MSI X670E Godlike - performance with RAM 128Gb and 3 SSD M.2

Haereticus

New member
Joined
Feb 8, 2023
Messages
12
Good morning, everyone.
A few months ago I switched from an Asus X670E motherboard to the top of the line MSI motherboard, i.e. the Godlike, attracted by some advertised features, including RAM support up to 256Gb, and 2DPC 2R Max speed up to 5400+ MHz (it said 6000+ when I bought the motherboard), and secondly the presence of three M.2 slots totally independent (i.e., unaffected by the presence of other peripherals-at least on paper), specifically slot 2_1, managed by the CPU, running at PCIe 5.0x4 and 2_2 and 2_3 managed by the chipset, running at PCIe 4.0 X4 (the fourth M.2 slot is connected to the third PCIe slot).

I thus transferred my 4 memory slots (128Gb) G.Skill EXPO 6000 and my 3 SSDs (one of which is 5.0) to the new motherboard.

Unfortunately, expectations were disappointed:
with four memory banks it is impossible to boot the motherboard except at the maximum frequency of 4000;
  • the performance of the three SSDs is about a third slower than what I was detecting on the Asus MB. To give an example with the 5.0 SSD, which must necessarily be installed in slot 2_1, I get a "score" in the 3DMark Storage Benchmark just above three thousand points, while on the Asus I was well above 4000. With synthetic benchmarks such as CrystalDiskMark I get read/write performance of ca.10,000 versus about 12,000 that the SSD is capable of (SSD temperatures do not seem to be an issue, remaining well below the threshold for Throttling)

Is there by any chance something I'm missing, or more simply are there "limitations" that are not properly advertised in the datasheet?

PS) The only other peripheral present is the GPU, obviously in the first PCIe slot.
 
DDR5 just highly dislikes if you populate all four slots, and AMD AM5 especially so. Whatever RAM size you need, you want to reach it with two modules. Treat the board like it only had slots A2 and B2 (2nd and 4th from the left) and treat the other two as decoration, if you want any kind of decent speed.

DDR5-6000 with four modules currently has little chance of working on AM5, due to twice the stress on the memory system compared to two modules. Even with just two modules, DDR5-6400 is already about the limit from where it can become harder to make it work on AM5. So DDR5-6000 with four modules is best not even considered, that kind of thing may work on Intel Z790 + 14th gen CPU, where the CPU's memory controller is more advanced, but it could be challenging there too. On AM5 at the moment, i don't see how it could work, unless your CPU happens to have an exceptionally good IMC (integrated memory controller).

Also see RAM explained: Why two modules are better than four / single- vs. dual-rank / stability testing.

What kind of speed did you get with four modules on the Asus X670E board, surely not the DDR5-6000 XMP?

Do you have any actual use for that much RAM? Because while such an amount of RAM can be necessary for professional applications, like rendering, video processing, lots of VMs, things of that nature, it will do absolutely nothing for gaming for example, most games don't even use more than 16 GB RAM. If you only want 128 GB because you think "more is better" (no offense, sometimes it is), but if you don't actually have any professional workloads that require it, you're only making things unnecessarily difficult and expensive, while providing zero benefit from that much RAM being available. So then i would stick to one kit of 2x 32 GB. That should work at XMP no problem.

top of the line MSI motherboard, i.e. the Godlike, attracted by some advertised features, including RAM support up to 256Gb

Gb would be Gigabit, but of course you mean GB. Thing is, that's not a feature of the GODLIKE, that's just something they added in a BIOS update, see here on the lowly PRO X670-P WIFI: https://www.msi.com/Motherboard/PRO-X670-P-WIFI/support

and 2DPC 2R Max speed up to 5400+ MHz (it said 6000+ when I bought the motherboard)

The IMC is the main sticking point in the memory system here, not the board or the RAM itself. On AM5 this is actually the most pronounced, since the memory controller inside the CPU is worse than on Intel. Once you populate all four slots, the attainable DDR5 speed goes down the drain pretty much.

The following factors all affect if the RAM can actually run at a certain setting:
1) The mainboard model (PCB layer count, PCB trace optimization, RAM slot topology and slot count, component selection, RAM VRM etc.)
2) The mainboard's BIOS optimizations and the BIOS settings
3) The CPU's integrated memory controller (IMC), quality depends on the CPU generation as well as on the individual CPU (silicon lottery)
4) The properties of the RAM modules.

