• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Cannot enable 4000MHz CL16 on Z790

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

yoadknux

Member
Joined
May 6, 2016
Hello everyone,

I decided to buy a B-Die memory: F4-4000C16D-32GTRSA - G.Skill Trident Z Royal DDR4-4000 CL16-16-16-36 1.40V 32GB (2x16GB) for my Z790 system. Unfortunately I cannot get it to run on the specified speed.

Some background - My CPU is i5 13600k (now at stock), my motherboard is Aorus Z790 Elite DDR4 (https://www.gigabyte.com/Motherboard/Z790-AORUS-ELITE-DDR4-rev-10).
Before the memory upgrade, I ran a very similar kit: F4-4000C18D-32GTZN Trident Z Neo DDR4-4000 CL18-22-22-42 1.40V 32GB (2x16GB). This kit is 24/7 stable, XMP runs perfectly fine right out of the box. This is a Hynix kit.

What happens with the new B-Die kit: when I enable XMP, it resets, takes a few bootless minutes, then shuts down, posts, and supposedly enables the XMP on the BIOS side. However monitoring software (such as HWInfo or CPU-Z) will show the default RAM speed (not XMP).

I always heard crazy things about how good B-Die is in terms of overclocking, and I have a reasonably OK motherboard on the fastest platform that will ever support DDR4, but I can't even get it to XMP-Stable.

I am using the exact same slots on the motherboard that I used with the old kit: #2 and #4 on the motherboard. I tried loading safe settings and re-enabling XMP a few times, no luck. When I try to manually set the speed to 4000MHz I get instant BSOD in Windows.

For now I went back to my old kit, and no surprise, everything works. That's really a bummer because the Royal kit looks so much better than the regular one. Can I get some suggestions on how to get this kit to work?

Here are the kits in question: Top is the old TridentZ Neo, Bottom is the new TridentZ Royal.



Thanks! 29TEabr.jpg
 
Last edited:
Try to add a bit of System Agent voltage. Tighter timings on this kit.

What bios are you on? Spring forward, or, try falling back and see if one works better than the other.

Any review of mine that uses ddr4 uses Trident Royale Z 4000 kit and I don't recall it not working through various bios over time.
 
Try to add a bit of System Agent voltage. Tighter timings on this kit.

What bios are you on? Spring forward, or, try falling back and see if one works better than the other.

Any review of mine that uses ddr4 uses Trident Royale Z 4000 kit and I don't recall it not working through various bios over time.
Thank you so much for the comment.

I just updated the BIOS a few days ago to try to address this issue but didn't see an effect.

I set VCCSA to 1.35V, VCCIN to 2V, VDDQ to 1.4V, and SA PLL Overvoltage to +120mV. Still no luck. Actually the reported RAM speed is one half of what it should be.

Here are some images from what I see at monitoring software and BIOS:

Wibm67X.jpg

fG1XV3Q.jpg

6oapKxk.png

Upqtmkn.png
 
Do bandwidth tests confirm those values? Are you seeing notay less throughput with those windows readings? Weird.
I didn't test with any benchmark because the timings are all wrong too. And I trust both CPU-Z and HWInfo - they show the right timing with my old kit:

sBH0q0F.png rIjVjfu.png

As for the current kit, I managed to boot it at 3600 CL16. But this behavior is really weird, I have a feeling the kit is faulty, because I've never seen anything like this.
 
Clear CMOS. Then enable only the XMP profile, and check how it acts. If it doesn't boot then set manually 3800 or 3866. If it doesn't boot then check 3733. I mean nothing manually, only XMP enabled.
If you won't make it work at XMP settings and something like 3733 then maybe something is with RAM or its memory profile, but I guess that it can be BIOS issue.
I would check if there are any beta BIOS versions as there can be some changes.
You can also leave everything at auto, set VDIMM at 1.45V and go up with the frequency. No XMP or anything else, except memory voltage. Then you will see how high the motherboard is training this memory kit. Maybe it doesn't like something in the XMP profile.
If you won't make it work then simply ask G.Skill support (they react quickly as I heard). I doubt that Gigabyte support will be helpful.
 
Clear CMOS. Then enable only the XMP profile, and check how it acts. If it doesn't boot then set manually 3800 or 3866. If it doesn't boot then check 3733. I mean nothing manually, only XMP enabled.
If you won't make it work at XMP settings and something like 3733 then maybe something is with RAM or its memory profile, but I guess that it can be BIOS issue.
I would check if there are any beta BIOS versions as there can be some changes.
You can also leave everything at auto, set VDIMM at 1.45V and go up with the frequency. No XMP or anything else, except memory voltage. Then you will see how high the motherboard is training this memory kit. Maybe it doesn't like something in the XMP profile.
If you won't make it work then simply ask G.Skill support (they react quickly as I heard). I doubt that Gigabyte support will be helpful.
Thanks Woomack. I followed your steps, and I can post with XMP on, however Windows gives me an endless BSOD loop:
TdGXPxZ.jpg

Some of these are "Memory Management", some are "less or equal", etc
And again - None of this happens with my TridentZ Neo Kit, which is very similar - Gear 1, 4000MHz, CL18, without touching VCCIO/VCCSA

I think I'll try to contact G-Skill support and see what they say.
 
