msi pro z790-p wifi DDR5 review

Admin

Administrator
Moderator
Messages
3,032
#1
I decided to get this motherboard due to the fact that i just couldn't find any particularly great alternative to it that was available. I had bad experiences with my previous gigabyte AM4 board so i didn't feel like waiting for them to release a better alternative (which may have been a mistake).

CPU: 13900KF
Cooler: as500plus
ram: 2x32 GiB sk-hynix M-die (final manufacturing by g-skill).

After finally putting my system together and booting into windows i found out that the internet didn't work at all. To fix that i needed to go online to the msi website to download new drivers (the CD they provided didn't work either). I tried booting into linux but at that time icouldn't even get that to boot since secure boot was enabled by default and i didn't find where to disable it. I ended up solving the issue by using my LG V20 to do USB-tethering (no longer supported on my LG G4).

So i would recommend to download drivers in advance or have a second computer available so you can put things you need on an USB.

They still haven't published any LAN drivers for windows 10, only windows 11. WIFI still works with windows 10 though.
 

Admin

Administrator
Moderator
Messages
3,032
#2
DDR5 overclocking
All i had to do to reach 6400 MT/s was to change the memory ratio from 60 to 64, i did not have to actively raise any voltages or lower any timings.

Unfortunately no attempt at getting 6600 to be stable was successful. I do not know if the board is to blame for that though.

The motherboard will automatically raise the memory controller voltage "CPU VDDQ" as you overclock the memory in "geat 2" and this may cause unsafe voltage to be applied to the cpu memory controller potentially causing degradation. This is not at all obvious to the user since the CPU VDDQ voltage is at a different place and you will not even see the new voltage until after rebooting the system. I do expect people to damage their CPU memory controllers by not knowing what they are doing.

CPU SA is automatically set to 1.29 @ 6400 MT/s.

Something that is missing is ability to just change all timings or group of timings (such as sub-timings) at once by % or cycles, that would make it a lot easier to achieve higher DDR5 frequencies. Instead you have to manually loosen timings to overclock further and there are a lot of these timings (more than 20) and just one being too tight can cause instability.

With dual-rank ram msi will set the CPU VDDQ to 1.45v at 6600 MT/s if you have that voltage to auto. I am not sure if that is too much or too little since i never got that 100% stable (i ran prime95 using the "large FFT" setting). At 6400 MT/s the default CPU VDDQ voltage is 1.35 which hasn't caused any stability-issues so far.

By default the board will retry failed ram overclock 2 times which is 2 times more than what it should be, it really doesn't make sense to retry an overclock that turned out to be unstable unless you are trying to set some overclocking records which isn't the target demographic for this board in the first place. This is just another example of a bad default setting.
 

Admin

Administrator
Moderator
Messages
3,032
#3
Terrible fan-control
It was much easier to adjust the fan-curve on my old gigabyte board. Now you need to manually change each point on the curve for each individual fan on the graph. You cannot just apply the same graph to multiple fans at once and i didn't find any way to adjust the graph using just the keyboard.

They also lack decent presets.

You can eventually get a good set-up but it just takes needlessly long time.
 

Admin

Administrator
Moderator
Messages
3,032
#4
Not respecting the intel power-limit
Intel officially has a 243W power-limit for the 13900K and 13900KF but msi ignores that completely.

At first after updating the bios the power-limit will be 4096W "unlimited" but after updating the bios i actually got to select between 3 options all of which would cause the CPU to throttle at 100 °C even with a good air-cooler. For their "boxed cooler" option it's 253W and it's much higher if you select "tower cooler".

The cooling capacity of the AS500plus seems to be around 180W if you are not willing to go above 100 °C, makes me wonder if there is something wrong with my installation (people recommended getting a contact frame but i didn't find it in stock for a decent price before getting the motherboard).

I will need to get a better cooler or face significantly diminished MT performance which was to be expected. The AS500plus was originally bought for a different purpose.
 

Admin

Administrator
Moderator
Messages
3,032
#5
Is the dual 8-pin actually useful?
I wasn't sure if my current corsair TX650 would be enough for a 3090 at 375W PL and a 13900KF. I was thinking that if i switch to a better PSU and cooler (such as custom loop) it would be very useful to to have two 8-pin ports so i could take the full advantage of that better PSU and cooler.

When i tested using my old TX650 PSU i tried my best to trip my PSU OCP but nothing i tried worked. I tried running luxmark while starting and stopping prime95 to somehow hit a PSU limitation but everything just kep working (except for the cooling, throttled quickly). Later i discovered that the single 8-pin wasn't even properly connected so it's a bit surprising i didn't run into any issue.
 