With 1) you got the best and then some, no doubt. But that's not the limiting factor on AM5. For 2) it's the same for all MSI boards with this chipset, as they all use the same codebase and AGESA. With 4), your RAM alone could do it, for example when only using two modules. So that leaves 3), this is the main reason why such configurations rarely work well on AM5. The memory controllers are less good than on Intel right now.

The board specs and QVL become more and more irrelevant in case of these two factors: Enthusiast-grade XMP speeds, or very challenging configurations like four big dual-rank modules. See the replies here, they were mostly about the first factor, but it applies just the same to the challenging configuration. At this point the QVL becomes more of a marketing tool, not a guarantee that it will work. For the QVL, MSI work with CPUs binned for top-grade IMCs, most users' CPUs will not have such a capable IMC. So the QVL always represents the best-case scenario.

There is not a single 4x 32 GB DDR5 kit that is meant for AMD AM5 with an EXPO profile. There's one Corsair kit like this (without and with RGB), but it's meant for Intel, https://www.corsair.com/us/en/p/mem...t-s-cl40-memory-kit-black-cmk128gx5m4b5600c40

If you can make do with 96 GB of RAM, get a kit of 2x 48 GB with an EXPO profile, for example:
This has the best chance of actually running at DDR5-5600.

and secondly the presence of three M.2 slots totally independent (i.e., unaffected by the presence of other peripherals-at least on paper)

X670E GAMING PLUS WIFI:
Screenshot 2024-03-06 at 18-38-13 X670EGAMINGPLUSWIFI_EN.pdf.png


PRO X670-P WIFI:
Screenshot 2024-03-06 at 18-37-58 PROX670-PWIFI.pdf.png


MAG X670E TOMAHAWK WIFI:
Screenshot 2024-03-06 at 18-38-29 MAGX670ETOMAHAWKWIFI.pdf.png


MPG X670E CARBON WIFI:
Screenshot 2024-03-06 at 18-38-54 MPGX670ECARBONWIFI.pdf.png


MEG X670E ACE:
Screenshot 2024-03-06 at 18-39-17 MEGX670EACE.pdf.png



I've been critical of the GODLIKE boards (since a few generations), their value-for-money is pretty bad to say the least. The maximum i'd ever recommend is the ACE, but if there's a UNIFY-X available with only two DIMM slots, that would actually be the best. Since most people should just stick to using two DDR5 modules anyway, and having only two DIMM slots results in superior signal quality over four slots with two empty ones. Sadly those two-DIMM-slot boards have become rare, even though four modules tend to run quite badly.

Coming to the weaker M.2 PCIe SSD benchmark scores on the MSI board. I'm of the opinion that you should try to rectify the RAM situation first, or at least benchmark the SSDs while using only two modules in slots A2 and B2 and thus at XMP speed. Because everything you do on the system has to go through the RAM first. DDR5 depends on higher speeds to really get going, because it has to compensate for its higher latencies compared to DDR4. This means that DDR5-4000 performs considerably worse than DDR4-3600 for example, so it becomes a bottleneck. You want to run DDR5-5200 or -5600 even for challenging configurations. To me, DDR5-4800 is the absolute minimum i would accept for a highly challenging configuration such as yours. Anything lower and you will have to change something, maybe go to 2x 48 GB if you require that much RAM.

Generally, i'm not sure what RAM speeds you managed on the ASUS, but it seems like it was a bit higher, so maye it's something about the MSI BIOS after all which prevents this configuration from going at least to DDR5-4800 or so. But going high into the DDR5-5xxx range or even up to XMP, i highly doubt that this was possible on the ASUS either. But yeah, i agree, the GODLIKE wasn't the best purchase for several reasons. Especially not when you're doing a sidegrade from the same platform on ASUS, that's just a waste of money at that point.

You could open a ticket with MSI about your issues, maybe they will ask you to try something about the RAM situation (but usually you can't force it to perform very well when it won't even go above DDR5-4000 now), and that would then perhaps also make the SSDs work better. But yeah, try that with just two RAM modules, curious if the scores improve.
 
Last edited:
Citay, thank you for your long reply, although it contains information I already knew. What turns my nose up at is the fact that all MB manufacturers include in the datasheet (or manual) the speed limitation that can be achieved when using all 4 memory slots.

