Navigation

TLDR: 8x is better than 16x, because the system can access more data at the same time

2021 - HyperX Impact DDR4 HX432S20IBK2/32
2023 - Kingston KF426S15IB1K2/32

Yes - Up to 1300 points in Timespy, which is a significant number. The exact gain will be highly dependent on what ram your model came with. Generally, the smaller the ram that is being replaced, the bigger the impact.
In one case, going from 16GB “slow” stock ram to HyperX Impact DDR4 HX432S20IBK2/32, Timespy reported 1300 point gain.

After seeing a lot of confusion here about 1×8 and 1×16 and people thinking it has to do with single and dual rank. This specifically has nothing to do with ranks but everything to do with how a x8 device is configured and how a x16 device is configured within a DIMM.
In the video, Linus touched upon it but I think didn’t explicitly state the technical exact reason for the difference in performance. He was very close to it, and I think gave a sufficient explanation for it but here I’ll try to elaborate if it interests you.
You can lay out the total capacity for a DIMM in various ways really, in this case you have two ways to achieve 8GB of total capacity (note the bit to byte conversion):

1 rank of 8 chips x 8Gb in density = 64Gb total density = 8GB total density
1 rank of 4 chips x 16Gb in density = 64 Gb total density = 8GB total density

The nomenclature for the device width is related to the total amount of data pins on a channel. DDR4 channels are typically 64-bits wide, so if you have a:

x8 device- it means a single rank within the channel has exactly 8 chips to make-up 64 data pins
x16 device- it means a single rank within the channel has exactly 4 chips to make-up 64 data pins

Now with the explanation out of the way, you can now go deeper into the architectural layout typically found in devices in DDR4:

x8 device has a total of 16 addressable banks (it has 4 Bank Groups and 4 Banks per Bank Group) with a 1KB page size [each page a bank is allowed to open at any given time]
x16 device has a total of 8 addressable banks (it has 2 Bank Groups and 4 Banks per Bank Group) with a 2KB page size

Having less banks available to the MC means you have less resources to interleave and lower chances of extracting the full BW available according to the rated maximum bus speed. The analogy Linus gave about the librarian searching for a book makes total sense with this in mind.

Now the next thing to affect performance is the page size:
Typically the page size usually dictates tFAW (Four Activate Window Time), so it’s usually worse for 2KB than it is for 1KB by several nano-seconds which can affect performance [need to look up data sheets for a given device by the given manufacturer to look up exactly what they are].
Anyway as for why this happens, this is directly tied to memory advancements really, over time memory chip suppliers make denser chips on smaller nodes. As times goes on they will have more of one kind than the other.
Given the density of the DIMM here is 8GB, that means as memory gets more advanced, they can make more wafers to supply everyone in the world (and a bonus is that its cheaper to include less chips on a module to sell).

TLDR: 8x is better than 16x, because the system can access more data at the same time

Make sure the default/JEDEC/PnP profile is DDR4-3200. It is not possible to use XMP or otherwise change memory settings in BIOS.
Also, prefer dual rank, 8bit device width (2Rx8) RAM or Single Rank, 8bit device width (1Rx8) Ram.

No, any DDR4 SO-DIMM 1.2V can be installed. A 64 GB kit has been tested, but 128 GB should also work, even with Windows Home