Experience with STRIX Z490-E Gaming / Maximus XII Hero / i9-10900k / 4000 MHz DDR4

[P-h-a-s-e]

Hi,

Here we go..

Summary

• For OC value, stick with the Strix and pass on the Maximus.
  
• If you want 4000 MHz get 2 sticks, not4
  

A bit more detail based on the past week or two:

The STRIX Z490-E Gaming was a better overclocker than the Maximus XII Hero (board silicon lottery?). The Strix was stable at 5.2 GHz at 1.35V and the Maximus needed 1.38 V (chip, cooler, and memory).

4 sticks of 4000 MHz is not a good idea with these boards. May be a chipset thing -- someone mentioned something something "daisy chain" in another forum. The Strix, with the timings and settings from XMP2, MCH Fast Boot enabled, Round Trip Latency enabled, and BIOS 0707 managed 3600 MHz -- DRAM, VCCIO and SA voltages had no effect (went up to 1.45, 1.25, 1.25, then backed down to 1.39375, auto, auto). The Maximus with timings 14-14-28, 1.39375, and BIOS 0607 managed 3400 MHz (that also worked for the Strix). I did not try the Maximus 0707 BIOS because when I did that with the Strix, coincidently(?) I had an SSD M2 crash (being RMA'ed). I do not know if the crash had anything to do with the BIOS, but I'll skip 0707 just in case.

I kept the Maximus on the case (differences did not justify to swap the strix back).

I lowered the voltage to 1.36 with LLC 6 for daily usage. The final settings are:

AI Overclock Tuner: Manual
ASUS Multicore Enhancement: Enabled
AUX Instruction Core Ratio Negative Offset: 0
CPU Core Ratio Synch: All Cores
All-Core Ratio Limit: 52
BCLK Aware Adaptative Voltage: Disabled
CPU Core/Cash Voltage: Manual Mode
CPU Core Voltage Override: 1.36 (down from 1.38 needed for Cinebench R20 stability)
DRAM Frequency: 3400
DRAM Voltage: 1.39375
Min. CPU Cache Ratio: 50
Max. CPU Cache Ratio: 50
CPU Core/Cache Current Limit Max.: 255.75
Long Duration Package Powe Limit: 4095
Short Duration Package Powe Limit: 4095
CPU Load Line Calibration: Level 6 (down from Level 7 needed for Cinebench R20 stability)
DRAM CAS# Latency: 14
DRAM RAS# to CAS# Delay: 14
DRAM RAS# ACT Time: 28
VMX: Enabled (not sure if it makes a difference, but I need it)

I could not get to 5.3 with AUX -1 and CPU voltage up to 1.38. R20 would freeze in the first iteration and HWInfo was showing voltages that seemed high for a daily driver -- don't recall exactly what they were.

R20 score is now 6834 -- though it is always a PITA to figure our which process to kill and services to disable. Both ASUS and NZXT-CAM software affect the score significantly. CAM must be on for the pump/fans to ramp up... so you kind of have to run a pre-test to rev things up then kill CAM to keep the cooling to the max while you run the actual benchmark.

[Falkentyne]

There is nothing in the BIOS called "AUX". Please check your spelling.

What is VMX? Why do you need it? Is that something for running VMware or virtual OS?

[Arne_Saknussemm]

Mixing kits even of "same" type is never going to help with high frequency RAM

My experience with these chips leaves me a bit nonplussed with the IMC...however 4000 I've got running on several

1.2+ IO and SA help with maybe 1.4 or 1.45 DRAM voltage

You would be very lucky to run 5.3GHz stable on a 10900K...maybe cinebench R15 would pass...but 20 will need AVX -2

Hero XII a better performer for me than the Strix E

[P-h-a-s-e]

Well I decided to return the 2x16 kit that is still under the return period, get a 2x32 kit and then sell the remaining 2x16 kit.

Arne Saknussemm wrote:
Mixing kits even of "same" type is never going to help with high frequency RAM

I heard that before, unfortunately I heard it after I decided to get a second kit. The cause, however, may be the daisy chaining, that while allowing the higher speeds for 2 sticks does not work that well with 4. Coincidentally "mixing" kits gives you 4 sticks. So whether mixing is a factor or not 4 sticks allegedly puts you at a greater disadvantage with the Z490 chipset, I'm told by a guy on the gSkill forums.

Arne Saknussemm wrote:
Mixing kits even of "same" type is never going to help with high frequency RAM

My experience with these chips leaves me a bit nonplussed with the IMC...however 4000 I've got running on several

1.2+ IO and SA help with maybe 1.4 or 1.45 DRAM voltage

Curious, was that with a Z490 chispset and 4 sticks? With those voltages I did not get 3800. I did get 3600 on the STRIX 0707, but not on the Maximux 0607. However, I back down the voltages on the STRIX and it hit the same 3600 at 3.9375 /auto /auto.

Arne Saknussemm wrote:
You would be very lucky to run 5.3GHz stable on a 10900K...maybe cinebench R15 would pass...but 20 will need AVX -2

Hero XII a better performer for me than the Strix E

Yeah, that is not going to happen. I had to go back to 1.37V and Level 7 to get 5.2 GHz stable on the Maximus. That is 0.02V more than the Strix, which, for all I know may be delivering the same power delivery to the CPU. Well actually 0.03V -- with R20 I had to bump it to 1.38V to get it pass 20 mins, while on the STRIX it took 1.35V, both at level 7.

