Cannot enable 4000MHz CL16 on Z790

[yoadknux]

Just an update over time. 4000MHz 16-16-16 1.45V in Gear 1 with manual timings was Karhu and Memtest stable, but fails Prime95 large FFT quickly and consistently. On the other hand 3600 14-14-14 is Prime stable, so I just stayed with it. I guess maybe I could try 3800 14-16-16 at 1.5V but RAM oc is such a tedious process.

 

[EarthDog]

but fails Prime95 large FFT quickly and consistently.

So do most things. If there's one test I never run, it's that one, lol. There comes a time when you need to use your PC as you normally would to see if it's stable. 90% of the time that failed, I was stable doing daily activity... unless you're running those stress test-like loads or your work needs that stability, it's can be a waste of time to test that thoroughly.

but RAM oc is such a tedious process.

I gave up years ago. The amount of effort versus returns was never worth it (to me) outside of the benchmarking realm. It's easier to buy the speed you need with XMP and go from there (though more costly). Looks like you made it though!!!

 

[magellan]

@Woomack and ED
Is the OP's issue w/his RAM sticks, his motherboard or CPU? It's a Raptorlake, the last DDR4 capable CPU from Intel and it can't get 4000Mhz.? That doesn't make any sense considering there are 8700k's that can get to 4000Mhz. and more.

For the original OP (yoadknux), you might consider looking for other people who own Aorus Elite Z790 motherboards and look at their BIOS settings for RAM sticks of similar density/size and speed and see if you can get a template of their settings. Alternatively, if Gigabyte has forums dedicated to their motherboards you could look there for more specific help w/your model+make of motherboard.

Just for chuckles, what are your IOL's/RTL's set at? Can you set RTL init or IOL init?

 

[EarthDog]

@Woomack and ED
Is the OP's issue w/his RAM sticks, his motherboard or CPU?

See post #7.

 

[yoadknux]

@Woomack and ED
Is the OP's issue w/his RAM sticks, his motherboard or CPU? It's a Raptorlake, the last DDR4 capable CPU from Intel and it can't get 4000Mhz.? That doesn't make any sense considering there are 8700k's that can get to 4000Mhz. and more.

For the original OP (yoadknux), you might consider looking for other people who own Aorus Elite Z790 motherboards and look at their BIOS settings for RAM sticks of similar density/size and speed and see if you can get a template of their settings. Alternatively, if Gigabyte has forums dedicated to their motherboards you could look there for more specific help w/your model+make of motherboard.

Just for chuckles, what are your IOL's/RTL's set at? Can you set RTL init or IOL init?

I spent a lot of time on this and wrote a "journal" in this post so I'll summarize it for you:

1. Woomack was on point with this, Z690+Z790 are just not DDR4-optimized platforms. Buildzoid also once said "XMP 4000 just generally doesn't work on all 12th gen CPUs. Anything rated over 3600 is matter of luck or skill."
2. I've found a lot of threads with Aorus Z690 and Z790 owners that had issues with XMP. Both with DDR4 and DDR5.
3. My conclusion is that it's a BIOS problem. The main issue was that default XMP set 4000 16-16-16 with Command Rate 1T. That's not gonna boot on almost anything. Even 4000 16-16-16 with 2T probably needs manual tweaking of VCCSA and RAM voltage.
4. My board can do 4000MHz. It's just a question of timings. My older Neo kit runs 100% XMP stable at 4000C18, no tweaking. This new Royal kit runs 4000C16 with tweaking, it just fails a particular stress test and passes the others. I've found stability on that particular test with 3600C14 and 3800C15.
5. I also think G-Skill released an overly-binned kit: 1.4V is low for 2x16 4000C16. I had more stability with 1.45V. For example, the kit in your signature (4222C17) is rated for 1.5V - I never went that high. Patriot kits are also 1.45V+ at these timings.
6. All my tests were under Gear 1, meaning memory frequency = controller frequency. Supposedly Gear 2 allows higher bandwidth at the cost of latency and I didn't want that. I guess speeds like 4200 would be achievable there. There's also a chance that my 13600k has worse IMC than 13700k/13900k.

So it's really the motherboard, but this RAM is also very picky

 

[yoadknux]

So do most things. If there's one test I never run, it's that one, lol. There comes a time when you need to use your PC as you normally would to see if it's stable. 90% of the time that failed, I was stable doing daily activity... unless you're running those stress test-like loads or your work needs that stability, it's can be a waste of time to test that thoroughly.


I gave up years ago. The amount of effort versus returns was never worth it (to me) outside of the benchmarking realm. It's easier to buy the speed you need with XMP and go from there (though more costly). Looks like you made it though!!!

I really agree with you on that. Honestly I wanted things to be exactly like you wrote: buy fast memory with XMP, enable, don't touch. But it just didn't work for me. I had to manual tweak otherwise it would be an expensive kit that sits unused in some drawer. And when you start doing it, you may as well do it right, that's why I kept posting updates. In the end it turned out fine.

