32GB 3600MHz single rank RAM - performance loss?

[TheRandomGuy7]

I just realised my 2x16GB set of Crucial Ballistix CL16 3600 RAM is single rank. I'm using a 5800x at 1440p so I'm worried I'm losing performance from this, in some benchmarks I've seen the hit could be pretty significant?

Should I try switch to a dual-rank 2x16GB kit, or maybe 4x8GB single-rank instead. Or alternatively would I be able to compensate by overclocking to 3800MHz or 4000MHz? (from what I gather single-rank is easier and more stable to OC)

I do have a spare 2x8GB CL16 3200MHz single-rank kit, would I get better performance adding 2 more 8GB sticks and using these (4x8gb single-rank) instead?

 

[EsaT]

Only certain dual rank 16GB DIMMs are those with Samsung B-die.
Meaning likes of 3200MHz CL14 and 3600MHz 16-16-16.

No point in changing to 3200MHz DIMMs.
Singly rank 3600MHz is equally fast in gaming as dual rank 3200MHz and in other areas that higher memory bandwidth and InfinityFabric bus speed from 3600MHz give advantage.

 

[TheRandomGuy7]

Only certain dual rank 16GB DIMMs are those with Samsung B-die.
Meaning likes of 3200MHz CL14 and 3600MHz 16-16-16.

No point in changing to 3200MHz DIMMs.
Singly rank 3600MHz is equally fast in gaming as dual rank 3200MHz and in other areas that higher memory bandwidth and InfinityFabric bus speed from 3600MHz give advantage.

I'm looking at swapping out to this RAM kit which is CL16, 3600MHz, Samsung B-die and should be dual-rank -- so it ticks all the boxes. Do you know about this RAM and if it's guaranteed to be dual-rank?

It's well worth the swap to me as I got my current set at a fantastic price and a friend would take it for exactly what I paid.

 

[Minstadave]

I'm looking at swapping out to this RAM kit which is CL16, 3600MHz, Samsung B-die and should be dual-rank -- so it ticks all the boxes. Do you know about this RAM and if it's guaranteed to be dual-rank?

It's well worth the swap to me as I got my current set at a fantastic price and a friend would take it for exactly what I paid.

Samsung B-die can only be dual rank in a 16GB mobule config.

 

[TheRandomGuy7]

Ah excellent thank you. I'd never read that before, only that samsung B-die is the 'best'

 

[Joxeon]

For gaming you will only really notice a difference at 1080p while at 1440p HW unboxed found the difference between dual and single rank 3600mhz ram to be 0.7%.

 

[EsaT]

I'm looking at swapping out to this RAM kit which is CL16, 3600MHz, Samsung B-die and should be dual-rank -- so it ticks all the boxes. Do you know about this RAM and if it's guaranteed to be dual-rank?

Samsung B-die is propably the oldest DDR4 chip still in production and limited to 8 Gbit capacity per chip.
That means 16GB DIMM needs 16 of them.
And with single chip having 8 bits wide bus that needs splitting DIMM into two ranks to fit into 64 bit wide memory bus.

 

[TheRandomGuy7]

For gaming you will only really notice a difference at 1080p while at 1440p HW unboxed found the difference between dual and single rank 3600mhz ram to be 0.7%.

I've watched a lot of videos and there are widly different results out there, but even at 1440p it seems the difference could be still up to 5% depending on the game. Personally now I have a 5800x I want the peace of mind that it isn't being bottlenecked whatsoever by something silly like mobo or ram.

Samsung B-die is propably the oldest DDR4 chip still in production and limited to 8 Gbit capacity per chip.
That means 16GB DIMM needs 16 of them.
And with single chip having 8 bits wide bus that needs splitting DIMM into two ranks to fit into 64 bit wide memory bus.

Thanks that makes sense. I've bought a sibling TeamGroup kit which is rated at 3200MHz instead but with lower timings (as the 3600 kit increased in price).

From my understanding, while it loses 400MHz in frequency, the timings move down to 14-14-14-31. Therefore due to lower latency this should be pretty much identical performance-wise to the CL16 3600MHz, PLUS I should be able to OC to, say 3800 at CL16. Please do let me know if I'm completely wrong, otherwise I'm sorted and will get to using my new kit.

 

[Joxeon]

've watched a lot of videos and there are widly different results out there, but even at 1440p it seems the difference could be still up to 5% depending on the game. Personally now I have a 5800x I want the peace of mind that it isn't being bottlenecked whatsoever by something silly like mobo or ram.

