rules of thumb :

  1. Do not mix RAM modules with different capacity/speed/manufacturer. A OWC memory kit with Hynix ICs won't work with an OWC memory kit with Micron ICs, even if they are "For Mac", with the same speed, timings, comes from the same manufacturer, and other specs.
So, given this, if mixing otherwise identical sticks (same size, speed, and timing) from different vendors can cause issues, doesn't that mean, logically, that just meeting spec isn't enough to ensure they all work together—i.e., that manufacturing variations can make the sticks different enough to not work together, thus contradicting the view you expressed here:?
It either meets the spec or it doesn't. There is no benefit to meeting the spec a little bit better.
Yes, in practice it may be that having the same part no. is sufficient, manufacturing variations aside. But even with that, it appears things are more nuanced than what your above quote indicates. It looks like manufacturing variations do matter, and that, between vendors, they can be large enough to make a practical difference, but within the same vendor and part no. they are not (?).
To address your original question: whatever you decide are "matched" DIMMs, then the DIMMs in Channel A (slots 1 and 2) should match the DIMMs in Channel B (slots 3 and 4), for full dual channel operation.
You're saying the pair in A should match the pair in B, which doesn't answer my question—I'm not asking about matching between pairs, I'm asking about matching within pairs. Here's what I asked:

If pair X consists of matching sticks X1 and X2, and pair Y of matching sticks Y1 and Y2, should it be configuration I or configuration II:

CONFIGURATION I
BANK 0/ChannelA-DIMM0: stick X1
BANK 1/ChannelA-DIMM1: stick X2
BANK 2/ChannelB-DIMM0: stick Y1
BANK 3/ChannelB-DIMM1: stick Y2

OR:

CONFIGURATION II
BANK 0/ChannelA-DIMM0: stick X1
BANK 1/ChannelA-DIMM1: stick Y1
BANK 2/ChannelB-DIMM0: stick X2
BANK 3/ChannelB-DIMM1: stick Y2
 
Last edited:

wilberforce

macrumors 68030
Aug 15, 2020
2,930
3,207
SF Bay Area
OK, fair enough about the Crucial article. But I want to understand this at more than merely a superficial level. So let's recall this quote from the OP, which warns of the problem of mixing otherwise identical sticks from different vendors. And it's not just the OP that warns agains this—it's a general recommendation.

So, given this, if mixing otherwise identical sticks (same size, speed, and timing) from different vendors can cause issues, doesn't that mean, logically, that just meeting spec isn't enough to ensure they all work together—i.e., that manufacturing variations can make the sticks different enough to not work together, thus contradicting the view you expressed here:?

Yes, in practice it may be that having the same part no. is sufficient, manufacturing variations aside. But even with that, it appears things are more nuanced than what your above quote indicates. It looks like manufacturing variations do matter, and that, between vendors, they can be large enough to make a practical difference, but within the same vendor and part no. they are not (?).

You're saying the pair in A should match the pair in B, which doesn't answer my question—I'm not asking about matching between pairs, I'm asking about matching within pairs. Here's what I asked:

If pair X consists of matching sticks X1 and X2, and pair Y of matching sticks Y1 and Y2, should it be configuration I or configuration II:

CONFIGURATION I
BANK 0/ChannelA-DIMM0: stick X1
BANK 1/ChannelA-DIMM1: stick X2
BANK 2/ChannelB-DIMM0: stick Y1
BANK 3/ChannelB-DIMM1: stick Y2

OR:

CONFIGURATION II
BANK 0/ChannelA-DIMM0: stick X1
BANK 1/ChannelA-DIMM1: stick Y1
BANK 2/ChannelB-DIMM0: stick X2
BANK 3/ChannelB-DIMM1: stick Y2
Configuration II, because then Channel A matches Channel B.

However, (for configuration II), if part number for X1 does not match part number for Y1, OR part number for X2 does not match part number for Y2, the speed will drop to 2133 MT/s. This applies only for the 2020 iMac.

Just to reiterate: you and I have different definitions of what constitutes a satisfactory "match." You have a much more rigorous definition than I do. I and others have not found this rigorous a definition necessary. I and others have found that only matching capacity (GB), clock speed and latency is necessary (and not manufacturer, batch, etc.). However, you are of course welcome to use your more rigorous definition - I just do not think it necessary.

