Wait, to replace actual DDR5 6000-32 (AM5)

[Sarajiel]

Question is,
if i upgrade to new 2x16 (or more) 6000>6400-30, next year when there is new DDR5 out proven to be an evolution over older DDR5, what should i look at in specs ?.. like rank... !? to get RAM that works better out of the box.

You can already get newer and "better" DDR5 in the form of 24MBit chips, aka 2x 24GB memory kits, but that won't let you magically run your system at DDR5-6400, because it mostly depends on the integrated memory controller inside your CPU. Having an 8-layer PCB, which your motherboard has, helps but unless AMD publishes an AGESA version that allows DDR5-6400 on any Ryzen 7000X3D CPU, no fancy memory stick will help you, or allow you to bypass that tedious manual tuning part.
Right now, you can expect that DDR5-6200 will work on most Ryzen 7000 CPUs, but that's about it. If yours is a potato or the motherboard doesn't like the memory sticks, you are stuck at DDR5-6000. Not to mention that a lot of the "overclocker" brands offer EXPO/XMP profiles that assume that your CPU is actually able to run the advertised speed. If it can't, you might be stuck with a kit that doesn't offer a lower tier EXPO/XMP profile, unlike what most Kingston and the better Corsair kits offer.
Generally speaking of DDR5, unless you need the amount of memory, only use two single-rank memory sticks. Also look for sticks that offer more than one speed bin in their EXPO/XMP profiles (e.g. 6000 & 6400) since you are obviously not in the position to manually tune memory yet. Avoid any memory kit that requires your CPU to run in 2:1 mode since that is usually slower than DDR5-6000 in 1:1 mode on Ryzen. This might change at some point, but right now, don't waste money on DDR5-6800/7200/7600/8000 kits.
If you actually want to optimize your memory timings, you'd have to buy a fancy overclocker board with only two DIMM slots, like the ASUS ROG Crosshair X670E Gene. However, that's more for those folks who really like pushing their memory speeds to the absolut maximum for competitive overclocks.

In theory, you could already buy a Fury Renegade DDR5-6400 48GB kit as an upgrade, but you won't have any guarantee that your CPU can actually run it at 6400, but at least you have a 6000 XMP as fallback. Is it woth it? Probably not, unless you really need those additional 16GBs of memory regardless of the speed.

The easiest way to learn about memory timings for AM5 with Hynix memory would be following a guy called BuildZoid on YouTube (channel: Actually Hardcore Overclocking) and reading the publicly available posts on his Patreon. He does a lot of weird memory overclocks, but also offers pointers for easy daily timings. However, he is a bit unstructured and rambles a lot in his videos. There are some other sources that use lots of BZ's findings, but I usually enjoy watching his rambling videos since that's fun for me.

For example, my R9-7950X3D can't run my DDR5-6400 M-die kit stable at 6200 since my fancy Gigabyte overclocker board doesn't like the PMIC of the G.Skill sticks. Not even an active cooler blowing onto the RAM helps. However using a DDR5-6000 A-die kit from Kingston does DDR5-6200 just fine.
My R9-7950X does DDR5-6400 with that cursed G.Skill kit on another Gigabyte board for normal people, both on XMP and manually tuned, but needs active cooling for the later. At the moment I run it at 6200 without active cooler, because I'm waiting for those nice looking Jonsbo NF-1 memory fans since that CPU is inside a white showcase build.

Just for reference, here are some older stable timings of my cursed G.Skill memory kit. Please don't copy any of those timings directly unless you understand what they are doing, mainly because Gigabyte uses slightly different voltages than Asus, and those timings were done months ago on a nowadays outdated AGESA. However, if you compare them setting by setting with your system, you'll notice that your memory can probably run much faster than it does now, if tuned manually.

 

[ir_cow]

latency is pretty much irrelevant for DDR5 as are most other primaries, but he definitely has a lot of headroom in the secondary and tertiary timings.

Not really true that CAS is irrelevant. All timings matter and AMD will benefit the most from the lowest latency.

 

[Sarajiel]

Not really true that CAS is irrelevant. All timings matter and AMD will benefit the most from the lowest latency.