[P-h-a-s-e]

Called gSkill and they confirmed that the 4-dimm speed reduction is likely due to the daisy chain technology. It is implemented in all ASUS Z490 mobos. I did not check other manufacturers, but the initial finding remains. If you to maximize the memory speed stick with 2 sticks and pass on 4.

BTW, none of the 2x32 GB in the QVL is out yet. gSkill said they are likely to hit the US market, just not yet.

[P-h-a-s-e]

A quick update...

I looked up LLC a bit.

At 5.2 GHz 1.37V LL7, according to the 1520mv - 1.1 * IOUT formula, I was 200 mv over spec with temps 90-91 C and a VID of 1518mv at 182 Amps. So I brought it down to 5.1 GHz 1.35V and LLC 4 and that made the CineBench R20 VID 1344mv and the current to 151 Amps. Which is just about within spec (1520mv - 1.1 * 151 = 1354mv).

To my surprise the temps are 20C lower, with CAMS reading 69-71 C. The R20 score went from 6744 to 6675 -- a 1% loss.

I'm gonna call good now.

[Arne_Saknussemm]

P-h-a-s-e wrote:
Curious, was that with a Z490 chispset and 4 sticks?

Yes...MXIIHero...4000 18-18-18-38 1.45Dram 1.27 SA and IO single 32 GB kit...passes Karhu RamTest

Obviously this assumes the RAM is good and the IMC semi capable

[P-h-a-s-e]

Arne Saknussemm wrote:
Yes...MXIIHero...4000 18-18-18-38 1.45Dram 1.27 SA and IO single 32 GB kit...passes Karhu RamTest

Obviously this assumes the RAM is good and the IMC semi capable

Ah, okay. that makes sense. If you look at the QVL you see 2-stick kits for higher speeds than 4-stick kits. I would guess that it is not only a market segmentation issue, but that there are technical challenges... i.e. 4-sticks at a given speed is inherently different/harder-to-make than 2-stiks for that same speed,

In the attached you see the highest speed for 4x8GB is 4500... yet there are a bunch of 2x8 GB kits for higher speeds, up to 4800 .


Likewise, for the 64GB total I was looking for. The highest 4x16GB kit is 3600 (which is the same speed we got with a bit of work for two 2x16GB 4000 kits on the ROG SKILL). However, if you want higher than that you have to go back to 2 sticks, and that gets you up to 4400 -- none of them are in the retail channels yet.


It seems that:

• It is harder to make a stick for a 4-stick-kit for the ___same_speed_and_total_capacity___ or ___same_speed_and_stick_capacity___
  
• The sticks in the 4-stick kit are inherently different (better?) (certainly less available) than the 2-stick kit for the same capacity
  

[pianobench]

Hi,

I have a couple of questions.

1. Where do you find the values of 1.1mOhm for load line calibration lvl 7? Just wondering.

2. Did you notice any differences in vcore reporting? I am very interested in the Maximus Hero XII with die-sense. So I am wondering which board gave you accurate vcore readings.

I have a Z490-F board right now and I wish I spent the extra 30 or 40 for the Z490-E. At the time I figured save $$$ on the motherboard because I didn't need the wifi from Z490-E. And I thought Asus XII Formula cost way way too much at $499 !!!

I came from an i7 930 with Asus Rampage III Formula. And I remember at the time the board was 200 to 300. So I stuck to that price range.

Thank you,

[Falkentyne]

pianobench wrote:
Hi,

I have a couple of questions.

1. Where do you find the values of 1.1mOhm for load line calibration lvl 7? Just wondering.

2. Did you notice any differences in vcore reporting? I am very interested in the Maximus Hero XII with die-sense. So I am wondering which board gave you accurate vcore readings.

I have a Z490-F board right now and I wish I spent the extra 30 or 40 for the Z490-E. At the time I figured save $$$ on the motherboard because I didn't need the wifi from Z490-E. And I thought Asus XII Formula cost way way too much at $499 !!!

I came from an i7 930 with Asus Rampage III Formula. And I remember at the time the board was 200 to 300. So I stuck to that price range.

Thank you,

LLC7 is 0.165 mOhm not 1.1 mOhm.

1.1 mOhm is Intel spec, which is basically "yeet vdroop". On Maximus 12 Extreme and Apex, that is LLC3. I heard this is LLC1 on the Hero (for some weird reason, LLC1 on the Apex/Extreme is 1.7 mOhm, probably because that is the stock mOhm for 6 core 10th gen processors) but I heard that second hand and i have not seen anyone test it (it's easy as long as you look in "Asus EC" in Hwinfo64 sensors, and the bios vcore set is given by the user, you can calculate the mOhm values perfectly, but ONLY on the "Maximus" series boards, and on Gigabyte boards that have "VR VOUT". That's currently NOT possible on the Strix because the Strix does not have die-sense vcore readouts (and probably not an Amps reading on Asus EC, either).

If Hero's LLC1 is 1.1 mOhm, then most likely Strix LLC1 is 1.1 mOhm also, but again I do not know for sure.