There is no trace of this on the Godlike datasheet, this is to me a serious lack of clarity.

However, this is a minor problem, because given the impossibility of using Expo 6000 mode, I limited the use to 64GB of memory in two slots (but I forgot to write that down).

And yes, I actually do need a large memory configuration, because my use of the PC is certainly not just gaming (which is actually residual), but I can easily 'make do' with 64GB.

Also everything written about SSD performance relates to the PC with only two memory slots...

Again according to the datasheet and the manual, with three M.2 SSDs there should be no loss of performance... but in the end that's not the case... However, searching a little online, I verified that this is a case history that others have already encountered...

In fact, I have opened a ticket, but as yet I have not received any feedback...
 
What turns my nose up at is the fact that all MB manufacturers include in the datasheet (or manual) the speed limitation that can be achieved when using all 4 memory slots.

There are 2 very different things here:
1) CPU IMC overclocking capabilities
2) motherboard RAM overclocking capabilities
That "2DPC 2R Max speed up to 5400+ MHz" is not related to your CPU.

Also, another known thing: all the current CPUs (AMD and Intel both) have terrible performances with 4 memory modules.

Even more, another rule here: more memory you have, lower speed you can achieve.

You didn't mention anything about your CPU, but let's take a look at one of the best AMD CPUs:
https://www.amd.com/en/products/apu/amd-ryzen-9-7950x3d#product-specs
Max Memory Speed
2x1R : DDR5-5200
2x2R : DDR5-5200
4x1R : DDR5-3600
4x2R : DDR5-3600


The bottom line: with all the current CPUs if you push the IMC to its limits, don't expect anything good.
So your 4000 speed is expected.
Other CPU samples can do more (4400 or 4800), others can do less (see the AMD guaranteed speed)
 
There is no trace of this on the Godlike datasheet, this is to me a serious lack of clarity.

You can see it on here: https://www.msi.com/Motherboard/MEG-X670E-GODLIKE/Specification
But you also quoted that figure before in your initial post, so maybe you mean something else?
And even those specs, like i mentioned (at least in something i linked), those are reached by their in-house overclockers, those are not "plug&play" numbers.


I limited the use to 64GB of memory in two slots (but I forgot to write that down).
Also everything written about SSD performance relates to the PC with only two memory slots...

Ok, yeah, didn't know that, so some of what i wrote was obviously redundant. In general, i thought i'd write it all down, no matter what you knew and what you didn't, so you get the whole picture as best as possible.

Are you using the latest BIOS version? Always worth trying if you're not satisfied with something.

Again according to the datasheet and the manual, with three M.2 SSDs there should be no loss of performance...

Hopefully you're benchmarking them one at a time, not all simultaneously? The SSDs going through the chipset will eventually be limited by the bandwidth available between CPU and chipset, if you tax several at once. The one that gets PCIe 5.0 x4 lanes from the CPU shouldn't have a limitation. The question is, is there something not up to par in general, or only the SSDs?

To that end, we could check the sensors with HWinfo64. Run it and open "Sensors", then expand all sensors by clicking on the little <--> arrows on the bottom, also expand the columns of the sensors a bit so everything can be read. Make it three big columns of sensors (or four, if the screen resolution is high enough). In the end, it should be a screenshot with all the sensors visible at once, like this:

yes.png


Make sure the power plan in Windows is on "Balanced". Do nothing on the PC for a while (couple minutes), so the "minimum" baselines for the values are established. After that time in idle, then produce full CPU load with Cinebench R24, and after completing a 10 minute run, when the CPU temperatures have stabilized at the highest level, take a screenshot of the sensor window and tell me the Cinebench score. On this screenshot, we can see if everything is alright with fully multithreaded CPU load, the highest normal load you can encounter in daily use without resorting to an artificial stress testing tool like Prime95.
 
I thus transferred my 4 memory slots (128Gb) G.Skill EXPO 6000 and my 3 SSDs (one of which is 5.0) to the new motherboard.
First, only put 2 sticks, make sure everything working as expected. Then, update the bios to the latest one. The latest one overclocks best, as least it is true for my MSI x670e Carbon.

