Why people say 2x16 is better than 4x8 even if it is faster in gaming?

[foxbill86]

Why people say 2x16 is better than 4x8 even if it is faster in gaming? 

 

I have asked this before a long time ago (wanted a 4x8 than a 2x16) that having 4 dimms will increase latency cause the signal would travel from A2 to A1 (same on the B side) before ending the sequence and thus having more latency. So I was like "oh okay, makes sense why people always use dual instead of quad and MOST consumer CPUs are dual channel anyway"

But I saw this video a while ago by ancientgameplays. So in short, still. 4 dimms is better than 2? I thought it would increase latency but shits faster. What gives? 

 

[Tan3l6]

Most platforms support 2x RAM better than 4x. More so with AMD CPUs.

In my case - A2 + B2 work as intended while A1 + B1 does not basically work at all. 

4 slots together work at very low speeds, if at all. Currently DDR5.

Spoiler

I have AMD platform now, previous Intel platform was much better considering 4 sticks - had no issues actually  (DDR4).

[YoungBlade]

If you can get them to run at the same speed and timings, then yes, 4 DIMMs is better than 2. Always has been.

 

The issue for DDR5 is that that's a big IF.

 

At the moment, memory controllers struggle when you run 4 sticks. You might not be able to achieve DDR5-6000 CL30 on all 4 and might have to settle for DDR5-5600 or 5200 speeds. In that case, any benefit goes out the window.

 

The risk of not being able to run the RAM at XMP/EXPO speed is why it's not recommended.

[starsmine]

what Im more shocked by is how "well" single stick is performing there. 

[WereCat]

More ranks

 

If the 16GB sticks were the same amount of ranks as the 8GB sticks then more sticks = more bandwidth

[Hinjima]

4 minutes ago, starsmine said:

what Im more shocked by is how "well" single stick is performing there. 

Thats because a single stick of DDR5 actually runs in dual channel mode.

In DDR5, each memory module is essentially built with dual-channel capabilities, so a single stick can still utilize both channels. However, the bandwidth is limited compared to using two sticks.

[porina]

Assuming same speeds and timings. could be something in the ranks, but it's complicated.

 

In DDR5 era, 8GB modules are probably 1Rx16 which is another way of saying junk performance. 16GB modules are likely 1Rx8 which is kinda baseline performance. 32GB are 2Rx8 which can give a performance advantage over 1Rx8 in memory intensive workloads. 4x 8GB configuration results in 2R/channel but poor ones, but maybe that is still enough to overtake regular 1Rx8. 

 

In DDR4 era, where it was easier to get 4 sticks running, they changed over lifespan. Early 8GB modules were 2Rx8, dropping to 1Rx8 and eventually to 1Rx16 towards the end of its life. 16GB modules were 2R but later ones were 1R. I think you have to go 32GB modules to guarantee 2R, by which point it can be easier and cheaper to go 4 sticks of 1Rx8.

 

Below is old testing I did with DDR4. 11700k, 1080 Ti, 1080p. Minor variations in gaming. I could have done with a faster GPU at the time so it might be more GPU limited. Look at memory heavy compute below for a bigger difference. I did set all ram to 2133 MT/s. Didn't note latency but whatever JEDEC mid bin is.

 

[emosun]

2 dimms = less to go wrong

 

Any computer that works will outrun any computer that doesn't 

[xg32]

4x8 is better only for ddr4, and i have to have direct airflow on it, the small difference isn't worth it, would have went with 2x16 if i had to pick again (or 2x32 for ddr5)

 

also by minmaxing performance it had to be the binned b-die, which is still going strong, but i'm not sure how much longer they'll last compared to 2x16 which runs much cooler (0/8 bricked so far)

[Tetras]

47 minutes ago, foxbill86 said:

But I saw this video a while ago by ancientgameplays. So in short, still. 4 dimms is better than 2? I thought it would increase latency but shits faster. What gives? 

 

49 minutes ago, foxbill86 said:

Why people say 2x16 is better than 4x8 even if it is faster in gaming? 

As mentioned already, the main reason 2x16 (or more specifically: 2 sticks, so this also applies to 2x32) is recommended is not due to their performance, but that you're less likely to have problems when trying to run them at their rated speed. More sticks and/or more memory ranks is harder to run. You can also largely replicate the performance of 4x single rank x8 modules (like 4x16) by running 2x dual rank x8 modules (like 2x32).

[freeagent]

For DDR4, 2x 32GB is dual rank, so it would be as fast as 4x8 would at the same speed and timings, 2x 16GB early sticks were also 2R, but later models were 1R. 2R is better than people give it credit for, they just have to remember that there is more to computing than just playing games, which are a relatively low load for a computer system, minus the GPU.

 

Basically, anything that is not C14 B-Die, is low end. And later models with the super loose timings, are just shit for benchmark performance, but totally fine for daily life.

[Mark Kaine]

1 hour ago, foxbill86 said:

So in short, still. 4 dimms is better than 2? 