Admin

Administrator
Moderator
Messages
3,032
#6
CPU overclocking is confusing but you do get a lot of options
Intel has settled for a needlessly complicated turbo boost system for the 13900K and 13900KF. I wanted to disable "Termal Velocity Boost" to and just have it just be power-limited instead but i don't actually know if TVB is actually disabled currently.

I opted to manually offset the voltage-frequency curve but the issue then is that you do not actually see the resulting voltage or the actual curve. All you can do it to edit the offset for the ratios and you cannot actually change these ratios, you can however change the base frequency which will also affect the ram (can be both for the better or worse).
 

Admin

Administrator
Moderator
Messages
3,032
#7
Trying to reach 6600
While reaching 6400 MT/s was very easy trying to reach 6600 has been very difficult.

I have at least gotten close to the point where i need to test again with a less aggressive voltage curve

Running "stressapptest -s 2009 -M 256 -m 8 -W" via linux i got:
Code:
Report Error: miscompare : DIMM Unknown : 1 : 80s
2022/11/13-23:21:21(CET) Hardware Error: miscompare on CPU 0(0x1) at 0x7fcfc4a024f8(0x4f8:DIMM Unknown): read:0xfffffffffffffffd, reread:0xffffffffffffffff expected:0xffffffffffffffff
DRAM voltage: 1.44
CPU VDDQ: 1.4
CL34,40,40,40

At 6400 the following is about how low you can go with the following timings (without getting stability issues)

tFAW: 32
tRFC2: 370

So losening the main timings is no longer enough at 6600, a lot of timings will become too tight at that higher frequency. The issue with losening timings is that it actually adds latency and if you loosen them too much you might end up with more latency at that higher frequency. It actually got more stable when i reduced the voltages and loosened timings (main and sub) more but i couldn't get it fully stable that way.

It is worth noting that there will always be a risk for hardware error but you want to at least be able to pass stability tests with decent reliability. I never ran into any stability issues at 6400 for CL32, 38, 36, 32, 32 though.
 

Admin

Administrator
Moderator
Messages
3,032
#8
Example of pointless DDR5 overclock
Sure you can boot to windows but what's the point if it's unstable. Not worth it just to get some screenshot to share (you could just use pinta for that).

meaningless.PNG


A lot of bad things can happen due to overclock being unstable. You do not want to risk getting important files corrupted or having a game crash during an important moment.
 

Admin

Administrator
Moderator
Messages
3,032
#9
BIOS overview
This doesn't cover all bios settings. I show how to disable secure boot among other things.

You can try tweaking the "CPU Lite Load" setting in the "Advanced CPU Configuration" section, (AC_LLC/DC_LLC presets). Your current value is 9 - try finding the lowest possible value your CPU will be stable at. Lower values provide lower temperatures/power consumption. Not sure how this will work while custom offsets are provided in the VF curve. The BIOS looks nice, it has everything one might need.
From what i have read you may need to change the "CEP" setting to make your make your undervolting take affect. I will look into that more after tweaking my ram timings.

https://www.igorslab.de/en/the-last...unify-x-test-with-adaptive-oc-and-teardown/6/
 

Admin

Administrator
Moderator
Messages
3,032
#10
Not even my 6400 memory overclock was fully stable
The issue was that prime95 large FFT simply wasn't good enough at spotting memory stability issues, this resulted in an unstable 6600 overclock appearing to be stable at first.

Currently under linux i am doing the following test which my older 6400 MT/s overclock couldn't pass. I have tried raising the voltages slightly and loosening timings but it still hasn't passed. This is worrying.
Code:
stressapptest -s 7200 -M 32768 -m 32 -W
I noticed that it actually became less stable at higher voltage indicating that it was a temperature issue. Lowering the voltage back to 1.4 (for the DRAM) made it close to fully stable.

1668548214529.png


The ram stick was really hot when i touched it so i need to reduce the voltage further or add a fan.

It did pass once at stock CPU settings
1668682843089.png

133*24
1.4v DRAM voltage
1.38v CPU VDDQ
1.26v CPU SA
 

Admin

Administrator
Moderator
Messages
3,032
#11
6400 MT/s finally stable?