 

[ir_cow]

All hail buildzoid and his disciple @Sarajiel

 

[Sarajiel]

All hail buildzoid and his disciple @Sarajiel

Let me guess, you also overclock DDR5 for a living and came up with all your timings by yourself?

 

[ir_cow]

Let me guess, you also overclock DDR5 for a living and came up with all your timings by yourself?

Nah, just a hobbyist with my own opinions. I just fundamentally disagree that CAS doesn't matter. Though it could be argued that frequency doesn't either in certain applications. Just depends on what your doing really.

 

[dgianstefani]

Buildzoid's argument is that cas timings don't matter as much as subtimings for DDR5, which is true. The title of the video is clickbait.

 

[Sarajiel]

Nah, just a hobbyist with my own opinions. I just fundamentally disagree that CAS doesn't matter. Though it could be argued that frequency doesn't either in certain applications. Just depends on what your doing really.

TBH, I wish there actually were better resources for DDR5 overclocking than piecing it together by watching some 30+ hours of videos by some overclocking nerd, while scrubbing Reddit for actual useful information. Especially since I usually run quad rank setups of 128GB or hopefully higher once my new Kingston 96GB kits arrive later this week. When it comes to that sort of setup, I probably know more about it than Mr. BZ, because I seem to be one of probably only seven crazy people who want to run 128+ GB on AM5 at speeds faster than DDR5-3600/4800 or EXPO/XMP.
I really wish, I could point the OP to a written in-depth DDR5 guide for AM5 similar to the guides that exist for DDR4 on AM4.

However, BZ seems to be pretty much spot on that tCL alone isn't worth aiming for. Especially in my current main setup (128GB A-die, DDR5-5600 30-36-30-30, 1.35V VDD & 1.2V VSOC) it's basically pointless to increase VDD to 1.45V or 1.5V to keep running tCL even lower, because quad rank is an absolute nightmare to cool above 1.4V VDD. Not to mention that high voltage mode is still annoying on Gigabyte boards, which I'm sort of stuck with for PCIe lane splitting reasons.
Sadly, there doesn't seem to be many other important timings that scale well with voltage. On A-die you are usually better off with running tCL == tRP for example instead of pushing tCL. Also, if your voltages and temps get too high, you might need to lower tREFI for stability.

When testing my settings, I definitely observed that you rarely notice tCL changes with real world benchmarks. Maybe it's more important for folks who competitively benchmark AIDA or similar stuff, but that's not relevant for my use case or doesn't seem to affect most games.

 

[ir_cow]

... because I seem to be one of probably only seven crazy people who want to run 128+ GB on AM5 at speeds faster than DDR5-3600/4800 or EXPO/XMP.

Now that 2:1 ratio is a thing again, above 4800 might actually work now. But really I don't see the point in high freq for 128GB+. You are maxing out the slots because you need the capacity. The applications that all use that much do not benefit much from tight timings or high freq. A saving 10 minutes on a 10 hour render is pointless. Especially if it might crash in the middle due to those said tweaked timings.

I really wish, I could point the OP to a written in-depth DDR5 guide for AM5 similar to the guides that exist for DDR4 on AM4.

Write on yourself. Or ramble on video for hours at a time

 

[Sarajiel]

Write on yourself. Or ramble on video for hours at a time

But I'm not some famous l33t overclocker with enough epeen points on HWBot.

The applications that all use that much do not benefit much from tight timings or high freq. A saving 10 minutes on a 10 hour render is pointless. Especially if it might crash in the middle due to those said tweaked timings.

Who said anything about 10h render jobs? That's not what I'm doing on my main machine. You know, render nodes are still a thing. If I wanted to do that one system for everything approach, I'd use a TR, but that doesn't really work for me. I'd rather have a bunch of systems for dedicated tasks. Not to mention that R9-5950Xs are not just dirt cheap nowadays, but also fairly energy efficient and easy to setup with 128GB of Micron E-die. The problems usually occur with the AM5 systems that I use interactively and aren't really feasible with just 48GBs of memory like you'd use for a gaming system.