You would figure a Raptorlake CPU shouldn't have any problems running memory at 4000Mhz. or higher.

I still have cold boot issues running my memory at 4000Mhz on a z390+9700k system, but repeatedly going into BIOS and hitting save settings eventually gets the memory to train, it sometimes takes up to three tries to get it to train this way. You might give this a shot w/your Raptorlake.
 
I guess it's more on the motherboard side, as not every BIOS is optimized for a specific IC. It shouldn't happen with Gigabyte, but less popular brands work well almost only with 8GB memory modules. It can also be a RAM problem, so it's better if G.Skill support says what they think about it.
Since Z690/Z790 was released, all manufacturers have been focusing on DDR5 and solving early DDR5 issues on their motherboards. They skipped fine-tuning for DDR4 because of lack of time, and simply DDR5 was one of the main reasons to sell their more expensive motherboards. The last fully DDR4-focused chipset was Z590.
 
Just a brief update. No response yet from G-Skill Support.

a) When I try to put DDR4-4000 MANUALLY I get the following two light going back-and-forth and no post until I reset CMOS:
b) When I enable XMP and change to DDR4-3600 MANUALLY I manage to post and get the following:

43Gh3VI.png

Maybe it's related to the Command Rate 1T?

EDIT:
c) Yes, I manage to boot it with XMP, DDR4-4000, if I change Command Rate from 1T to 2T.
What does this imply then?

zJizM08.png
 
Last edited:
It implies you need to use 2T instead of 1T. :)

What does the SPD tab say. Is it supposed to be 2T at those speeds?


That's the EZ Debug thing. I'm guessing it's flashing between DRAM and CPU?
Yes, that's what the flashes indicate. As for SPD, Command Rate is greyed out, so I don't know if it's supposed to run at 1T or 2T.
I'd like to provide an update: I'm not 100% stable yet but there is improvement.
First of all I found that I need to increase memory voltage from 1.4 (XMP) to 1.45. And I also pushed VCCSA to 1.35V, VCCIO to 2V, VDDQ to 1.4V.
Then I noticed that I have to change Command Rate from 1T to 2T, otherwise I have errors even in lower RAM frequencies.
Gear did not make an important difference here, so I'm running at Gear 1 (Mem Controller Freq = DRAM Freq).
I left it overnight and Karhu RAM Test showed 0 errors:
Pmw9HUj.png

But, on the other hand, follow-up test with P95 Large FFT + AVX had some workers fail, which makes me wonder if I'm really stable or not. Are there any other recommended tests?
 
Yes, that's what the flashes indicate. As for SPD, Command Rate is greyed out, so I don't know if it's supposed to run at 1T or 2T.
I'd like to provide an update: I'm not 100% stable yet but there is improvement.
First of all I found that I need to increase memory voltage from 1.4 (XMP) to 1.45. And I also pushed VCCSA to 1.35V, VCCIO to 2V, VDDQ to 1.4V.
Then I noticed that I have to change Command Rate from 1T to 2T, otherwise I have errors even in lower RAM frequencies.
Gear did not make an important difference here, so I'm running at Gear 1 (Mem Controller Freq = DRAM Freq).
I left it overnight and Karhu RAM Test showed 0 errors:
View attachment 362137

But, on the other hand, follow-up test with P95 Large FFT + AVX had some workers fail, which makes me wonder if I'm really stable or not. Are there any other recommended tests?

CR 1T usually doesn't work above 3600-3733. It depends on the motherboard. Some top OC motherboards run at 4000-4200 max. It's still weird, as full auto/XMP settings should have CR2 by default. I haven't seen any DDR4 with CR1 in XMP. Probably it's a BIOS issue, but you figured it out.

You can check higher SA voltage or something like 16-17-17 or 16-18-18. Newer IC/last top G.Skill was usually in 4000 kits at 16-18-18/16-19-19, while older IC was at 17-18-18. I'm not sure what was the real difference in IC, but this was also how it was scaling. I mean, tRCD/tRP may need to be higher than CL.
I also had problems stabilizing 4000 below some voltage on some kits. For example, I had to set 1.45V only to make 4000 work on the Team Group kit. Regardless if it was CL15, 16, 17, or 18, it needed 1.45V.
Gear 1 will probably run 1-2 ratios higher. Gear 2 is not worth it on Samsung IC.
In the end, you have to check it yourself as there are many variables.
 
CR 1T usually doesn't work above 3600-3733. It depends on the motherboard. Some top OC motherboards run at 4000-4200 max. It's still weird, as full auto/XMP settings should have CR2 by default. I haven't seen any DDR4 with CR1 in XMP. Probably it's a BIOS issue, but you figured it out.