You should be able to run 3800/16 providing the CPUs memory controller is decent although it would be interesting if you could post some fps results for the dual rank B-die vs crucial kit which also should be able to do 3800/16.

 

[TheRandomGuy7]

You should be able to run 3800/16 providing the CPUs memory controller is decent although it would be interesting if you could post some fps results for the dual rank B-die vs crucial kit which also should be able to do 3800/16.

Yeah the crucial kit is fine and would probably OC decently well. But the single rank vs dual rank alone should in theory make a pretty notable difference.

I'll do a few synthetic tests and find some games with built-in benchmarks before the new kit arrives and then compare to the new RAM at stock XMP values (3200MHz, CL14) to see if the dual-rank makes any difference. Even if not then I'd just OC it to be faster than the crucial anyway.

 

[EsaT]

From my understanding, while it loses 400MHz in frequency, the timings move down to 14-14-14-31. Therefore due to lower latency this should be pretty much identical performance-wise to the CL16 3600MHz, PLUS I should be able to OC to, say 3800 at CL16. Please do let me know if I'm completely wrong, otherwise I'm sorted and will get to using my new kit.

While 14 cycles at 3200MHz (~8,75 ns) is actually tiny bit faster timing than 16 cycles (~8,89 ns) at 3600MHz, CPU's internal InfinityFabric bus runs at equally much lower clock lowering its bandwidth/increasing its latency.
Also communication between cores in different CCXes/CCDs depends on IF speed.
So overall 3600MHz 16-16-16 is better in more situations.

 

[TheRandomGuy7]

While 14 cycles at 3200MHz (~8,75 ns) is actually tiny bit faster timing than 16 cycles (~8,89 ns) at 3600MHz, CPU's internal InfinityFabric bus runs at equally much lower clock lowering its bandwidth/increasing its latency.
Also communication between cores in different CCXes/CCDs depends on IF speed.
So overall 3600MHz 16-16-16 is better in more situations.

I see, so would I be better off running it at something like 3600 or 3800MHz at CL16 instead? I'm assuming there should be zero stability issues with a basic overclock on a RAM kit as highly acclaimed as this.

 

[EsaT]

I see, so would I be better off running it at something like 3600 or 3800MHz at CL16 instead? I'm assuming there should be zero stability issues with a basic overclock on a RAM kit as highly acclaimed as this.

3600MHz is easy for memory controller and InfinityFabric of (nearly 100% of) Zen2/Zen3 CPUs.
Including manual overclocking, for which 3200MHz CL14 B-die is excellent.
For example settings given by Ryzen DRAM calculator should work well in most cases.
Auto settings in BIOS could give too much various voltages actually preventing stability.

3800Mhz is more demanding and can't be quaranteed/would need lot more manual tweaking.
There's already more than tiny chance that IF of your individual CPU couldn't do it.
And at that resolution wouldn't be worth the effort/time needed.

 

[TheRandomGuy7]

3600MHz is easy for memory controller and InfinityFabric of (nearly 100% of) Zen2/Zen3 CPUs.
Including manual overclocking, for which 3200MHz CL14 B-die is excellent.
For example settings given by Ryzen DRAM calculator should work well in most cases.
Auto settings in BIOS could give too much various voltages actually preventing stability.

3800Mhz is more demanding and can't be quaranteed/would need lot more manual tweaking.
There's already more than tiny chance that IF of your individual CPU couldn't do it.
And at that resolution wouldn't be worth the effort/time needed.

To be honest I just wanted a solid 3600MHz kit to set XMP on and forget about it. But since it ended up a lot cheaper I've now got a créme de la créme overclocking kit - so I may as well at least put a few hours into tweaking it. I've tinkered with and overclocked everything else but never bothered with RAM before.

If I use values from the DRAM calculator and it works out stable should I just use that or try manually OC higher? Motherboard should handle anything with extreme reliability, it's a freshly installed Strix B550-XE.

 

[EsaT]

If I use values from the DRAM calculator and it works out stable should I just use that or try manually OC higher?

No point in that.
Finding tightest settings with stability would fast take really lots of time.

 

[TheRandomGuy7]

it would be interesting if you could post some fps results for the dual rank B-die vs crucial kit which also should be able to do 3800/16.

No point in changing to 3200MHz DIMMs.
Singly rank 3600MHz is equally fast in gaming as dual rank 3200MHz and in other areas that higher memory bandwidth and InfinityFabric bus speed from 3600MHz give advantage.