EDIT: The issue with OP's post about problems with mixing Apple RAM with third party RAM stems from two things: the third party RAM is typically larger capacity than the base Apple RAM (2x4GB), and the part number for the Apple RAM is not the same as the part number for the third party RAM (obviously). This makes it impossible to configure the RAM such that Channel A matches Channel B, AND the part numbers in slots 1 and 2 match, AND the part numbers in slots 3 and 4 match. In fact, it is impossible to configure 4 sticks satisfactorily if they are not all the same size. (it will function but not at full speed, although one person found an exception to this by getting RAM from OWC that had blank part numbers recorded in the sticks.)
 
Last edited:
  • Like
Reactions: theorist9

phr2

macrumors newbie
Dec 8, 2019
9
8
Well, the requirement for matching part numbers in each memory channel is present in Intel Datasheet even for 8-th generation CPUs:

"Memory down of all technologies (DDR4/LPDDR3) should be implemented homogeneously, which means that all DRAM devices should be from the same vendor and have the same part number. Implementing a mix of DRAM devices may cause serious signal integrity and functional issues."
"The DRAM device technology and width may vary from one channel to the other."


The difference is, that 2019 iMACs don't enforce this requirement properly, while 2020 iMACs do.

Therefore, configurations like 4,16,4,16 combining in both channels base Apple RAM (4GB single rank) with larger modules (16GB dual rank) are by principle unsupported by Intel no matter if it's 2019 or 2020 iMAC.

Configurations like 16,16,4,4 are supported in Flex mode. That will hovewer operate dual-channel only for part of RAM present in both channels (4,4,4,4) i.e. 16 GB in this example, while the remaining 24 GB will be operated in single channel mode.


Snímka obrazovky 2023-05-07 o 8.24.40.png


For full dual-channel mode, you need to populate both channels with the same memory amount.

Full performance is ensured, if you have e.g. 16GB of memory from Vendor X in Channel A (slots 1,2) and 16GB of memory from Vendor Y in Channel B (slots 3,4) but you should never put different part numbers in the same channel (neither on 2019 iMAC).

To sum up:

- if you only use 2 DRAM modules, put them into DIMM0 slots of both channels (slots 2,4 on 2020 iMAC).
- if you use 4 DRAM modules, buy them in pairs and put first pair into slots 1,2 and second pair into slots 3,4
 
Last edited:
  • Like
Reactions: wilberforce

Xylopyrographer

macrumors newbie
Jul 20, 2024
3
0
Adding to the info pool...

Upgraded a 2020 iMac 27" i7 from 64GB to 128GB.

In the 64GB configuration, slots 1 and 3 each had a 32GB stick from the same mfg with the same P/N (call this mfg A). Total memory reported as 64GB at a speed of 2667 MHz.

Installed the two new 32GB sticks from a different mfg with their same P/N (call this mfg B) into slots 2 and 4. Total memory reported as 128GB at a speed of 2133 MHz.

Removed the sticks from mfg A and moved the sticks from mfg B into slots 1 and 3. Total memory reported as 64GB at a speed of 2667 MHz.

Reconfigured with mfg A in slots 1 and 2, mfg B in slots 3 and 4. Total memory reported as 128GB at a speed of 2400 MHz.

Didn't try mfg B in slots 1 and 2, mfg A in slots 3 and 4.

Quite the game of whack-a-mole.
 

theorist9

macrumors 68040
May 28, 2015
3,880
3,059
If pair X consists of matching sticks X1 and X2, and pair Y of matching sticks Y1 and Y2, should it be configuration I or configuration II:

CONFIGURATION I
BANK 0/ChannelA-DIMM0: stick X1
BANK 1/ChannelA-DIMM1: stick X2
BANK 2/ChannelB-DIMM0: stick Y1
BANK 3/ChannelB-DIMM1: stick Y2

OR:

CONFIGURATION II
BANK 0/ChannelA-DIMM0: stick X1
BANK 1/ChannelA-DIMM1: stick Y1
BANK 2/ChannelB-DIMM0: stick X2
BANK 3/ChannelB-DIMM1: stick Y2

Configuration II, because then Channel A matches Channel B.

To sum up:

- if you only use 2 DRAM modules, put them into DIMM0 slots of both channels (slots 2,4 on 2020 iMAC).
- if you use 4 DRAM modules, buy them in pairs and put first pair into slots 1,2 and second pair into slots 3,4
So @wilberforce says Configuration II is correct, while @phr2 says Configuration I is correct.

I'm not sure who is right.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.