Ryzen Above: Best Memory Settings for AMD's 3000 CPUs, Tested

[mitch074]

I just double checked, and the timings are all definitely still set to "Auto" in the BIOS. The motherboard I'm using is actually one that has been reviewed here, a Gigabyte Aorus Eilte - one of the cheapest X570 mobo's around.
https://www.tomshardware.com/reviews/gigabyte-x570-aorus-elite-atx-motherboard,6252.html

At one time, motherboards wouldn't have XMP profiles to work with - I'm also not sure how XMP would work there, because 1T/2T depends on the total amount of ranks you have.
Long story short, 1 rank per channel usually defaults to 1T, anything more than that (2, 3, up to 4 ranks possible!) the motherboard could switch to 2T. Some kits and motherboards might allow you to go back to 1T with dual rank, but motherboard makers usually try to play it safe with non-simple configurations.
From my (DDR1) experience, 1T on 3 ranks or more usually doesn't work.

 

[sizzling]

Did you click the link I showed you? It's about 2 rank vs 4 rank on Intel. It's from 2017. It's a follow-up to another review that showed the same pattern.

Edit: Come to think of it, could you imagine the wailing and gnashing of teeth had we titled this article "NO, AMD does NOT meaningfully benefit from faster RAM: Get 4 Ranks Instead!" There would be blood...

My intention is not to argue with anyone but it was you that took my post and made it about RAM speed. You seem to be on an attack.

 

[Xolhid]

I don't know why everyone is saying that the mclk and fclk automatically decouple after 3600mhz(1800mhz). Even saying it is the "sweet spot". The is completely untrue. It will only decouple if mclk does not match the fclk frequency. Please fix this in your review.

There are some people who are lucky to get 3800mhz(1900mhz) 1:1 stable. I myself was only able to hit 3733mhz(1866mhz) 1:1 stable. About a 2-3% performance improvement over 3600mhz(1800mhz) for my hardware silicon lottery.

 

[Crashman]

I don't know why everyone is saying that the mclk and fclk automatically decouple after 3600mhz(1800mhz)...It will only decouple if fclk is not manually adjusted to match the mclk frequency. Please fix this in your review

So it decouples automatically and it can be fixed by manually setting FCLK? That's what we told people
CPU-Z reported "northbridge frequency" at half the memory clock frequency until we manually made that adjustment.

 

[JamesSneed]

Thanks for this writeup. As someone that has a Gen 1 Ryzen 1800x paired with G.skill Flarex 3200mhz C14 memory that is considering upgrading to Ryzen 3000 down the road I know exactly what i'm going to do first which is buy another two sticks of memory. Long live the x370 The 3200mhz CAS 14 memory is right with the fastest when 4 sticks are used at 1T and I know the Flarex b-die kits will hit it. Thanks again.

 

[webjapm]

Thanks for this writeup. As someone that has a Gen 1 Ryzen 1800x paired with G.skill Flarex 3200mhz C14 memory that is considering upgrading to Ryzen 3000 down the road I know exactly what i'm going to do first which is buy another two sticks of memory. Long live the x370 The 3200mhz CAS 14 memory is right with the fastest when 4 sticks are used at 1T and I know the Flarex b-die kits will hit it. Thanks again.

You'll tell us how you're doing with the 4 modules.
I also have flarex 3200 cl14.

 

[alextheblue]

So it decouples automatically and it can be fixed by manually setting FCLK? That's what we told people
CPU-Z reported "northbridge frequency" at half the memory clock frequency until we manually made that adjustment.

The charts AMD put out seem to imply it won't automagically decouple until you exceed 3733.

Going by that, it would seem you don't need to manually override anything until you exceed 3733... maybe some mainboard manufacturers BIOS revisions changed this? Anyway yeah I've heard some people can run stable 1:1 at 3800 but thus far it doesn't seem widespread. I agree that 3600 is the highest rated RAM I'd consider due to cost.

 

[alextheblue]

Ryzen 1/2xxx was really dependent on RAM quality to work well, while Ryzen 3xxx just works well with any RAM, same as Intel.

Yeah, Zen and Zen+ definitely benefited quite a bit more from memory speed. Zen 2, not so much... just get a "decent" kit on the cheap and you'll be within spitting distance of more expensive setups. I'd still prefer 4 ranks, if you can do so and still hit 1T, but even then it's not as drastic a difference in most software as it was with older Zen.

Obviously that doesn't apply to the upcoming bandwidth-hungry APUs - Zen 2 mobile APUs in particular are going to get a huge boost from LPDDR4X... if the manufacturers actually utilize decently high clocked RAM! So easy to hamstring an APU-powered laptop by penny pinching on RAM. Like the older AMD APUs where some OEMs would sell the models with the most CUs in single channel config, in some cases even using single channel only boards.

 

[Crashman]

The charts AMD put out seem to imply it won't automagically decouple until you exceed 3733.

Going by that, it would seem you don't need to manually override anything until you exceed 3733... maybe some mainboard manufacturers BIOS revisions changed this? Anyway yeah I've heard some people can run stable 1:1 at 3800 but thus far it doesn't seem widespread. I agree that 3600 is the highest rated RAM I'd consider due to cost.

