5950x Infinity Fabric Problems

SDNYC09

Limp Gawd
Joined
May 26, 2006
Messages
197
Hello.

I am having an issue with my new 5950x. I have tried three ram kits ( 4000, 3600, 3600 ) at first I thought it was the ram not being able to achieve 4000mhz clock ( having the Infinity fabric at 1800 ) Knowing this isn't 1 to 1 I wanted to test the memory first. After 3 kits and no luck, I clocked the Infinity fabric down to 1600 and tried again, all the ram works at all settings, 4000 mhz 3600mhz, etc. I cannot get the PC to post with any other infinity fabric setting over 1600, no matter the ram speed, I can keep it at stock 2133 and it still will not post. Does anyone have an idea why this will not post with an IF over 1600? Supposedly this chip should do 1800 at least and even 1900-2000 with future AGESA patches.]


My current setup :

Msi MEG Ace x570
Gskill Trident z F4-4000c19d
Also have Rip Jaws - 3600mhz CL16
5950x - on custom water cooling
3090 Msi Trio - Waiting for block
1200 watt corsair ps
 

Brackle

Old Timer
Joined
Jun 19, 2003
Messages
7,719
There is a known bug with the newest MSI bios that will not allow you to run past 1600mhz on the 5000 series CPU's. If you popin a 3000 series you have no issues at all.

I too have the same problem with my 5600x. Once I read the forum thread at overclock.net that the MSI Unify is having this issue, other users also reported the same issue on other MSI motherboards (like the ACE).

https://www.overclock.net/forums/amd-motherboards.11/

Just going to have to wait for fix :/ I would LOVE to see my 3800mhz CL14 memory on these new Ryzens work :(
 

Brackle

Old Timer
Joined
Jun 19, 2003
Messages
7,719
I agree that it is up to you, but I would keep the Ace because it is still a top notch motherboard. Just wait for the bios bugs to be ironed out.
 

Putz

I have a custom title
Joined
Jul 8, 2002
Messages
5,311
I have a 4000 kit as well but its the only 4x8 on the QVL for the Asus X570 ROG Strix and so far I have not be able to get 4000 mhz, also 5950 and the latest Beta bios for a couple days ago

bios showing 1866 memory and IF, Ryzen master is showing 1900IF/1900 (3800 mem)

I have another QVL kit coming in a 2x16 format to try to see if that's any better, was hoping that 4000mhz ram/2000 infinity was going to be attainable but so far not looking that way

@ my 3733 memory I am seeing 55333 read/ 55510 write and , and Cinebench coming it around 9700 all cores score, seems a little low
 

thesmokingman

Supreme [H]ardness
Joined
Nov 22, 2008
Messages
6,435
I have a 4000 kit as well but its the only 4x8 on the QVL for the Asus X570 ROG Strix and so far I have not be able to get 4000 mhz, also 5950 and the latest Beta bios for a couple days ago

bios showing 1866 memory and IF, Ryzen master is showing 1900IF/1900 (3800 mem)

I have another QVL kit coming in a 2x16 format to try to see if that's any better, was hoping that 4000mhz ram/2000 infinity was going to be attainable but so far not looking that way

@ my 3733 memory I am seeing 55333 read/ 55510 write and , and Cinebench coming it around 9700 all cores score, seems a little low

The read/writes don't really matter, it's the latency at any speed. A lot of ppl jumped the gun for 4000mhz ram but that shit takes a lot of fiddling with a mature bios and we are far from a matured bios. You might not have any more success with a different kit...
 

Baker_God

[H]ard|Gawd
Joined
Apr 27, 2003
Messages
1,346
There is a known bug with the newest MSI bios that will not allow you to run past 1600mhz on the 5000 series CPU's. If you popin a 3000 series you have no issues at all.

I too have the same problem with my 5600x. Once I read the forum thread at overclock.net that the MSI Unify is having this issue, other users also reported the same issue on other MSI motherboards (like the ACE).

https://www.overclock.net/forums/amd-motherboards.11/

Just going to have to wait for fix :/ I would LOVE to see my 3800mhz CL14 memory on these new Ryzens work :(
Wow, I'm actually relieved to hear this. I've been pulling my hair out trying to get my UNIFY to post at XMP settings tonight. Glad that it's a known issue and not just my bunk hardware!
 

exlink

Supreme [H]ardness
Joined
Dec 16, 2006
Messages
5,480
Apparently MSI released a new beta bios. People on OCN are hitting 1900 stable on the infinity fabric now. Will hopefully keep getting better with each bios revision.
 

thesmokingman

Supreme [H]ardness
Joined
Nov 22, 2008
Messages
6,435
Apparently MSI released a new beta bios. People on OCN are hitting 1900 stable on the infinity fabric now. Will hopefully keep getting better with each bios revision.