I'm pretty sure that my settings are stable enough for my use case. 72+ hours of a mix of y-cruncher VT3, linpackXtreme, and prime95 large FFTs while cooking my system with Furmark is usually a decent stability test. There is also quite a difference in setting up an overclock for benchmarks or for daily use.

 

[DemonicRyzen666]

Buildzoid's argument is that cas timings don't matter as much as subtimings for DDR5, which is true. The title of the video is clickbait.

All other times are for the "interior chips"(IC's) on the stick itself & how it interacts with the I/O dies memory controller.
It's so the IC's are more aware of the need to be open & available which leads to less wait times between IC's which generally means higher throughput.
So in general buildzoid is wrong as Cas timing still dictate latency overall anyways. Which can be shown by math.

 

[InVasMani]

What's more important than cas or frequency is the true latency. Also important in regard to that is best true latency you can eek out at the highest frequency and/or tightest sub timings. It's not really one thing it's the sum all things and how it comes together.

I'm still trying to dial in my memory settings and figure out what's working best with my CPU/MB and memory kit. I'm pushing 7000MT/s CL30 and pretty moderately loosened up sub-timings rn since Sleeping Dogs had a few occasional hiccups during game play. I was doing 6600MT/s CL26, but it was less stable though better true latency.

I might revisit it though since I think I got the voltages a little better dialed down. I think I can probably dial them in a little more now that I kind of have them pinpointed better in terms of which voltages need more emphasis and emphasis. I'm pretty happy with 7000MT/s CL30 though just the same especially if it remains stable. It's not too shabby for a kit rated for 7000MT/s CL34 to shave down CAS by 4 at the same frequency.

 

[AsRock]

Hello,

My PC is :
KF560C32RSK2-32 (that i would upgrade when/if DDR5 became improved, signle/dual rank.......?)
7800X3D
B650E-E
4070ti
1000w shift

About the RAM:
... i think about to upgrade to the best possible sticks (2x16), sooner or later... when/if there will be DDR5 that could really be better, for this computer setup specifically (CPU)...
... so, i ask you what are the important specs in PDF page (url of my sticks) ?.. and how much time should i wait (if needed) to be sure to grab improved DDR5 all in all compard to my actual ones ?
I search the best/perfect sticks for this setup, that's it... also there is the XMP/DOCP thingie, actually XMP didn't bothered my MB, i think, even, it's a full AMD (not the GPU btw) !

i really don't understand even one thing about RAM memory, i need to be advised

Based on a 7800X3D !

Going from CL32 to CL30 is 100% waste of money, even going CL28 is worthless.

 

[Klemc]

So, to test the CPU, what do i do first, to know if it has a good memory controller, or is potato !

Spoiler

I'm playing pathologic 2, this game not so impressive at all grphically, asks a lot of ressources !

 

[ir_cow]

Back to the original question of what would be better compared to the 5600 kit already in use. I would say because of the X3D already in the the system, in this case "faster" memory might show zero improvements.

Once again it comes back to what your doing. If it's gaming, the actual games and what resolution your at is important.

 

[Klemc]

1440p/240hz, mostly 120 FPS, 90 if needed (if fans spin too fast), unlimited if the PC handles well enough the game (120+).

100% games, i do only gaming there on my PC.

 

[dgianstefani]

1440p/240hz, mostly 120 FPS, 90 if needed (if fans spin too fast), unlimited if the PC handles well enough the game (120+).

100% games, i do only gaming there on my PC.

Fast ram certainly helps with 240 Hz. Even with a 7800X3D.

This is what you can do with 1.6 V and other tweaks on M die.

 

[Klemc]

Fast ram certainly helps with 240 Hz. Even with a 7800X3D.

This is what you can do with 1.6 V and other tweaks on M die.

View attachment 334326

i'll see, thanks...

... whatever, i'm playing a game that pushes any PC out there on his knees, Pathologic 2, yeah, DOCP ON or OFF, doesn't changes anything.

 

[Launcestonian]

Fast ram certainly helps with 240 Hz. Even with a 7800X3D.