Pinging you both in case you're interested in the results below

 

[TheRandomGuy7]

So I received the new kit and I did some tests with stock XMP, and then more tests with a manual OC using mostly DRAM calculator values.

All tests were of course done in identical conditions, even down to the ambient temperatures in my room being similar each time.


OLD KIT: Crucial Ballistix, stock XMP (3600MHz, 16-18-18-38, SINGLE-RANK)
NEW KIT: TeamGroup 8pack edition, stock XMP (3200MHz, 14-14-14-31, DUAL-RANK)
NEW KIT OC: 3600MHz, 14-15-15-30, dual-rank

Metro Last Light Redux (benchmark tool, 2560x1440, max everything, 3 runthroughs)
Average Framerate: OLD KIT: 109.67 NEW KIT: 109.67 NEW KIT OC: 110.33
Min. Framerate: OLD KIT: 35.85 NEW KIT: 33.64 NEW KIT OC: 37.26

Civilization VI (Gathering Storm AI benchmark)
Average Turn Time: OLD KIT: 31.22 NEW KIT: 31.14 NEW KIT OC: 30.52

Warhammer II: Total War
Battle benchmark: OLD KIT: 101.3fps NEW KIT: 101.7fps NEW KIT OC:106.4fps
Campaign map benchmark: OLD KIT: 91.8fps NEW KIT: 93.4fps NEW KIT OC: 96.8fps
Skaven benchmark: OLD KIT: 91.7fps NEW KIT: 93.9fps NEW KIT OC: 97.7fps

3DMark Timespy
Timespy score: OLD KIT: 15912 NEW KIT: 15968 NEW KIT OC: 16102
CPU score: OLD KIT: 11341 NEW KIT: 11728 NEW KIT OC: 12150


Performance is pretty much on par with the sort of results reviewers were getting. I noticed I don't actually own any newer games with built-in benchmarks so I did a selection of older titles. Metro proved to be a pretty useless comparison but Civ VI and Warhammer II should be decent indicators of RAM performance
since they're both heavily CPU bound. Newer titles would surely see a greater gap open up, but these results are still interesting.

Results are very similar between the kits at stock settings, though the dual-rank did take a small lead of up to 3% in the more meaningful CPU-bound tests. This suggests negligible improvement due to dual-rank alone --although if Zen 3 prefers 3600 over 3200 this could have had an impact. The gap would also be greater at 1080p and perhaps greater still had I tested recent AAA games.

After OCing the new kit the improvement from the Crucial kit becomes significant, the sort of gains I was aiming for with the switch. Timespy CPU score jumped 7%, the Total War battle benchmark increased 5%, while even Metro saw a nice boost to the minimum fps floor.

All in all, youtubers seem to be correct that dual-rank vs single-rank will only yield a <5% improvement at 1440p and in some cases the difference is non-existent. Switching to dual-rank combined with a lower CAS of 14 has yielded very good improvements.

 

[Joxeon]

@TheRandomGuy7 So it appears the overclock added more than dual rank alone, did you try the best OC possible on the crucial kit which should also be able to do 3600/14 and possibly even 3800/14 if the Depending on the memory controller?. Interesting results Though.

 

[TheRandomGuy7]

@TheRandomGuy7 So it appears the overclock added more than dual rank alone, did you try the best OC possible on the crucial kit which should also be able to do 3600/14 and possibly even 3800/14 if the Depending on the memory controller?. Interesting results Though.

I got rid of the Crucial kit immediately since I was able to recoup all the costs. I didn't get round to testing any OC on that kit as I really didn't have the time. From the results I did get, it seems that the Crucial would perform similarly if I managed to get it stable at 3600/14 -- however the ranking alone would certainly have some effect of up to a few %.

I could definitely squeeze a lot more performance out of this TeamGroup kit but at the moment I just want 3600 C14 and perfect stability. It seems to require at least 1.44v to run properly at 3600 C14, and although stable in games it does get some errors in OCCT. I don't really want to run it as high as 1.5v and definitely not higher than that under any circumstances. Not really sure if voltage is the only thing I should be tweaking when getting mem errors. Also not sure if 3800/ C16 would be easier to get stability with. I'm completely new to all this so I might have to make a separate post for OC help.

 

[Joxeon]

I had to stick 1.49v through my 2x8 3200/14 kit before it was error free when running it at 3800/14 although I've now changed it to 3800/16 @1.42v as there was no real difference to fps at 1440p with a 3080 and with the card pumping out so much heat compared to the 1070ti I had previously it would cause the occasional crash as B-die is quite temperature sensitive.