For 128GB setup:
You need Hynix A die. Not all 6000 kit can do 128GB above 5600.
Also I found not all A die kit can run above 5600.
The first thing I would try is Enable EXPO (5600 or 6000) with 128GB, before saving the bios, down clock it to 5200 . If that doesn't work, you got wrong kit and they will never work above 5600. Return that kits and try different brand and model.
ASUS got the best bios, but on AM5, they don't clock memory well comparing other brands (except x670e gene, that is 2 dimm board)
MSI, hardware-wise is excellent, but the bios is buggy with AM5. Don't expect EXPO 128GB will work out of the box. To make it work 5600 above, you need to tune all the voltages and timings.

The highest I can boot into windows with 128GB is 6800, not fully stable. The maximum stable frequency you can expect is around 6200.
 
Max Memory Speed
2x1R : DDR5-5200
2x2R : DDR5-5200
4x1R : DDR5-3600
4x2R : DDR5-3600


The bottom line: with all the current CPUs if you push the IMC to its limits, don't expect anything good.
So your 4000 speed is expected.
Other CPU samples can do more (4400 or 4800), others can do less (see the AMD guaranteed speed)
I was not aware of these limitations determined by the CPU (and not the motherboard). Very interesting.
 
You can see it on here: https://www.msi.com/Motherboard/MEG-X670E-GODLIKE/Specification
But you also quoted that figure before in your initial post, so maybe you mean something else?
And even those specs, like i mentioned (at least in something i linked), those are reached by their in-house overclockers, those are not "plug&play" numbers.
>4x DDR5, Maximum Memory Capacity 256GB
>Memory Support DDR5 8000+(OC)/ 7800(OC)/ 7600(OC)/ 7400(OC)/ 7200(OC)/ 7000(OC)/ 6800(OC)/ 6600(OC)/ 6400(OC)/ 6200(OC)/ 6000(OC)/ 5800(OC)/ 5600(OC)/ 5400(OC)/ 5200(OC)/ >5000(OC)/ 4800(JEDEC) MHz

What I mean to say is that it does not say that using all four modules poses limitations (in fact, it would seem the opposite)...
 
What I mean to say is that it does not say that using all four modules poses limitations (in fact, it would seem the opposite)...

The limitations are both on the CPU and motherboard sides.
But for the motherboard (a "static" player here) they're not huge:
https://www.msi.com/Motherboard/MEG-X670E-GODLIKE/Specification
Max. overclocking frequency:
• 1DPC 1R Max speed up to 8000+ MHz
• 1DPC 2R Max speed up to 6400+ MHz
• 2DPC 1R Max speed up to 6400+ MHz
• 2DPC 2R Max speed up to 5400+ MHz
 
04b.png
Make sure the power plan in Windows is on "Balanced". Do nothing on the PC for a while (couple minutes), so the "minimum" baselines for the values are established. After that time in idle, then produce full CPU load with Cinebench R24, and after completing a 10 minute run, when the CPU temperatures have stabilized at the highest level, take a screenshot of the sensor window and tell me the Cinebench score.
01.png
03b.png
R24.png
 
>You need Hynix A die. Not all 6000 kit can do 128GB above 5600.
Unfortunatly I don't know if my G.Skill F5-6000J3040G32GX2-TZ5NR have Hynix A die

>The first thing I would try is Enable EXPO (5600 or 6000) with 128GB
With Expo enabled and 4 dimm, the motherboard won't post (Ab error)
It only starts if I manually set the frequency to 4000 or 4200
 
Last edited:
The limitations are both on the CPU and motherboard sides.
But for the motherboard (a "static" player here) they're not huge:
https://www.msi.com/Motherboard/MEG-X670E-GODLIKE/Specification
Max. overclocking frequency:
• 1DPC 1R Max speed up to 8000+ MHz
• 1DPC 2R Max speed up to 6400+ MHz
• 2DPC 1R Max speed up to 6400+ MHz
• 2DPC 2R Max speed up to 5400+ MHz
Hi, @RemusM

From my understanding, the performance claims on the specs page are the highest kits that listed in the QVL.
Whether a kit works or not is related to the quality of the IMC. If MSI uses a golden sample Ryzen 7950x to validate the QVL, there will be no way that let average CPU work at that frequency.
The question, what percentage of the CPU do you expect will work based on MSI's QVL?
 
