9800x3D - 6400 CL32 1:1 not stable

[berkekolik]

Hello,
I have Gskill DDR5 Trident z5 rgb 6400 32gbX2 CL32
CL32-39-39-102 1.40V
DOCP1 Settings in bios

MB: Asus rog x870e-e gaming wifi
Chip: 9800x3D

When I increase the ram to 6400 it works in 1:2 ratio.
Also 6000 - 2000 fclk 1:1 stable.
When I change it to 6400 1:1 FCLK 2000/2133/2200 etc. the system becomes unstable OCCT test gives too many errors. How can I use my 6400 RAM in 6400 without any problems, please help. I don't want to use it in 1:2 6400.

This stable:

This is not stable:

 

[phanbuey]

Same here - not all IMC can do 6400 1:1

Best is to tighten timings to Cl30-36-36-36 6200 and go 1:1 - same performance essentially.

 

[ShrimpBrime]

VDD/Q try 1.45v

Use more fans, cool the memory directly, see if that helps.

 

[SyWillCry]

Try 1.3V Vsoc. If it cannot do 6400 1:1 at that, it will not do 6400 ever.

 

[ir_cow]

The FLCK is probably not stable at 2133. Try 2000Mhz instead

 

[RootinTootinPootin]

your VDDIO is too much for just 6400 MT's, more isn't better or guarantee stability, if this is on AUTO, then try lowering it manually, at the same time increase your memory VDD and VDDQ, and just try using 2000FCLK 1st to sort things out, lastly your ODT's are a mess..

 

[berkekolik]

your VDDIO is too much for just 6400 MT's, more isn't better or guarantee stability, if this is on AUTO, then try lowering it manually, at the same time increase your memory VDD and VDDQ, and just try using 2000FCLK 1st to sort things out, lastly your ODT's are a mess..

All settings are automatic, I don't know how to set it manually. If you help me, I can change it.

 

[berkekolik]

The FLCK is probably not stable at 2133. Try 2000Mhz instead

Still not stable.

Try 1.3V Vsoc. If it cannot do 6400 1:1 at that, it will not do 6400 ever.

When I try CPU Soc manuel set, Bios says u must use Cpu_OV pin.

 

[_roman_]

Do not be lazy. There are plenty of guides. I read pletny of am4 ryzen guides in the past. That konwledge still was usful with my Am5.

Just because it boots and has POST screen does not imply a stable overclock.

Overclocking is out of warranty, I hope you can deal with the costs and issues.

6000 MT/s is also not guaranteed and supported as far as i know.

 

[berkekolik]

Do not be lazy. There are plenty of guides. I read pletny of am4 ryzen guides in the past. That konwledge still was usful with my Am5.

Just because it boots and has POST screen does not imply a stable overclock.

Overclocking is out of warranty, I hope you can deal with the costs and issues.

6000 MT/s is also not guaranteed and supported as far as i know.

So your suggestion is to stay at 6000?

 

[ir_cow]

View attachment 406862

Still not stable.

You have extremely bad luck. Bottom of the barrel IMC. It happens. Two options left. Crank up the CPU voltages or lower to 6200.

I am also assuming this is 2x 32GB and matching set in the correct slots.

 

[berkekolik]

You have extremely bad luck. Bottom of the barrel IMC. It happens. Two options left. Crank up the CPU voltages or lower to 6200.

I am also assuming this is 2x 32GB and matching set in the correct slots.

Yes 2-4th slot and 2x 32gb

 

[Cowboystrekk]

Dual rank needs more soc voltage than single rank. I need 50mv more on my dr kit vs sr kit. I would go for 6200 at highest stable fclk. Check with linpack large if you get regression. You may get lucky and do 2200 or unlucky and barely do 2100. I would then tune timings. Try Rcd and rp 38 or 37, rfc 512, refi 50000.

 

[Waldorf]

@berkekolik
ryzen like low timings, and you can easily make up for "less" clocks (as long as it is 1:1), by tuning (secondary/rfc) timings.

 

[_roman_]

So your suggestion is to stay at 6000?

You may bother reading about stable testing. And to get a feeling what is a most likely work setting in uefi and what is just someone was lucky setting.

5600 MT/s or 6000 MT/s as a base point with stock settings with 1.25V DC. Stable test.
Than #14 suggests reduce the other settings and check the impact. Not all reduced values will improve performance or will keep stability.

 

[dgianstefani]

All settings are automatic, I don't know how to set it manually. If you help me, I can change it.

There is zero chance you will get 6400 1:1 2x 32 GB with good timings on auto settings with a four slot motherboard, without it being unstable.

There is essentially low chance you will get 6400 1:1 at all with that setup.

The difference between 6400 and 6000 1:1 is essentially meaningless for actual games with a 9800X3D, unless you play at 1440p with a 5090, especially if your FCLK isn't also raised to be in sync, @ir_cow suggestion of lowering it would likely work though, alongside some manual voltage and stability tuning if you have a week to spare. For productivity it can matter slightly.

For 2x32 on a four slot motherboard you should be thankful 6000 1:1 works at all on auto voltages.

Focus on finding a nice 1.2 V or lower all core OC, for instance I can do 5.5 GHz locked on 1.2 AC, and 5.6 in winter with 1.22 V. You'll see more consistent performance gains than chasing 400 MT on a CPU where memory is no longer the bottleneck most of the time.