It's 1% faster? And who says latency doesn't suck?

Also excuse me, "DX11nklvmerror, your system has rebooted due to memory controller breakdown the 20th time today "

 

4 sticks is *not* better, especially on Ryzen. ¯⁠\⁠_⁠(⁠ツ⁠)⁠_⁠/⁠¯

[freeagent]

10 minutes ago, Mark Kaine said:

4 sticks is *not* better, especially on Ryzen. ¯⁠\⁠_⁠(⁠ツ⁠)⁠_⁠/⁠¯

It is. Just because you didn't try, or couldn't get it to work doesn't mean it isn't better.

 

On Zen 3 4 sticks ran just fine for me, just as easy as 2 sticks. overclocking to 1900 1:1 was not a problem, and my 58X3D will do 4 sticks at 1933 1:1.

 

I see guys running 4 sticks of DDR5 at 6000 meanwhile I cant get anything past 4800 stable.

[porina]

26 minutes ago, freeagent said:

I see guys running 4 sticks of DDR5 at 6000 meanwhile I cant get anything past 4800 stable.

Don't know what changed, but I've noticed more "high speed" 4 module kits advertised now. It's not that high in absolute terms, but high for 4 sticks, like around 6000. Is there a new chip out, are enthusiast module makers getting better at tweaking timings, or something in recent AGESA?

 

If I hate myself, I have "same" DDR5 kit 2x32 + 2x16 I could throw into one system and see what happens. Like AMD post times aren't slow enough already.

[StDragon]

14 minutes ago, porina said:

Like AMD post times aren't slow enough already.

That's only for the initial DIMM training at POST upon first enumerating new modules. It's always been that way for the Zen series AFAIK. I think it's part of the AGESA code which makes this known process so prevalent.

[podkall]

50 minutes ago, freeagent said:

It is. Just because you didn't try, or couldn't get it to work doesn't mean it isn't better.

 

On Zen 3 4 sticks ran just fine for me, just as easy as 2 sticks. overclocking to 1900 1:1 was not a problem, and my 58X3D will do 4 sticks at 1933 1:1.

 

I see guys running 4 sticks of DDR5 at 6000 meanwhile I cant get anything past 4800 stable.

AM4 is different from AM5

[freeagent]

Training 4x 16 DDR5 take probably close to 5 minutes lol. Change VSOC, mem voltage, timings, anything memory related including its subsystems is a painful experience.. especially if you let it try settings more than once lol..

 

Just brutal..

[freeagent]

Just now, podkall said:

AM4 is different from AM5

I have both. OP is talking about DDR4.

[podkall]

1 minute ago, freeagent said:

I have both. OP is talking about DDR4.

uh-huh, DDR4 + 7500F, see it many times:

 

 

[freeagent]

1 minute ago, podkall said:

uh-huh, DDR4 + 7500F, see it many times:

 

 

Is there a problem?

[starsmine]

12 minutes ago, freeagent said:

Is there a problem?

7500F is only on AM5
AKA DDR5, not DDR4

[freeagent]

2 minutes ago, starsmine said:

7500F is only on AM5
AKA DDR5, not DDR4

I wasn't the only one talking about DDR4, I was expanding on what they were saying. I also kicked in some of my short experience with DDR5.

[podkall]

18 minutes ago, freeagent said:

Is there a problem?

relevancy?

 

 

 

 

 

 

• It's like if OP was asking "are 95C temps fine?" - others say they aren't, but you jump in and say they're fine because AM5 CPUs can stay pinned at 95C (but OP in this theoretical post has 13th/14th gen Intel CPU with outdated BIOS)

[porina]

24 minutes ago, StDragon said:

That's only for the initial DIMM training at POST upon first enumerating new modules. It's always been that way for the Zen series AFAIK. I think it's part of the AGESA code which makes this known process so prevalent.

It seems particularly bad on AM5. I don't recall AM4 era taking anywhere near that long.

 

For normal use, MCR makes it tolerable. If tinkering with ram settings, you have to have MCR off and it takes forever to reboot. Adding more ram will make it even slower.

 

My Zen 3 laptop (fast boot disabled, no hibernation, no sleep) is often well on the way to completing loading Windows before my Zen 4 desktop even gets past post with MCR on. Laptop does only have 32GB of ram compared to 64GB on the desktop though.

[freeagent]

Just now, podkall said:

relevancy?

 

 

 

 

 

 

• It's like if OP was asking "are 95C temps fine?" - others say they aren't, but you jump in and say they're fine because AM5 CPUs can stay pinned at 95C (but OP in this theoretical post has 13th/14th gen Intel CPU with outdated BIOS)

I don't know what your problem is, but if you quit focussing on what I said about DDR4, you will see I also spoke about DDR5. 

 

So I fucked up. 

 

Do you rub a puppies nose in the carpet when he pisses? If you do, that makes you an asshole.