Overclock.net banner
21 - 40 of 70 Posts
Discussion starter · #21 ·
MSI's LLC numbers go opposite ASUS's. Mode 3 = flat. Mode 4+ starts to introduce Vdroop.
The reason why I'm telling you to start at 3 is because it helps rule things out for now. You can always increase the Vcore and introduce more Vdroop later on.

If those settings can pass y-cruncher's CST with all tests enabled at a reasonable temperature (under 80-85C), you can either push it harder, or try to reduce the Vcore gradually.
Don't go above 80-85C daily though.
What is the best LLC in MSI bios? I am at 6 but everyone either says 3 or 4? I want to understand if it will help me temp wise
 
Discussion starter · #22 ·
Go to CPU settings and disable "CPU Undervolt Protection". This will fix your ring clock being at 3600.
Drop the e-cores to 40x and try LLC Mode 2. Try increasing the VRM switching frequency to 500Hz+.
Try to go from LLC6 to LLC5 and temps almost increase 2/3 degrees.
Why is a good thing to decrease it to LLC Mode 2?
 
Try to go from LLC6 to LLC5 and temps almost increase 2/3 degrees.
Why is a good thing to decrease it to LLC Mode 2?
Not sure about the LLC 2 recommendation. That causes overshoot.

I've found that LLC 5 and higher seems to be harder to stabilize and gives me worse results. So I stick with LLC 4.
 
Discussion starter · #24 ·
Not sure about the LLC 2 recommendation. That causes overshoot.

I've found that LLC 5 and higher seems to be harder to stabilize and gives me worse results. So I stick with LLC 4.
For me with same voltage 1.32v when I switch to LLC6 BIOS MSI to LL5 -- i immediately go to almost 100 degrees.
Just from LLC6 to LLC5.
Maybe I cant go that lower.
 
Discussion starter · #25 ·
You might have a poor sample.

Use BIOS V1.22 from February.
Set Per Core and plug in 51x for all of the P-cores, and 40x for all of the E-cores to start.
Set Ring to 40x.
Set Vcore to Override and plug in 1.28V.
Set LLC to Mode 3 or 4.
Set VCCSA to 1.35V.
Enable XMP and set it to Gear 1.

Run y-cruncher's Component Stress Test with all tests enabled.
If it passes, boost the P-cores to 52x. If not, reduce them to 50x. Repeat.
If it still doesn't pass, raise Vcore or lower the P-cores and/or E-cores. Experiment a bit.
Monitor your BSODs and temps throughout, as they will give you an indication for where the issue lies.

If the tests constantly red error but don't BSOD in particular, there's also a possibility that the IMC is weak and VCCSA might be too low.
You could either boost it to 1.40V and risk degradation, or you could reduce the memory multiplier.

There's a lot for you to do.
This config above fails (LLC MODE 4) in y cruncher N32 test.
cinebench goes to 93 degrees also.
VCCSA affects temps? i left it at 1.35v.
 
Honestly, I think you might have a mounting issue or paste wasn't applied right. With an Arctic, you shouldn't hit that temp at just 1.28V.

You're using the LGA1700 bracket right? Try tightening the screws some more.
 
Discussion starter · #27 ·
Honestly, I think you might have a mounting issue or paste wasn't applied right. With an Arctic, you shouldn't hit that temp at just 1.28V.

You're using the LGA1700 bracket right? Try tightening the screws some more.
Yes, I have the LGA1700 bracket and also the thermaltake stuff. maybe it-s not completely tight?
I can repaste and do it again for sure.
 
Discussion starter · #28 ·
Honestly, I think you might have a mounting issue or paste wasn't applied right. With an Arctic, you shouldn't hit that temp at just 1.28V.

You're using the LGA1700 bracket right? Try tightening the screws some more.
these bing temp jumps also increased the moment I have disabled the "Cpu under voltage protection".
The ring went from 3.6 to 4.0 and the temps spiked.
 
Try y-cruncher's main test by pressing 0 > 1 > 8. If you can pass that a few times, then it sounds like there is a cooling issue.
You'll have to experiment and figure out where the issue is with the cooler.
 
This config above fails (LLC MODE 4) in y cruncher N32 test.
cinebench goes to 93 degrees also.
VCCSA affects temps? i left it at 1.35v.
N32 is RAM or IMC related. It heats up the memory as much as LinX or TM5.
What component of RAM IDK.
SFT (AVX2) is core related. If you made it through SFT but failed on N32 check the memory.
You can also loop SFT independently. Usually if you have a weak core, SFT will always fail on the exact same core every time (very similar to prime95 small FFT AVX or FMA3, without needing to resize the windows). AFAIK, a N32 fail may show random cores failing or the 'coefficient too large' stuff.
 
