F4-4000C16D-16GVK & i7-9700K on Gigabyte Aorus Z390 Master - unstable

[DarkErazor]

Hello,

this week I installed above mentioned two sticks in slots 2 and 4 of the Gigabyte Aorus Z390 Master.
Never had any stability issues before, with everything on auto in the bios, which is the latest version F11n.
Now I set most of the advised values manually from the gigabyte overclocking guide for these boards, but clocks are on stock levels.
XMP is enabled. Memory enhancement to normal.
Voltages are 1.25 for VCCIO and VCCSA. LLC is set to Turbo. DRAM voltage 1.4V.

See attached screenshots for the memory timings and readout of the RAM from Thaiphoon.
I increased timings 2-4 due to the instructions in this guide:https://github.com/integralfx/MemTes...d#overclocking

Even at these "stock" settings from XMP, the system is unstable in memory tests like OCCT memory SSE after a few minutes. Immediately errors from Memtest5, which I dont want to use because of missing documentation for the custom configs. Prime errors too after a short time.

What advice can you give me, when going for stable settings at stock clocks?
When that is established, how would you advise to move forward, tightening the RAM timings?
And what I dont understand, why isn't there a complete liste of all the (sub-)timings for the different RAM sticks available from the manufacturer?

If you need more information, please tell me.

Thank you in advance

[GrimReaper]

I'm so glad I found you. Especially it made me smile once I saw "today" as date posted. I only found this topic on 4th Google page, almost gave up finding anything about this kit.
Let me explain. Today I was testing my F4-3600C14D-16GTZNB 3600 14-15-15-35 1.45V kit, which is supposed to be the top B-Die bin.
But with B version of 16GTZN they increased the voltage from 1.40V to 1.45V. And only after Woomack from overclockers.com said it is best you can get, I decided buying it.

So far I have mixed feelings about it. It's definitely good, but me going after 4000 14-14-14 I would say it's average as it doesn't do that. Separate sticks probably do, so maybe it's also down to my mobo. I don't think it's the IMC as I tried going higher with VCCSA. But I'm starting to believe when people say 4000 CL14, they actually mean 4000 14-15-15 which isn't a problem and works with same voltage as 4000 15-15-15.
Then it also does 3600 14-15-15 with only 1.32V. This makes me believe I actually got the same bin as the discontinued F4-3600C14D-16GTZN. Memory kits from factory usually have 0.05-0.08V headroom and mine has a whole 0.13V.

Now I'm getting back to your topic. I also tried all other XMPs of all the best G.Skill B-Die kits. And these are the hardest to run:
3600 15-15-15 - mine need 1.32V, G.Skill made 1.35V F4-3600C15D-16GTZ. But this one is long discontinued and is known to be the highest bin G.Skill made, so no problem here.
4266 16-19-19 - mine need 1.45V, G.Skill made 1.45V kits
4000 16-19-19 - mine need 1.37V, G.Skill makes 1.40V, for example Ripjaws V F4-4000C16D-16GVK which is the one you have.

And this F4-4000C16D-16GVK got the most attention in me because it's very cheap, only 160€, end of December it was even 136€.
And so I was wondering, how can this cheap kit be better than my of one the best 2x8 GB kit. It was too good to be true. And I already thought about buying it in a month or so.

But now I see it's not even stable at 1.40V. That would actually make sense as for this money you shouldn't get a better bin than my 230€ kit.
If I were you I would just first try 1.45V, if stable going down in 0.1V or 0.2V steps, if not stable, going up in 0.1V or 0.2V steps.

As of what is going on with this kit, I hope G.Skill representative can answer to both you and me. I already read about some kits not doing XMP, but that's usually down to older not that good motherboard or something similar.

As I already said, all kits should have a factory headroom. My F4-3600C16D-16GVK 1.35V work 1.27V and F4-3200C14D-16GVR 1.35V work 1.30V.
If you think about it, it can't be any other way. No factory can afford to bin kits to just 0.01V. And they sure don't want anyone to experience what you did, that's bad for company rep.
Let's say your kit would be the same bin as mine. That would mean best ones would need 1.35V and worst ones would need 1.40V or even more. And this happened to you.
There are 2 possibilites here:
1. It really is the best bin, even better than 3600 14-15-15-35, 3800 14-16-16-36, 4000 15-16-16-36, 4000 17-17-17-37 and 4133 17-17-17-37 and only a few out of 10.000 like yours don't do XMP voltage. But these are all top bins and sell for 200-300€, so I would be very surprised.
2. It's an average bin and the XMP voltage they choose is too low. Would be very surprised G.Skill made such a mistake.

I was also wondering what's going on with that 19-19. I mean, B-Die is probably B-Die and all other kits work no problem 16-16-16 with only 0.01-0.02V more. I get it for 3600 16-19-19 which is a Hynix kit.

So if you can, please test:
1. Higher/lowest stable voltage on XMP settings 4000 16-19-19-39, at least half hour of Prime95 Large, but better is 1 hour as a lot of errors/BSODs can happen at 30-45min mark
2. 4000 16-16-16 lowest stable voltage
3. 4000 15-15-15 lowest stable voltage

You don't need 16-20-20-40, I'm quite sure it's just down to voltage.