You can check higher SA voltage or something like 16-17-17 or 16-18-18. Newer IC/last top G.Skill was usually in 4000 kits at 16-18-18/16-19-19, while older IC was at 17-18-18. I'm not sure what was the real difference in IC, but this was also how it was scaling. I mean, tRCD/tRP may need to be higher than CL.
I also had problems stabilizing 4000 below some voltage on some kits. For example, I had to set 1.45V only to make 4000 work on the Team Group kit. Regardless if it was CL15, 16, 17, or 18, it needed 1.45V.
Gear 1 will probably run 1-2 ratios higher. Gear 2 is not worth it on Samsung IC.
In the end, you have to check it yourself as there are many variables.
Thanks Woomack. So, it's not unusual to customize all the timings on these type of kits?
Do XMP timings generally expect Gear 1 or Gear 2? Is it better to have loose timings on Gear 1 compared to strict timings on Gear 2?

EDIT:
OK. Nevermind. I am now almost certain that this TridentZ Royal is faulty. I will return it.
I reached this conclusion by manually putting the timings from my old kit to the new kit.
Just a reminder, my old kit (TridentZ Neo) 100% XMP stable at 18-22-22-42-2T.
My new kit (TridentZ Royal) is supposed to run at 16-16-16-36-52 and it cannot even manage the timings from my old kit without giving an error.
DtjZRd1.png

Thanks for all the help guys
 
Last edited:
OK, I found out what's the issue. This RAM actually taught me so much about my BIOS settings and modern RAM overclocking. So it goes like this - some of this stuff I already shared:

a) On this board, on rare cases, some voltage settings (VCCSA, VDDQ) are not correctly applied. I had an instance where after loading factory default (no XMP), and manually setting VCCSA to 1.35V and VDDQ to 1.4V the board would not boot.
b) Auto secondary timings are sometimes applied and sometimes not after enabling XMP. I've had an instance where tRFC, tREFI, tCWL and other timings kept their "stock" value after applying XMP.
c) This combination of RAM + motherboard sets the default Command Rate to 1T which is very unstable without tweaking

In order to get XMP to work correctly, I first apply the voltages manually, set Command Rate to 2T, then apply XMP and go straight to BIOS to see that secondary timings and voltages were applied correctly.
XMP is generally stable. However since this requires so much tweaking then it loses the purpose of XMP, so I just manually apply all the timings.

My settings are as follows:
  • VDDQ 1.4V
  • DRAM Voltage 1.45V
  • VCCSA 1.35V
  • VCCIN 2V
And the following timings:
  • CAS Latency 16
  • tRCD 16
  • tRP 16
  • tRAS 36
  • tRC 52
  • tWR 24
  • tCWL 16
  • tRRD_S 4
  • tRRD_L 10
  • tWTR_S 5
  • tWTR_L 16
  • tCCD_S 4
  • tCCD_L 8
  • tRFC 700
  • tRTP 12
  • tFAW 48
  • Command Rate 2
  • tREFI 15600
  • Memory Gear 1
I managed to pass 8 hours of Karhu memory test (with Cache and FPU) and 3 cycles of TestMem5 (Extreme1@anta777). So, I think it's stable, but time will tell.
I guess the next steps are to play with other timings like tRFC and tREFI. But other than that, I don't think there's much more to be done. From online research, 2x16 Dual-rank Gear 1 for 24/7 stability peaks at ~4200. So I can try something like 4200 CL16 or 4000 CL14. But it's gonna require more voltage and LONG stress tests for little gain.

Here's what I'm currently getting on AIDA64:

0yEYp1t.png
 
Last edited:
Memory gear in this case refers to the base memory clock right? For my z390 I have two choices 133Mhz. and 100Mhz. are these the gears referred to in this thread?
Is VCCIN the same thing as VCCIO for z390 and z370? Or is VDDQ the same thing as VCCIO?

Don't raptor lake CPU's usually get better than 4000Mhz. on their memory sticks? Or is that only for < 32GiB?

Could yoadknux get potentially higher frequencies if he went with 4x8GiB instead of 2x16GiB sticks?
 
are these the gears referred to in this thread?
This was something new added to platforms after Z390(?). Memory controller and memory module ratio. Most ran 1:1 below ~3600 or so, then would flip at higher speeds (and possibly capacities).

Is VCCIN the same thing as VCCIO for z390 and z370? Or is VDDQ the same thing as VCCIO?
VCCin is the input voltage to the FIVR. VCCio is the ring/cache voltage. VDDq is a (signal) voltage on the PMIC of DDR5. They don't change the name of X to Y for giggles. :)

Don't raptor lake CPU's usually get better than 4000Mhz. on their memory sticks? Or is that only for < 32GiB?
Yes x2.

ould yoadknux get potentially higher frequencies if he went with 4x8GiB instead of 2x16GiB sticks?
Woomack covered that with you already.....but in a different thread. :)

 
Last edited:
Back