Easy answer: The chart was wrong. There were also errors in its chipset block diagram. We've pointed these things out a couple times and AMD updated some of its documents.

Edit: I should point out that some of these small errors were due to changes made in core firmware after the original charts were produced. CPU tester Paul had fits over the 3733/3600 decoupling speed change when he was writing up the first review.

 

[StefanoAkebano]

The point I think is that if you get 2x16GB you can run the optimal four ranks at 1T, as opposed to grabbing 4x8GB which you will need to run at 2T - the article shows there is a small but real difference between 1T and 2T. Therefore, there should be a slight performance difference between these two options (not to mention that 4x 8GB sticks generally cost more than 2x 16GB sticks), right? I'm not sure, but it makes sense.

Hi.
Nice article, sums current knowledge about Ryzen 3000 and memory.

But is missing one thing: difference between T-Topology and Daisy Chain boards.
Its quite well explained here:

View: https://www.youtube.com/watch?v=3vQwGGbW1AE


and here:
https://www.techpowerup.com/review/amd-ryzen-memory-tweaking-overclocking-guide/4.html

and here:
https://www.tomshardware.com/news/amd-ryzen-3000-cpus-memory-recommendations-asrock,39756.html

TL;DR:
As a rule of a tumb, its easier on X570 boards (I think all are Daisy Chain) and a lot of X470 boards to run 2x16 GB (2 sticks of dual rank) with higher speed/tighter timings then than 4x8gb (4 sticks of single rank). It shouldn't matter if its 4x8GB or 2x16GB on T-Topology motherboards.

 

[Crashman]

Hi.
Nice article, sums current knowledge about Ryzen 3000 and memory.

But is missing one thing: difference between T-Topology and Daisy Chain boards.
Its quite well explained here:

View: https://www.youtube.com/watch?v=3vQwGGbW1AE


and here:
https://www.techpowerup.com/review/amd-ryzen-memory-tweaking-overclocking-guide/4.html

and here:
https://www.tomshardware.com/news/amd-ryzen-3000-cpus-memory-recommendations-asrock,39756.html

TL;DR:
As a rule of a tumb, its easier on X570 boards (I think all are Daisy Chain) and a lot of X470 boards to run 2x16 GB (2 sticks of dual rank) with higher speed/tighter timings then than 4x8gb (4 sticks of single rank). It shouldn't matter if its 4x8GB or 2x16GB on T-Topology motherboards.

It's not missing, it doesn't belong here. This isn't a motherboard selection guide for extreme memory overclocking.

There's the limit for CPU I/O synchronization that tanks performance past DDR4-3600 unless you manually overclock a couple things, and then most users are stuck with no more than DDR4-3733 anyway. We could try to overshoot the handicap with ultrahigh memory data rates, but that would be material for an extreme overclocking guide, which this is not.

 

[webjapm]

So according to the tests of this article, 4 single rank modules or 2 double rank modules are better than 2 single rank modules using ryzen 3000.

 

[Crashman]

So according to the tests of this article, 4 single rank modules or 2 double rank modules are better than 2 single rank modules using ryzen 3000.

Correct.
It's a pattern we noticed repeatedly on our old Intel-based memory review platforms being repeated by AMD. AMD was first to get the full set of numbers.

 

[Ninjawithagun]

Traditionally when running 4 sticks in 2 channels, you are usually forcing the command rate to 2T rather than 1T, making the ram latency worse.

Heres what the article says:
"Command rate is the number of cycles a command must be presented before a memory operation commences, so that 2T incurs an additional cycle of latency to every memory access. It makes sense that most people would shoot for 1T, except that higher data rates often make this unworkable, particularly when more ranks of memory are used (ie, two “double-sided” or four “single-sided” DIMMs). The reason that we’ve used only half of our four-DIMM kit thus far is that it won’t run all four DIMMs at DDR4-3600 and 1T, hence you’ll want to see the penalty of 2T before we add four-DIMM configurations to the mix."

I see no way for 4 sticks to be faster as they have no more bandwidth than 2 sticks. If you look at the far right 2, you see the performance is roughly the same, which is very surprising. I have seen 4x8gb would have less compatability with ryzen in the past, but not sure of now.

Ironically, as the newer BIOS AGESA versions are released, so are advantages and improvements to the DDR4 configurations that can be ran with Ryzen. For example, when AGESA 1.0.0.3 AB was released, I was suddenly able to run @ 1T with all four DIMMs occupied and the FSB was running 1:1 @ 3600 with CL16 @ 1.35V with my old Corsair Vengeance RGB memory. I have since upgraded to the G.Skill Z Neo DDR4 3600Mhz CL14 1.4V kit (only running 2 x 8GB) and have NOT regretted it one bit. Amazing memory, but is expensive.

 

[NightHawkRMX]

I will agree agesa updates have done a lot to benefit Ryzen.

Memory compatability is a big thing. Boost behavior is a big thing agesa uodates have been changing lately.

 

[maddogfargo]

Can't believe I missed this article.