Hi, @RemusM

From my understanding, the performance claims on the specs page are the highest kits that listed in the QVL.
Whether a kit works or not is related to the quality of the IMC. If MSI uses a golden sample Ryzen 7950x to validate the QVL, there will be no way that let average CPU work at that frequency.
The question, what percentage of the CPU do you expect will work based on MSI's QVL?
I would have assumed otherwise...
The cpu/memory QVL combinations posted on the site as 'compatible' should always work...
But I don't know
 
The question, what percentage of the CPU do you expect will work with MSI's QVL?

Statistically speaking, only 10-20% of the CPU samples can achieve those speeds (using decent voltages).
Again, talking about the current CPUs:
For Intel, anything above 5600 is overclocking
For AMD, anything above 5200 is overclocking
And yes, there are CPUs (AMD and Intel both) capable of DDR5-8000, but using a small amount of memory and insane voltages (not suitable for a 24/7 usage).
 
I was not aware of these limitations determined by the CPU (and not the motherboard). Very interesting.

That's mostly just AMD (and Intel) covering their bases, so nobody can point the finger at them. Meanwhile they know that almost everyone will use XMP / EXPO (XMP is even an Intel invention, EXPO an AMD one). So just by looking at what is officially supported from the CPU side, we cannot immediately say, "well it says only DDR5-5200 (two modules) or DDR5-3600 (four modules) there". Because those are their low-ball numbers so they can't be blamed for any problems. I linked a video before in this thread, and at this particular point in the video, it explains how and why Intel and AMD come up with these low numbers. In reality, there's speeds that are generally accepted to work well, there's a "sweet spot", and then there's speeds and configurations that are likely to cause trouble. On Intel (13th and 14th gen), all of these are slightly higher than on AM5.

With two modules on AM5, if you prioritize problem-avoidance over everything else, DDR5-5200 or -5600 would be ok. I'd say DDR5-6000 is the sweet spot. DDR5-6400 can be hit or miss on AM5, anything higher and you really need to have lucked out with your CPU's IMC. Now, with four modules, the only thing worth even trying on AM5 is 4x 16 GB, that should work into the DDR5-5xxx range. But even that is nonsense, you'd use 2x 32 GB instead which will work much better again. 4x 32 GB or 4x 48 GB is just causing so much stress on the memory system that it's mostly not even worth trying. If you need that much, just go Intel 13th or 14th gen on Z790, period.

The whole idea of relying on the manufacturer's numbers (CPU or board manufacturer) to tell you what you can expect, this should be thrown out the window more or less. The CPU manufacturer is low-balling the number, the board manufacturer is listing a marketing number. The best bet is to look at the RAM manufacturer's lists if they offer one. But then you have an additional problem: You are combining two kits of 2x 32 GB that were never meant to run together, certainly not on AM5. G.Skill don't even offer a single 4x 32 GB kit of DDR5. So whatever numbers you find at G.Skill, they don't apply to you, since 4x 32 GB immediately doubles the stress on the memory system and makes everything much worse.


What I mean to say is that it does not say that using all four modules poses limitations (in fact, it would seem the opposite)...

That's always been the case on boards with a daisy-chain memory slot layout though. Of course the board makers are hesitant to highlight this fact. People only tend to find out later, when using four modules runs badly. That's why i wrote and called my thread "RAM explained: Why two modules are better than four".


Coming to the HWinfo / Cinebench screenshots. You didn't mention that you had a 7950X previously, and now went to a 7950X3D. The latter clocks the CCD with the 3D cache about 400 MHz lower than the former. It's important to use the latest AMD Chipset drivers and the latest Win11, so threads don't land on the slower CCD (when it's not full multithreaded load). Furthermore, if you re-use a Windows installation from a non-X3D CPU and then use it with an X3D CPU, it's a must to re-install the latest Chipset drivers, because only then will it install the proper driver for the thread assignment on the X3D.

Anyway, your Cinebench R24 score is completely normal for this CPU, nothing unusual there, no reduced performance. HWinfo, well, it should always be set to English. But nothing unusual there either, sensors look ok. The GODLIKE's completely overbuilt VRM stays ice cool too, as expected.

