PRO Z890-A WIFI - iGPU hardware video acceleration issue with RAM clocks above 5600MT/s

P

persiz154202d8

Member
Joined
Aug 22, 2024
Messages
69
Recently got a Pro Z890-A WIFI but any RAM clocks above 5600MT/s and NGU clock above 3000MHz breaks the hardware video acceleration.
For example that video drops a lot of frames on 4K, even on 1440p. At 8K is unwatchable and sometimes crashes.

Any voltages, timings or gears doesn't helps.
Tried by removing the old drivers with UDD and reinstalling the new ones but it didn't help either.
RAM clocks at or under 5600MT/s and NGU clock at or under 3000MHz works fine and the video is smooth with no frames drop at any resolutions.
Using the latest 7E32v2A73 BIOS, haven't tried with different ones.

The CPU is 265K, the RAM is CMK64GX5M2B6600C32
 
Last edited:
Sort by date Sort by votes
Hello,

Seems to be some issues with the iGPU and high frequency RAM.

Scroll down to "DDR5 Memory"
www.guru3d.com

Review: Core Ultra 9 285K Processor: Blazing Speed, But Energy Efficiency Falls Short (Page 30)

Intel today unleashes its Core Ultra 200 series processors, we start with two reviews, this one will entail the Core Ultra 9 285K. The processors offer eight performance cores and 16 energy-friendly cores.
www.guru3d.com www.guru3d.com

However, when we switched from dedicated graphics to the processor’s integrated graphics, we began to see minor artefacts on the Windows Desktop. Running games and benchmarks caused significant screen corruption. Please see the example below:
Click to expand...
Experienced reviewers will quickly recognize these artefacts of memory-related corruption. Since the processor's integrated graphics depend heavily on the memory subsystem, the artifacts may be caused by memory running at too high a speed, a faulty memory controller on the processor, or a software bug. Reducing the memory speed to 6400 MHz eliminated the problem. High-frequency CUDIMMs may be placing significant strain on the memory controllers. For an optimal experience, it is advisable to use a 6400 MHz memory kit with low latency timings. This is my personal recommendation. During the initial boot-up, the BIOS performs a memory training process that can take a few minutes. Switching the BIOS to a preferred profile, if compatible, optimizes the memory timings and requires another training cycle. Aside from the integrated graphics issue, the memory kit demonstrated flawless stability otherwise overall and when using a dedicated graphics card. Update: we have replicated this issue on several brands of motherboards, several memory kits and even different processors. Ergo we suspect Intel has some driver work ahead of them, that or the IGP will simply not work with super-fast CuDIMM frequency memory.
Click to expand...
 
Upvote 0 Downvote
I would be happy if 6400MT/s was eliminitaning the issue. The problem is that to solve it I need to reduce the speed to 5600MT/s. Apart from that, the NGU clock above 3000MHz also causes the same issue regardless of the RAM speed.
 
Upvote 0 Downvote
persiz154202d8 said:
the NGU clock above 3000MHz also causes the same issue regardless of the RAM speed.
Click to expand...

The NGU (Next Generation Uncore) default clock is 2.6GHz
Using the 200S Boost profile it can reach 3.2GHz
So the "playing" range here is 5200-6400.
Obviously the "safe" range is 5600-6000.
As I said several times in the past 20 years in this forum: overclocking is a faulty activity. :biggrin:
 
Upvote 0 Downvote
RemusM said:
So the "playing" range here is 5200-6400.
Obviously the "safe" range is 5600-6000.
As I said several times in the past 20 years in this forum: overclocking is a faulty activity. :biggrin:
Click to expand...
So, you're suggesting that it's normal for DDR5 memory to not be able to work properly at more than 5600? Seriously?!
 
Upvote 0 Downvote
persiz154202d8 said:
So, you're suggesting that it's normal for DDR5 memory to not be able to work properly at more than 5600? Seriously?!
Click to expand...

I'm suggesting you should contact Intel and see what they have to say here.
Just one hint for you: https://forum-en.msi.com/index.php?...el-core-ultra-arrow-lake-ddr5-support.373119/
Intel guarantees 5600 for UDIMM and 6400 for CUDIMM.
Your UDIMM (cmk64gx5m2b6600c32) with overclocked, undertimed and overvolted DDR5-5600 chips won't help at all here.
 