If you are testing with DDR4 3200 CL 16 and DDR4 3600 CL 18...you are testing with the wrong RAM. Most informed enthusiasts are going to go for higher quality memory, especially now that prices have improved.

High performance Samsung B-Die or similar really benefits Ryzen and should spec out around DDR4 3200 CL 14 and DDR4 3600 CL 16. They also usually have better sub timings.

Not to mention they are easy to OC & tighten the timings for even better performance.

 

[Crashman]

Can't believe I missed this article.

If you are testing with DDR4 3200 CL 16 and DDR4 3600 CL 18...you are testing with the wrong RAM. Most informed enthusiasts are going to go for higher quality memory, especially now that prices have improved.

High performance Samsung B-Die or similar really benefits Ryzen and should spec out around DDR4 3200 CL 14 and DDR4 3600 CL 16. They also usually have better sub timings.

Not to mention they are easy to OC & tighten the timings for even better performance.

If only you'd read the article before commenting: It compares latency while controlling for frequency, and it compares frequency while controlling for latency. Look up the term "scientific method" for the reasons it was done this way.

 

[Chuck017]

So why when I look at my motherboard manual does it tell me that I can have a higher clock speed with single rank RAM? I have a Matisse board, look at link below under memory specifications.

https://www.asrock.com/mb/AMD/X570M Pro4/index.us.asp#Specification

 

[Chuck017]

So why when I look at my motherboard manual does it tell me that I can have a higher clock speed with single rank RAM? I have a Matisse board, look at link below under memory specifications.

https://www.asrock.com/mb/AMD/X570M Pro4/index.us.asp#Specification

Matisse CPU* sorry, and I'm running Ryzen 3950x with G.Skill 3600 RAM, 64GB (4x16)

currently can't operate monster hunter: world over default (~1863 MHz) without crashing after about an hour or two; but all other games seem to be okay, if I change clock rate on RAM in BIOS then I get different behavior in MHW.

 

[Crashman]

Matisse CPU* sorry, and I'm running Ryzen 3950x with G.Skill 3600 RAM, 64GB (4x16)

currently can't operate monster hunter: world over default (~1863 MHz) without crashing after about an hour or two; but all other games seem to be okay, if I change clock rate on RAM in BIOS then I get different behavior in MHW.

Did youtry leaving all the other BIOS settings to "default" and just enabling XMP?

 

[Chuck017]

Did youtry leaving all the other BIOS settings to "default" and just enabling XMP?

Trying this out today, still curious as to why the board manual prefers single rank RAM.

 

[Crashman]

Trying this out today, still curious as to why the board manual prefers single rank RAM.

Because it's easier for the memory controller to address. One rank per channel gets you more stability headroom than two, which is why you can use faster memory in that configuration.

You have four dual-rank modules, which is the hardest for your memory controller to address. On the other hand, you'll get better performance at any particular data rate compared to fewer ranks.

 

[Chuck017]

Thanks! Monster hunter world still crashing (nothing else does so it may just be the game, still need to checkout all my drivers too its a fresh build) but glad I have an answer. I may change out the RAM as this specific RAM model number was not on the QVL list, maybe single rank will be less demanding and fix.

 

[Crashman]

Thanks! Monster hunter world still crashing (nothing else does so it may just be the game, still need to checkout all my drivers too its a fresh build) but glad I have an answer. I may change out the RAM as this specific RAM model number was not on the QVL list, maybe single rank will be less demanding and fix.

Well then you lose capacity. The top capacity for mainstream ICs is 8Gb (eight gigabits), so you end up having 8GB per rank.
The new higher-density 16Gb chips aren't being used on single-sided modules yet, so you're not finiding 16GB single-rank DIMMs. And the reason is that memory companies think you're better off with 16GB dual-rank DIMMs.

There is another option, two 32GB DIMMs, using the higher-density ICs. You currently have four ranks per channel (two dual-rank DIMMs). Using a pair of 32GB DIMMs, you could drop from four ranks per channel to two per channel, without reducing capacity. It probably gives you a bit more OC room before the memory controller <Mod Edit> itself. BUT!

The CPU you're using should be fine with four ranks per channel. It's much solider than its predecessors. I think you probably have a different problem.

 

[Chuck017]

Right, I haven't been able to find ANY sticks over 8gb that is single rank, we just aren't there yet.

And I have only two ranks per channel, (sorry my initial reply says 4x16 but thats referring to ''# of sticks x amount of memory per stick'') they are 2Rx16, so two ranks per.

I've decided I don't need all 64gb (little overkill anyways?) so I'm going to try with these new four 8gb sticks that I ordered that are actually on the QVL- hyper x 3600. I was not aware that the ASRock x570m board was so particular with the QVL, very limited list. I noticed my RAM model # was off by a single letter, does this matter? Probably not... but it's easy to troubleshoot and also best to stick with manufacturers recommendations. PLUS the model number i DID find on the QVL list (that was off by a letter) should only be done with the 2 stick configuration....as I said I installed 4... and no i haven't tried just two sticks of my current RAM b/c i feel like at that point I'd rather just get new RAM thats on the QVL thats also supposed to be faster.