I did run it twice. For reason unclear to me the speed ended up lower this time.
Code:
stressapptest -s 7200 -M 32768 -m 32 -W
1668719602179.png

Code:
stressapptest -s 7200 -M 60000 -m 32 -W
1668719644309.png

Code:
stressapptest -s 7200 -M 32768 -m 32 -W
1668774182539.png

I switched back to the default "sub timings configuration" and it passed first attempt further indicating that the issue was "Turn Around Timing Configuation" (might be just one subtiming that was too tight).
1668818122802.png

It passed again with the cas latency lowered to 32
1668848772335.png
 

Admin

Administrator
Moderator
Messages
3,032
#12
6464 MT/s
It almost passed using the same voltages

1668733968029.png


It failed again at 1.405v but has so far passed once at 1.41v (DRAM voltage).

1668789678259.png


It did however fail the next time.

1.41v: failed first test
1.415v: passed the first test (ended early) failed the next test (4 hardware incidents).

I ended up switching to tuning my 6400 overclock instead (too much work just for 1%). Could be that some subtiming became too tight again or less likely "CPU VDDQ" or "CPU SA" needed to be raised.
 

Admin

Administrator
Moderator
Messages
3,032
#13
6400CL32,40,38,38,76
Only tRCD and "Turn Around Timings" loosened. Still poor transfer rate though.

1668863740403.png


1668863774384.png
 

Admin

Administrator
Moderator
Messages
3,032
#14
Progress at last
Thanks to buildzoid i was informed that a good stability test for ram is the y-chruncher VST test. That made the stability-testing process significantly faster and more reliable. Luckily the ram-settings that i thought was stable since earlier did also pass the y-cruncher test and i were able to identify find 8 turnaround timings that could be tightened at least back to XMP.

After updating the bios to the latest version (that at first couldn't be downloaded even though the download button appeared on their website) and loading the XMP profile the following turnaround timmings (in bold) were close to stable stable at 6400

Code:
Timing   'Stable'  unstable
tRDRDSG  18
tRDRDSG  12        10
tRDRDDR  16
tRDRDDD  16
tWRWRSG  32        30
tWRWRDG  8
tWRWRDR  16
tWRWRDD  16
tRDWRSG  21        20*
tRDWRDG  21        20*
tRDWRDR  24
tRDWRDD  24
tWRRDSG  60
tWRRDDG  52
tWRRDDR  16
tWRRDDD  16
*not tested separately

Interestingly the bandwidth in testing is significantly faster by tightening these timings back by just 2 so the timings that were actually important for performance does not seem to have been the ones that needed to be loosened, at least not most of them.

Unfortunatily tRDRDSG is very important for the read bandwith so i need to find a way to tighten it further. As a first resort i am reducing the CPU VDDQ to 1.35 and CPU SA to 1.2'

tWRWRDG is important for the write bandwith, you don't want to set it higher than 8.
 

Admin

Administrator
Moderator
Messages
3,032
#15
Close to stable at 6400
The lowering of CPU SA to 1.2 and VDDQ to 1.35 allowed the ram to pass 1 hour of y-cruncher VST without error with tRDRDSG set to 8 but unfortunatily it wasn't actually stable. This stability issue disappeared after increasing tRFC2 and tRFCPB to 520

1669314613872.png


Main timings: CL32, 40, 38, 38, 72

Unfortunatily it failed the next test

1669329802580.png
 

Admin

Administrator
Moderator
Messages
3,032
#16
6600 MT/s stable with A-die
Someone else got 2x32 GiB stable at 6600 via the following settings

DRAM VDDQ 1.42v
CPU SA 1.15v
CPU VDD2 1.34v
CPU VDDQ 1.34v

- load your default XMP settings, put memory frequency to 6400 and set primary timings to CL34-40-40-96

- change tRFC to 520 (lower values can lead to errors when the memory temperature is higher, you will optimize this later if you pass the long test)
I'm currently testing a config at 6600. Yesterday It failed after 7 hours of Karhu, now testing with the same settings but with a bit higher voltage.

I also found out that CPU VDDQ 1.200v works very well with my memory.
This indicates that the board itself can actually do 6600 but that you might not reach that due to m-die having trouble being stable at 6600 and this will be even worse when you use dual-rank in part due to having twice the heat to cool off for the same voltage and frequency.

The individual later found out that he/she had somehow gotten A-die when buying this kit
My DIMM temperatures never go over 45C. I log them with hwinfo through the whole test. If an error occurs at some point during the test I just increase the tRFC with 10 cycles and it gets stable. The same was when I was testing at 6400. I couldn't explain the reason for that. I also couldn't max out the tREFI. After I found out that it is an A-die everything started to make sense. A-die can't stand low tRFC, it has also likes lower tREFI. On M-die, you can lower the tRFC by around 100 cycles and boost tREFI to 130560 or even 262143. I couldn't do that.
It doesn't make sense for g.skill to use A-die on a 2x32 6000 kit but they may have done that anyway.