This is what you can do with 1.6 V and other tweaks on M die.

View attachment 334326

But that's using BCLK overclocking to get to the unorthodox 6150 MT/s. Not sure if the OP board has its own clock generator?

 

[dgianstefani]

But that's using BCLK overclocking to get to the unorthodox 6150 MT/s. Not sure if the OP board has its own clock generator?

Neither does mine, you don't need it.

 

[Klemc]

But that's using BCLK overclocking to get to the unorthodox 6150 MT/s. Not sure if the OP board has its own clock generator?

https://www.reddit.com/r/ASUSROG/comments/11yega8

 

[Sarajiel]

But that's using BCLK overclocking to get to the unorthodox 6150 MT/s. Not sure if the OP board has its own clock generator?

AM5 Motherboards Sheet (X670/B650/A620) Should help to figure this kind of thing out.

This is what you can do with 1.6 V and other tweaks on M die.

Just out of curiosity, doesn't your CPU do 6200MHz with your memory kit?

 

[purecain]

With 6000mhz G.Skill RGB TridentZ5 with 4 16gb sticks I am only able to achieve 5600mhz 100% stable at cas30. If I took the cas to 32 and upped the voltage to 4.0v from 1.35v I think I could push 6000mhz. But would my OS die out of the blue one day soon. PROBABLY ^^ Correction, all it took was 1.4 for 4 sticks of ram to hit 6000mhz. Just testing for stability now.

Its just not worth losing all that latency for 200mhz...

Although im only in my first few days of the build so I will no doubt be back looking to push some more performance out of the ram over the next couple of weeks.

If anyone sees any obvious stuff I should be reading about and changing please advice. BTW I wanted to return this kit for 32gb modules or 24gb modules. Big mistake on my part. Although the Latency sucks on the old 32Gb modules. As soon as I heard new IC's were coming I tried to return the 16gb but was too late.

 

[dgianstefani]

With 6000mhz G.Skill RGB TridentZ5 with 4 16gb sticks I am only able to achieve 5600mhz 100% stable at cas30. If I took the cas to 32 and upped the voltage to 4.0v from 1.35v I think I could push 6000mhz. But would my OS die out of the blue one day soon. PROBABLY ^^

Its just not worth losing all that latency for 200mhz...

Seems realistic.

High voltage and low temperature are your friends though with ram OC.

 

[purecain]

Seems realistic.

High voltage and low temperature are your friends though with ram OC.

If I get a fan blowing on the sticks I could up the voltage, but being realistic with 64gb of ram, I'm probably lucky to of gotten them running at 5600mhz, although 5800mhz is where Im going to aim for next.
It Boots and is rock solid at 5600mhz, @5800 it wasn't perfectly stable in windows. I'll spend a day on that soon and see what info I can find out. I think with 4 sticks we are still finding the limits of the Chipset.

When I opened the bios to find they split the Chipset in two... I hadn't researched the dual chiplet design of either of the new High end designs. X670 and X670E. I understand this has led to BCLK overclocking.
That always seemed to mess up my windows installations over time. I've steered away from BCLK oc since the first Ryzen X1800...
Just looking at your use of it, does this mean its stable now??? Using this method for a permanent OC. Just looking at your post. Also can you OC an X3D Chip this way. I know the 5800X3d hated any kind of extra heat from the CPU. It was extremely temperamental.

Ok so decided today was the day and have the 64gb modules running at 5800mhz.... Just running a few stress tests ang then 6000mhz.

IMO the Hynix M-die definitely needs more than 1.35v... 5800mhz is now 100% stable with 1.37v.... Going to try 1.4v and 6000mhz next.

Well a simple 4.0v set in the bios gained me an easy 6000mhz... im still running the stress test for it but the pc feels stable as a rock.

When I set it up on EXPO on the first day of the build. My os nearly exploded. I thought Id corrupted the lot. Fortunately everything was recovered ok and I ended up running 5600mhz...

Pretty impressed with the pc running 6000mhz... I wonder how long it will take for this novelty to wear off and 6400mhz become absolutely necessary for life to continue.