Upvote 0 Downvote
Found the problem. It's called SOC SA Voltage. Obviously the voltage on Auto is too high. Seems like classic SA bug. The problem here is that there is no way to monitoring that voltage. Everything under 1.00v works fine. I'm testing it on 0.800v and so far so good.
 
Upvote 0 Downvote
persiz154202d8 said:
Found the problem. It's called SOC SA Voltage. Obviously the voltage on Auto is too high. Seems like classic SA bug. The problem here is that there is no way to monitoring that voltage. Everything under 1.00v works fine. I'm testing it on 0.800v and so far so good.
Click to expand...

Are you running the memory at 4800?
You cannot run DDR5-6600 stable with VSOC = 0.80 ... 1.0V
You need something around 1.20V

P.S.
@citay what's the meaning of your :biggthumbsup: for all kind of stupidities?
 
Upvote 0 Downvote
The snapshot system in the new BIOSes is completely insane. I have no time to deal with this now. There is shot from the phone:
c0Vhe4g.jpeg
 
Upvote 0 Downvote
persiz154202d8 said:
The snapshot system in the new BIOSes is completely insane. I have no time to deal with this now. There is shot from the phone:
c0Vhe4g.jpeg
Click to expand...

VCCSA = 1.10V is not enough for 64GB DDR5-6600.
And in most of the cases all this "tinkering by ear" does not bring anything good.
You don't even know the correct SOC SA and SOC NGU voltages for your system.
If the default computed values are not good, you need to contact MSI.
They will contact Intel and these things will be fixed properly.
Not based on opinions and guesswork.
 
Upvote 0 Downvote
The initial tests are:
y-cruncher - about 30 minutes
karhu - 1000%
TM5 13.1 @1usmus_v3 - 30 minutes
Prime95 Large FFT's - 30 minutes
All passes without errors. Pending more tests in depth.
Thanks for letting me know that VccSA = 1.10V is not enough for 64GB DDR5-6600. And thank you for your help.
 
Upvote 0 Downvote
persiz154202d8 said:
Thanks for letting me know that VccSA = 1.10V is not enough for 64GB DDR5-6600. And thank you for your help.
Click to expand...

Please contact MSI and tell them your findings.
That "blind mode" for SOC SA and SOC NGU voltages is not a good thing anyway.
good luck
 
Upvote 0 Downvote
At the end of the day it turns out to be a great system, as long as the voltages are kept at the low end.
2x32GB Dual rank at @7333MT/s (CMK64GX5M2B6600C32)
khmLJZ0.png

Gg22PeA.png

91VoKYq.png

For some reason the Asrock Timing Configurator bugs the RAM temps in HWiFO.
v2JJ0cm.png


IA CEP - Disabled
CPU Lite Load - Mode 13 (0.800/1.200 mOhm AC/DC LL)
P Cores ratio - Auto
E Cores ratio - Auto
Enhanced turbo - Auto (Disabled)
Ring clock - Auto
D2D - 3300
NGU - 3200
RAM - 7333 (6600 XMP)

VCC Core voltage - Offset -0.140v
P Cores voltage - Offset -0.100v
E Cores voltage - Offset -0.080v
Ring voltage - Offset -0.150v
SOC SA - 0.900v
SOC NGU - 0.950v
VccSA - 1.070v
VnnAON - 0.780v
VDD2 - 1.200v
CPU IO - 1.250v

RAM VDD - Auto (1.400v)
RAM VDDQ - 1.350v
 
Upvote 0 Downvote
Is there a "proper" fix for this already?
2x32GB Kingston Fury with XMP causes the same problem for me.

Could there be correlation that with XMP enabled, memory is showing as quad-channel in hwinfo?
Without XMP, dual-channel, iGPU working fine
265k on a Z890 Tomahawk Wifi

Everything running at default Bios settings as of now, which is stable and no issues.
Only thing I would like to get XMP working without issues.
 
Upvote 0 Downvote
Back
Top