Discussion starter · #31 ·
N32 is RAM or IMC related. It heats up the memory as much as LinX or TM5.
What component of RAM IDK.
SFT (AVX2) is core related. If you made it through SFT but failed on N32 check the memory.
You can also loop SFT independently. Usually if you have a weak core, SFT will always fail on the exact same core every time (very similar to prime95 small FFT AVX or FMA3, without needing to resize the windows). AFAIK, a N32 fail may show random cores failing or the 'coefficient too large' stuff.
It showed both and yes it is the only case that fails. I will paste and tight again the screws. I also have the thermaltake mount and will tight it again. If then it's the same I will assume it's not an heating issue. I also have another set of ram that can try instead the one I currently have.
 
Okay, then that's evidently too low Vcore.
If you can't even do 5.1 GHz with 1.28V, you have a subpar bin.
You'd need to push something like 1.33V+ just to run 5.2 GHz all-core, but you are already thermally limited.
You could try to disable the E-cores to gain some more thermal headroom, but it won't be by a significant amount.
I think you should just settle for 5.1 GHz and move on.
 
Clock watchdog timeout when running that.
That is an extremely basic test (0-1-8), although a good starting point before you start running sweaty tests. If you aren't passing that, you are extremely unstable.
 
Discussion starter · #36 ·
Okay, then that's evidently too low Vcore.
If you can't even do 5.1 GHz with 1.28V, you have a subpar bin.
You'd need to push something like 1.33V+ just to run 5.2 GHz all-core, but you are already thermally limited.
You could try to disable the E-cores to gain some more thermal headroom, but it won't be by a significant amount.
I think you should just settle for 5.1 GHz and move on.
Probably will settle with this config.
  • Set all P-Cores to 5.1
  • Set all E-Cores to 4.0
  • CPU Ratio Mode: Fixed Mode
  • CPU Ratio Offset When Running AVX: 0
  • CPU Voltage at 1.29 / Override mode
  • Extreme Memory Profile XMP: Enabled
  • Disabled Intel C-State
  • Disabled EIST
  • Disabled Intel Turbo Boost
  • Disabled Intel Speed Shift Technology
  • Disabled TVB Ratio Clipping
  • Disabled TVB Voltage Optimizations
  • Under DigiALL Power Set LLC > MODE 6

It's not even breaking 82 degrees.5.1.
The only thing i dont get is the LLC mode stuff? I cant run a LLC4 because it will go to 90s. but with LLC MODE as 6 i dont even pass 82 degrees.

Why is that?
 
Probably will settle with this config.
  • Set all P-Cores to 5.1
  • Set all E-Cores to 4.0
  • CPU Ratio Mode: Fixed Mode
  • CPU Ratio Offset When Running AVX: 0
  • CPU Voltage at 1.29 / Override mode
  • Extreme Memory Profile XMP: Enabled
  • Disabled Intel C-State
  • Disabled EIST
  • Disabled Intel Turbo Boost
  • Disabled Intel Speed Shift Technology
  • Disabled TVB Ratio Clipping
  • Disabled TVB Voltage Optimizations
  • Under DigiALL Power Set LLC > MODE 6

It's not even breaking 82 degrees.5.1.
The only thing i dont get is the LLC mode stuff? I cant run a LLC4 because it will go to 90s. but with LLC MODE as 6 i dont even pass 82 degrees.

Why is that?
Because it's introducing more Vdroop, and causing Vcore to go below the level of stability required to actually pass y-cruncher.
Also, I usually don't disable Intel Turbo Boost. Screws with the multipliers on load. If you run HWiNFO on the side and check, you might not even hit 5.1 GHz on the cores.

What is your System Agent (SA) Voltage set to?
 
Discussion starter · #39 ·
Because it's introducing more Vdroop, and causing Vcore to go below the level of stability required to actually pass y-cruncher.
Also, I usually don't disable Intel Turbo Boost. Screws with the multipliers on load. If you run HWiNFO on the side and check, you might not even hit 5.1 GHz on the cores.

What is your System Agent (SA) Voltage set to?
Ok what should i try LLC wise? because auto only puts me on 96 or above.
System Agent (SA) Voltage > still in auto. what should i try?
Intel turbo boost - on? ok will test. but my cores dont fluctuate honestly.. always at 5.1
 
Discussion starter · #40 ·
Also
Enable XMP and set it to Gear 1.
nope. only auto, gear 2 works. DDR4 4000 CL19 patriot viper steel (4x8gb).19 19 19 39.1.35v DRAM.
Gear 1 is not ok and returns errors.

The memory doesnt appear as compatible in QVL in the mobo page so maybe i need to replace with another.
 
21 - 40 of 70 Posts