On reflex I checked Asus for a new biois' and notta. :(
 

martyr8

n00b
Joined
Jul 6, 2016
Messages
1
I too am having problems with my ram.

I have Gskill c18 4000mhz 4 x 8gig and an Asus Crosshair 8 Wifi. I can not enable DOCP mode with a stick of ram in the first bank slot in any configuration. If I have 2 sticks installed in bank 2 and 4 I CAN enable DOCP mode. I tried my old ram Gskill c16 3200mhz 4 x 8gig and it works fine in DOCP mode with all 4 sticks.
 

alyjen

Weaksauce
Joined
Mar 24, 2011
Messages
78
FCLK 1:1 requires some tweaking on all motherboards, for some it required BIOS update, e.g. on my Asus TUF Gaming B550 voltages were set by default as attached below (auto voltages are actually 1.1, 1.0, 1.05, 0.9 but there's some voltage drop as LLC still on auto).
I had to tune VDDG CCD to 0.975 to get rid of any WHEA errors in OCCT.
OCCT with small data set + SSE enabled is super fast in picking WHEA errors. I could run TM5 for hours without errors, yet OCCT took less than 5 minutes to generate some. Funny thing, apart from this single WHEA error in heavy load/synthetic benchmarks I had no other stability issue.

I think there's still some room for improvement so I'll stay on my settings and wait for next BIOS

1605599709007.png
 

Astral Abyss

2[H]4U
Joined
Jun 15, 2004
Messages
2,834
FCLK 1:1 requires some tweaking on all motherboards, for some it required BIOS update, e.g. on my Asus TUF Gaming B550 voltages were set by default as attached below (auto voltages are actually 1.1, 1.0, 1.05, 0.9 but there's some voltage drop as LLC still on auto).
I had to tune VDDG CCD to 0.975 to get rid of any WHEA errors in OCCT.
OCCT with small data set + SSE enabled is super fast in picking WHEA errors. I could run TM5 for hours without errors, yet OCCT took less than 5 minutes to generate some. Funny thing, apart from this single WHEA error in heavy load/synthetic benchmarks I had no other stability issue.

I think there's still some room for improvement so I'll stay on my settings and wait for next BIOS

View attachment 300147

No other voltage changes needed? Good ol WHEA error is what I am getting pretty consistently.

I got tired of messing with it and put my 4x8 CAS 14 3200MHz GSkill back in for now.
 

Baker_God

[H]ard|Gawd
Joined
Apr 27, 2003
Messages
1,346
I just grabbed the new beta bios for my UNIFY off of MSI's site. It did the trick, stress testing at XMP speeds right now.
 

alyjen

Weaksauce
Joined
Mar 24, 2011
Messages
78
No other voltage changes needed? Good ol WHEA error is what I am getting pretty consistently.

I got tired of messing with it and put my 4x8 CAS 14 3200MHz GSkill back in for now.
it's either voltages or newer BIOS, my Asus sets all other voltages pretty high so the only one I needed to adjust was VDDG CCD. I can now think of lowering some of the values, but they are somehow connected so if you touch one you may mess up the whole thing.
And yes these 4 are the main culprits when it comes to Infinity Fabric clock stability, VSOC being the main reason followed closely by VDDG
 

Astral Abyss

2[H]4U
Joined
Jun 15, 2004
Messages
2,834
it's either voltages or newer BIOS, my Asus sets all other voltages pretty high so the only one I needed to adjust was VDDG CCD. I can now think of lowering some of the values, but they are somehow connected so if you touch one you may mess up the whole thing.
And yes these 4 are the main culprits when it comes to Infinity Fabric clock stability, VSOC being the main reason followed closely by VDDG
VSOC is the one I've heard the most, but this is good info. Thanks for posting and wish me luck on my way to stability.
 

E4g1e

Supreme [H]ardness
Joined
May 21, 2002
Messages
7,243
If I were you I'd get rid of that POS AMD and go back to intel as I'm sure that 14Nm++++++++ bios is matured by now.
Actually, both companies have been making POS parts for quite some time. Even 14nm+++++++++++++ isn't really designed for a CPU that has more than 6 cores and 12 threads. Intel's newer CPUs have been burning out (overheating) more than they should, even with monster cooling. In fact, I would dare say that some of those might need LN2 cooling just to even run at stock speeds. And that's all because Intel had to resort to cramming in more cores and more threads into a CPU that wasn't designed to have more than 6 cores and 12 threads to begin with, since the company has been having problems with its next-gen 10nm process. Even today, 10nm Intel CPUs are restricted to lower-power, lower-performance mobile parts that have no more than 4 cores and 8 threads.

By the way, the upcoming Rocket Lake (11th-Gen) desktop CPUs will still be produced on a 14 nm process, but it will be a new-from-the-ground-up architecture rather than a tweaked-over-and-over-again variant of Skylake. And as a new architecture, it will be limited to 8 cores and 16 threads on release.
 
Last edited:
Top