The individual later found out that the memory overclock became unstable when the CPU was clocked higher likely due to IMC overheating (the stable temperature can go down as you clock it higher) which is a strong indication that you will be IMC-limited before you are board-limited when running dual rank in gear 2
I'm starting to suspect that IMC overheating is causing those errors which I had after 2 or 3 hours of testing. I read it somewhere in the Russian forums that when the IMC heats up it becomes a source of various instabilities. And I'm inclined to believe that because all the timing misconfiguration errors are always detected very fast, after a minute or two with Karhu, but those strange errors after several hours might be caused by the IMC overheating. This is why at the lowest CPU clock speeds memory is much more stable and when you overclock your CPU it is harder to stabilize it because the temperatures get higher.

Just for the experiment, I bumped my CPU clock from 5.5. to 5.6GHz, and guess what - my memory overclock is no longer stable. When I get back to 5.5GHz, the memory overclock is stable again. I remember also reading this in the Russian overclocking forums but I never paid attention to it. So there is a strong relation between the CPU temperature/speed and memory settings. I am also observing that CPU SA is a very, very important voltage that can improve memory stability.
I'm currently in the second hour of running Karhu with memory clocked at 6600 and CPU ratios bumped 1 step up (5.6 all P cores / 5.9GHz two cores). My CPU VDD2 is set to 1.35v, CPU VDDQ 1.3v, CPU SA 1.25v. Interestingly, what improved stability for this test was the CPU SA which I increased from 1.2 to 1.25. Before that Karhu was failing after 4-5 minutes. I'm still trying to figure out what impact on the stability each of those voltages has, but CPU SA is very important, at least on my system. I need to do some more extensive experiments with this voltage.
 

Admin

Administrator
Moderator
Messages
3,032
#17
6400 stable with loose timings
Lowering the CPU VDDQ to 1.2 made it close to fully stable (using the default CPU clocks/voltages) using the timings i had tuned overrelying on y-cruncher VST, after over 3 hours of testing the FFT test found an error

6400 FFT-fail.PNG


After loosening a bunch of timings i wasn't able to get that error again but that might have been a coincidence, the overclock might still be slightly unstable even with the looser timings. I reduced CPU VDD2 to 1.35 and changed to CL32, 40, 38, 38, 76 after over 5 hours of testing failed to find any error (i first ran over an hour of just FFT and didn't get any error).


1669569158400.png

1669576396820.png

tRAS back to 72:
1669584058225.png


I tried raising tREFI then but didn't get it stable

1669597624181.png


1669621527129.png


Due to the rare nature of the error i went back raising primary timings to CL32,40,40,76 which may have fixed the issue.

Edit: still got an error during 4.57 hours of testing

1669642004416.png
 

Admin

Administrator
Moderator
Messages
3,032
#18
Gear 4 tried
I assumed for over a week that i had bought dual rank M-die but then when i checked the label on the ram it did indicate that i have A-die which does line up with the timing behavior (difficulty increasing tREFI or lowering tRFC)

OR48KRS821A

typhoon1.PNG

I am not sure what this means.

I tried loosening all timings and running the same voltages in gear4 but it wouldn't even boot, raising the DRAM voltage to 1.44 also didn't work. It seems like the gear4 mode is completely useless.
Same here, two days ago I tried 6800 in gear 4 - couldn't boot. I'm getting bored tweaking this memory.
It is worth nothing that i actually got 6800 to boot earlier in gear 2 but of course it wasn't stable.
 

Admin

Administrator
Moderator
Messages
3,032
#19
10 hour stability test passed
I ended up resetting tWRRDSG and tWRRDDG and only after that did it pass the 10 hour test

1669713387298.png


The higher copy speed is too to be changing tREFI to 32247 with tRFC currently at 520 (needlessly high probably).

Code:
Timing   Stable  unstable
tRDRDSG  18
tRDRDSG  8      
tRDRDDR  16
tRDRDDD  16
tWRWRSG  32      
tWRWRDG  8
tWRWRDR  16
tWRWRDD  16
tRDWRSG  21      
tRDWRDG  21      
tRDWRDR  24
tRDWRDD  24
tWRRDSG  70      64
tWRRDDG  48    
tWRRDDR  16
tWRRDDD  16
 
Top