So it really seems mostly something to do with the SSD performance. Is it filled to about the same level? Have you tried with a TRIM command before running the benchmark (Defrag "Optimize" and then waiting a bit)? Is the drive maybe starting to thermally throttle on the GODLIKE? Try running HWinfo in the background again (set to English), then run CrystalDiskMark which should heat up the drive pretty good. I'm curious what happens to the temperatures.
 
Last edited:
AMD Ryzen™ 7 7800X3D

Specs:

System Memory Type: DDR5
Memory Channels: 2
Max Memory Speed:
2x1R DDR5-5200
2x2R DDR5-5200
4x1R DDR5-3600 <<< there 4 module speeds Guaranteed for your CPU (1600Mhz drop from 2)
4x2R DDR5-3600 <<< there 4 module speeds Guaranteed for your CPU (1600Mhz drop from 2)

the CPU's IMC (Memory Controller) is the King in the situation of RAM Speed

all the Board does is has a bunch or slots and controlled Impedance Traces and the Speed the board says it can do hell yes it can do that as that is what the Traces on the board allow but if the CPU cant do that then its never going to happen as no BIOS update for firmware can fix what the memory controller in the CPU can do as it is what it is as its Hardware and is set in stone.

all updates can really do is try and improve incompatibilities and add RAM Specific Sub Timings that can only help a small amount.
 
Coming to the HWinfo / Cinebench screenshots. You didn't mention that you had a 7950X previously, and now went to a 7950X3D. The latter clocks the CCD with the 3D cache about 400 MHz lower than the former. It's important to use the latest AMD Chipset drivers and the latest Win11, so threads don't land on the slower CCD (when it's not full multithreaded load). Furthermore, if you re-use a Windows installation from a non-X3D CPU and then use it with an X3D CPU, it's a must to re-install the latest Chipset drivers, because only then will it install the proper driver for the thread assignment on the X3D.
New motherboard, new cpu and clean windows 11 install. All drivers are up to date. I also updated the ssd firmware.

Anyway, your Cinebench R24 score is completely normal for this CPU, nothing unusual there, no reduced performance. HWinfo, well, it should always be set to English. But nothing unusual there either, sensors look ok. The GODLIKE's completely overbuilt VRM stays ice cool too, as expected.

So it really seems mostly something to do with the SSD performance. Is it filled to about the same level? Have you tried with a TRIM command before running the benchmark (Defrag "Optimize" and then waiting a bit)? Is the drive maybe starting to thermally throttle on the GODLIKE? Try running HWinfo in the background again (set to English), then run CrystalDiskMark which should heat up the drive pretty good. I'm curious what happens to the temperatures.
 
I also updated the ssd firmware.

Sometimes they fix bugs and it performs a bit worse afterwards. For example if the throttling is more aggressive afterwards to prevent problems. Doesn't hurt their sales, because the sales of the drive are mostly driven by the launch reviews that have long been done on the initial firmware. But let's see how high the temperatures get.
 
Sometimes they fix bugs and it performs a bit worse afterwards. For example if the throttling is more aggressive afterwards to prevent problems. Doesn't hurt their sales, because the sales of the drive are mostly driven by the launch reviews that have long been done on the initial firmware. But let's see how high the temperatures get.
I updated the firmware hoping it would solve the problem, but it did not.
There was actually a small improvement in the data...

A01.png
A02.png
MSI_vs_Asus.png
 
53°C maximum for the MP700? How does it stay that cool? So it's definitely not anywhere near throttling.

But i don't know what ASUS were doing. The numbers from MSI seem more consistent with the review benchmarks.

Screenshot 2024-03-06 at 22-33-17 Corsair MP700 im Test.png
Screenshot 2024-03-06 at 22-33-10 Corsair MP700 im Test.png



Screenshot 2024-03-06 at 22-39-10 Corsair MP700 Gen5 M.2 SSD Review.png


Screenshot 2024-03-06 at 22-39-43 Corsair MP700 2 TB Review - 10 GB_s Gen 5 SSD Tested.png


The MP700 often trails behind the Crucial T700 a bit (same Phison controller and everything), because Crucial seem to have optimized their firmware better. Of course, apart from theoretical benchmarks, these numbers don't give that much of a real-world advantage to something like your SN850X (got one of those as well, they're good), considering the price difference (almost double the price). But i'm really starting to wonder if the MSI numbers are not the more representative ones after all. Only then i don't know how ASUS got to those numbers.
 
Back
Top