 

[EarthDog]

I had to manual tweak otherwise it would be an expensive kit that sits unused in some drawer.

That's an interesting way to think about it... if you run the sticks rated speed and don't overclock them, they're as good as being sitting unused in a drawer?????!!

 

[yoadknux]

That's an interesting way to think about it... if you run the sticks rated speed and don't overclock them, they're as good as being sitting unused in a drawer?????!!

If I can't enable XMP with a kit, then yeah, either I resolve the issue with manual timing (like I did) or it goes into some box. A 4000C16 kit without XMP defaults at 2133C15? That's useless.

 

[EarthDog]

If I can't enable XMP with a kit, then yeah, either I resolve the issue with manual timing (like I did) or it goes into some box. A 4000C16 kit without XMP defaults at 2133C15? That's useless.

Oh yeah, forgot you couldn't run XMP out the box. I have similar sticks (2x8gb royale 4000 cl 16) and they ran on that board for testing so it's pretty weird.

You could've returned them for something on the QVL/that's guaranteed to work too, but the royale are pretty...just not on the qvl.

 

[Create.Destroy]

Well, I could have easily helped you if I found this back then lol
Are you still running this kit?

Let me know if you want help with your subtimings and even a few primaries to get proper performance. The ones you lasted on the last page are not great

 

[yoadknux]

Well, I could have easily helped you if I found this back then lol
Are you still running this kit?

Let me know if you want help with your subtimings and even a few primaries to get proper performance. The ones you lasted on the last page are not great

I've been running this for the last few months


DRAM Voltage 1.45V, VCCSA 1.3V, VDDQ I think 1.4 or 1.35V

 

[Create.Destroy]

I've been running this for the last few months
View attachment 367782

DRAM Voltage 1.45V, VCCSA 1.3V, VDDQ I think 1.4 or 1.35V

Oh nice, you tightened a few things. Do you happen to have Asrock Timing Configurator for a better screenshot of the timings? A few are missing here.

Also, was tREFI unstable at 65000? Or mostly plugged in half of max to keep things simpler?

Any reason you would be against raising DRAM Voltage if it'll let you push it further or be more stable? You can push it high for B-die but maybe 1.5v-1.54v would be the limit without having a fan pointing on the dimms to cool them off.

You have a great kit, so personally I would squeeze out what I can for fun and for performance. If you don't want to spend too much time you don't have to push the extreme though haha

 

[yoadknux]

Oh nice, you tightened a few things. Do you happen to have Asrock Timing Configurator for a better screenshot of the timings? A few are missing here.

Also, was tREFI unstable at 65000? Or mostly plugged in half of max to keep things simpler?

Any reason you would be against raising DRAM Voltage if it'll let you push it further or be more stable? You can push it high for B-die but maybe 1.5v-1.54v would be the limit without having a fan pointing on the dimms to cool them off.

You have a great kit, so personally I would squeeze out what I can for fun and for performance. If you don't want to spend too much time you don't have to push the extreme though haha

I think the Asrock only worked with a certain software version or bios version, so I'm using the HWInfo one.
As for tREFI, I didn't really try to over-tigthen. Same with tRFC. Both values can probably improve. I was just happy that I got it to work and put "loose tight" timings if that makes sense.
I don't think it's good to put DRAM Voltage over 1.5V daily on a gaming rig. My 4090 releases a lot of hot air into the case, I wouldn't be surprised if the RAM hits 45-50c after an hour of a gaming session.

 

[EarthDog]

wouldn't be surprised if the RAM hits 45-50c after an hour of a gaming session.

If that's what you're seeing, you have plenty of headroom...

 

[Woomack]

The typical temperature for dual-rank DDR4 at ~1.4V under higher load is ~60C, so if you have 45-50°C then like ED said, there is plenty of headroom. How high it will work without stability issues depends on frequency, timings, and IC quality. My TridentZ Royal works up to 75°C+ without stability issues. Some kits end at 65°C after overclocking.
You can still check how high tREFI will go, as it often works up to the max value in BIOS and it gives more than main timings. I mean, in benchmarks like AIDA64, it's often 8-10ns lower latency and 5GB/s higher bandwidth, because of only this one timing. It's more than going from CL16 to CL14.
Dual-rank Samsung B kits are pretty random. Some dislike more than 1.45V; some won't run at CL14 regardless of timings, and some will hit a wall at ~4133 regardless of timings or voltages. I was passing that with multiple kits. Somehow, single-rank kits are much more predictable, but also perform worse.

You can play with timings, but since you already have 4000 CL16-16-16, you won't gain much from tweaking/overclocking. We are still on Overclockers, so I won't tell you not to touch anything. Save a stable profile in the BIOS and play some with various settings, as it will at least give you the satisfaction of having something optimized, and maybe a bit higher FPS in some games.