These chips are rated for 5600 MT, anything beyond that is luck.

You will get far more performance from running stable MT, with subtimings tweak.

Bump your TREFI up and cut down your TRFC.

TRAS and TRC are also laughably high.

This is the current tune I play around with. I can do 6000 MT with a little more voltage but I'm experimenting with 1.435 VDD.

Note, this setting gives significantly better gaming memory performance than your 6400 setup with much looser timings.

Experimenting with Nitro settings and other advanced mobo settings can also help with stability.

Also, the limited on die ECC will actually correct many of the errors that start to show up in a slightly unstable build, but the result will actually be lower performance even as you push memory further. You can only tell by actually benchmarking the tune, not just it passing a stability test. Stability is also highly dependant on temperature, especially with tREFI. So what is "stable" for a reboot, quick stability test, will likely not be "stable" for a several hour gaming session with the GPU putting out hundreds of watts of heat.

If you want to be sure something is "stable", 12/24 hrs of Karhu, plus a day of full GPU load while monitoring performance and WHEA errors.

Or, you could be happy with 6000 MT 1:1, which is actually good for a 2x 32 setup on a four slot mobo with auto settings.

Bear in mind, you have to do stability testing each time you change a setting.

 

[berkekolik]

There is zero chance you will get 6400 1:1 2x 32 GB with good timings on auto settings with a four slot motherboard, without it being unstable.

There is essentially low chance you will get 6400 1:1 at all with that setup.

The difference between 6400 and 6000 1:1 is essentially meaningless for actual games with a 9800X3D, unless you play at 1440p with a 5090, especially if your FCLK isn't also raised to be in sync, @ir_cow suggestion of lowering it would likely work though, alongside some manual voltage and stability tuning if you have a week to spare. For productivity it can matter slightly.

For 2x32 on a four slot motherboard you should be thankful 6000 1:1 works at all on auto voltages.

Focus on finding a nice 1.2 V or lower all core OC, for instance I can do 5.5 GHz locked on 1.2 AC, and 5.6 in winter with 1.22 V. You'll see more consistent performance gains than chasing 400 MT on a CPU where memory is no longer the bottleneck most of the time.

These chips are rated for 5600 MT, anything beyond that is luck.

You will get far more performance from running stable MT, with subtimings tweak.

Bump your TREFI up and cut down your TRFC.

TRAS and TRC are also laughably high.

This is the current tune I play around with. I can do 6000 MT with a little more voltage but I'm experimenting with 1.435 VDD.

View attachment 407634

Note, this setting gives significantly better gaming memory performance than your 6400 setup with much looser timings.

Experimenting with Nitro settings and other advanced mobo settings can also help with stability.

Also, the limited on die ECC will actually correct many of the errors that start to show up in a slightly unstable build, but the result will actually be lower performance even as you push memory further. You can only tell by actually benchmarking the tune, not just it passing a stability test. Stability is also highly dependant on temperature, especially with tREFI. So what is "stable" for a reboot, quick stability test, will likely not be "stable" for a several hour gaming session with the GPU putting out hundreds of watts of heat.

If you want to be sure something is "stable", 12/24 hrs of Karhu, plus a day of full GPU load while monitoring performance and WHEA errors.

Or, you could be happy with 6000 MT 1:1, which is actually good for a 2x 32 setup on a four slot mobo with auto settings.

Bear in mind, you have to do stability testing each time you change a setting.

I have Msi 5090 Suprim, I just play games 4K. Now I am using 6000mhz 1:1 2033 fclk. There is no problem any games without Valorant. And sometimes when I press shutdown, windows freezing. What do you prefer, 6400 1:2 or stay in 6000 1:1 ?

 

[freeagent]

6000 1:1.

 

[dgianstefani]

I have Msi 5090 Suprim, I just play games 4K. Now I am using 6000mhz 1:1 2033 fclk. There is no problem any games without Valorant. And sometimes when I press shutdown, windows freezing. What do you prefer, 6400 1:2 or stay in 6000 1:1 ?

At 4K it's irrelevant.

But 6000 1:1 2000 FCLK. Tune the relevant subtimings and test stability.

 

[berkekolik]

At 4K it's irrelevant.

But 6000 1:1 2000 FCLK. Tune the relevant subtimings and test stability.

Now I changed 6200 1:1 there is no error in test. Looks stable. Can I continue with this?

 

[dgianstefani]

Now I changed 6200 1:1 there is no error in test. Looks stable. Can I continue with this?

If you're happy with that and it actually is stable, why not?

 

[phanbuey]

there is minimal difference at the same timings between 6200 and 6400 1:1 - if on 6200 you can keep CL30 - 36 - 36 timings, then it's actually a little bit faster from my experience than 6400 CL32 1:1.

But either way, 6000 1:1 w/ 2000 flclk is 98% there.

 

[berkekolik]

there is minimal difference at the same timings between 6200 and 6400 1:1 - if on 6200 you can keep CL30 - 36 - 36 timings, then it's actually a little bit faster from my experience than 6400 CL32 1:1.

But either way, 6000 1:1 w/ 2000 flclk is 98% there.

If there is no exaggerated difference, it seems more logical to stay at 6000

 

[freeagent]

Just a little lower latency, and a little extra bandwidth. Probably not enough to notice in a daily situation.