 

[yoadknux]

Well, you guys asked nicely, so I thought why not. I changed tREFI to max value - 65535.
It booted with no issues, so a good start.
Haven't had the change to fully test it yet, but so far so good.



I like 4 hours of Karhu, followed by TestMem anta777, then one hour P95 large FFT - we'll see if it survives that.
Voltage is still 1.45V.

 

[Flamethrower1972]

Well, you guys asked nicely, so I thought why not. I changed tREFI to max value - 65535.
It booted with no issues, so a good start.
Haven't had the change to fully test it yet, but so far so good.

View attachment 367789

I like 4 hours of Karhu, followed by TestMem anta777, then one hour P95 large FFT - we'll see if it survives that.
Voltage is still 1.45V.

I hope everything works out for you I know what it's like to get so frustrated and can't solve a problem.

 

[Create.Destroy]

I think the Asrock only worked with a certain software version or bios version, so I'm using the HWInfo one.
As for tREFI, I didn't really try to over-tigthen. Same with tRFC. Both values can probably improve. I was just happy that I got it to work and put "loose tight" timings if that makes sense.
I don't think it's good to put DRAM Voltage over 1.5V daily on a gaming rig. My 4090 releases a lot of hot air into the case, I wouldn't be surprised if the RAM hits 45-50c after an hour of a gaming session.

Most of the time you'll start getting errors at around 60c, so if you are well below that's great. I understand the sentiment, but factually you can run it above 1.5V safely with no issues if your temps are within reason.

I would run 1.5V for now. Change your tFAW to 16 and you can prob turn tRAS down to 32. Your tWTR_S 8 and tWTR_L 4. tWR 12 and tRTP 9. All those 8's can probably be 7s. Can tCKE do 4?

Your biggest hurdle is your IMC, so depending on voltage your limit will be 4000-4266 for Gear 1.

I'm running a 3200 cl14 kit at 4266 cl16-16-16-32. I would push your kit with higher voltage. It's binned well compared to mine.

And revival thread!

Post magically merged: Jul 6, 2024

The typical temperature for dual-rank DDR4 at ~1.4V under higher load is ~60C, so if you have 45-50°C then like ED said, there is plenty of headroom. How high it will work without stability issues depends on frequency, timings, and IC quality. My TridentZ Royal works up to 75°C+ without stability issues. Some kits end at 65°C after overclocking.
You can still check how high tREFI will go, as it often works up to the max value in BIOS and it gives more than main timings. I mean, in benchmarks like AIDA64, it's often 8-10ns lower latency and 5GB/s higher bandwidth, because of only this one timing. It's more than going from CL16 to CL14.
Dual-rank Samsung B kits are pretty random. Some dislike more than 1.45V; some won't run at CL14 regardless of timings, and some will hit a wall at ~4133 regardless of timings or voltages. I was passing that with multiple kits. Somehow, single-rank kits are much more predictable, but also perform worse.

You can play with timings, but since you already have 4000 CL16-16-16, you won't gain much from tweaking/overclocking. We are still on Overclockers, so I won't tell you not to touch anything. Save a stable profile in the BIOS and play some with various settings, as it will at least give you the satisfaction of having something optimized, and maybe a bit higher FPS in some games.

75c and stable is badass lol yeah subtimings will be the real gains here. I'm sure gear 1 is close to the limit unless he got lucky with his IMC.

Post magically merged: Jul 6, 2024

Well, you guys asked nicely, so I thought why not. I changed tREFI to max value - 65535.
It booted with no issues, so a good start.
Haven't had the change to fully test it yet, but so far so good.

View attachment 367789

I like 4 hours of Karhu, followed by TestMem anta777, then one hour P95 large FFT - we'll see if it survives that.
Voltage is still 1.45V.

Also try this for the Asrock Timing Configurator: https://drive.google.com/file/d/1KnfI8jGU7zNxvXMMpupe1KQuoqwDWcJR/view?pli=1

It's a rar so use 7-zip or something. It's supposed to be an updated and proven version.

 

[Woomack]

75c and stable is badass lol yeah subtimings will be the real gains here. I'm sure gear 1 is close to the limit unless he got lucky with his IMC.

It's a G.Skill Royal 4400 CL17. At some point, I had enough of multiple other kits that couldn't pass 4000-4133 or had stability problems at 1.45V+ (two TG kits were throwing errors at more than 1.45V regardless of temps or settings, and both couldn't even post at more than 4133). Since I couldn't get a review sample as all kits went for sale and the amount was really limited, I bought the 4400 kit.
I'm not using DDR4 anymore, but I'm keeping 3 different kits, "just in case".
My experience with various kits (30+ probably, I don't even remember) is that dual-rank Samsung B has a higher tolerance to temps than single-rank, it also runs hotter in general. Maybe it's just my luck, but I had no problems with any dual-rank Samsung B up to at least 70°C. Some lower binned single-rank kits had problems at 65°C and 4000+. Of course, it's a matter of luck. Hynix or Micron is much more predictable.