That should be enough info for me to not buy this kit. As now I'm somehow intrigued by it, especially by that low voltage.
If you can do that for me, I can help you with subtimings, I know everything from that Github guide.

VCCIO you can lower to 1.20V, that should be more than enough for 4000.

And what I dont understand, why isn't there a complete liste of all the (sub-)timings for the different RAM sticks available from the manufacturer?

This depends on motherboard and every particular serial number of a stick. Factories don't have time to test each stick down to every subtiming. That's what overclocking is for. One person with one kit has a lot more time to find the max stable frequency and lowest timings.
Sure, XMP could have at least a few subtimings more or at the least tRFC, and more relaxed ones which every kit should do.
My MSI Z490 Tomahawk doesn't even take tRFC (I found this one in my mobo BIOS under XMP info and in AIDA64, only 631 though, board sets 630 anyway on Auto and on XMP), tFAW, tRRDS and tRRDL from XMP, only the 4 primary timings and voltage. All subtimings are the same with XMP and with Auto in ASRock Timing Configurator.
So XMP is actually worthless for me, I only tried it to check if it sets some subtimings. I can type in the timings from the box and voltage is higher than necessary anyway.

[DarkErazor]

Hey GrimReaper,

Thanks for your informative reply.
After wasting the sunday, trying to find stable settings with "Memory Enhancement" in the bios on "Auto" or "Normal", I switched to "Relax OC".
You can see the differences it makes in regards to the timings in the screenshots attached.
This "Relax OC" is 1 hour stable in TestMem5 @ 1.45V VDIM (not Vcore) and 1.25V VCCIO/VCCSA, without a single error. Finally!
I just finished 10 minutes in TestMem5 @ 1.4V Vcore and 1.2V VCCIO/VCCSA without any errors.

The normal memory enhancement gave me very quick errors, mostly in the first 2 minutes, test 4 or 6 of the extreme config.
I will keep testing what you asked for in regards to lowest stable voltages and get back to you again.

[GrimReaper]

Your subtimings are quite relaxed now. I checked which were very low before and I'm almost sure these 2 were the problem:
tRDRD_SG - 7
tWRWR_SG - 7

As I was setting subtimings a lot and these 2 were one of the few I first had to change to get them to lowest stable.
I think you should be good with 8/8. That's also the safe value per that guide.

Lowest I can get on 4000 CL16 (actually they're the same from 3200 to 4266) is:
tRDRD_SG - 7
tWRWR_SG - 8

But Auto on my MSI Z490 Tomahawk is not a problem, it sets them relaxed enough. I haven't tried anything else except everything Manual with the help of that Github guide.

By Vcore 1.45V you probably mean VDIMM, because I wouldn't suggest running that high vcore.

EDIT:
Actually I also have 7/7 now that I just have primaries set. Needing 7/8 is probably tied to some other subtiming.
In any case I would try lower let's say 10 at a time to previous values so you can find which ones are the problem. Then when you get to those 10, you can find the exact one(s) that are too low. Because now you're really high.

But I would still start with these on the right of ASRock timing config as only these are different now. tWTR_S was quite high even before.
You can leave DR and DD on Auto.
DR - dual rank sticks, for example 2x16GB
DD - dual dimms per channel, for example 4x8GB
You don't have either of these so no need to set them.

tWRRD_DG I think it was also fine before, usually it needs to be 6 lower than SG.

When you have XMP timings stable with subtimings closer to what were before, then you can try to lower VDIMM. It should work 1.35V.

[DarkErazor]

You were right. I meant VDIMM. Corrected above.
So my findings are as follows:
1. Higher/lowest stable voltage on XMP settings 4000 16-19-19-39, at least half hour of Prime95 Large, but better is 1 hour as a lot of errors/BSODs can happen at 30-45min mark
=> 1.38 VDIMM
2. 4000 16-16-16 lowest stable voltage
=> 1.38 VDIMM
3. 4000 15-15-15 lowest stable voltage
=> 1.46 VDIMM
VCCIO/VCCSA both @ 1.2V
TestMem5 more than 30 minutes without errors.
I will try your suggestions next.

[GrimReaper]

That is the same as my F4-3600C14D-16GTZNB. Very good for 50€ less.
I expected it can do 16-16-16 with same voltage, it's B-Die after all. Don't really know why they used the same timings as in 3600 16-19-19 with Hynix CJR/DJR.
There is not a single G.Skill XMP which won't benefit with manual tuning, but tRDC and tRP +3 from what it is capable is the stupidest of all.
And voltage they used really is on the borderline to 1.40V. So I wasn't really wrong suggesting you need higher voltage.
Their XMPs are really strange, my 3600 14-15-15 1.45V works with 1.32V, so they could also market it as 1.35V. But more common sense would be 1.45V for yours and 1.40V for mine. Usual headroom is 0.05-0.08V and not 0.02 or 0.13V.

You can also test:
3900 14-14-14 - mine need 1.48V
4000 14-15-15 requires same voltage as 4000 15-15-15, so 1.46V. I think this should be the best setting for you.
4266 16-16-16 should also work on same voltage as 4000 15-15-15 - 1.46V. But you need higher VCCSA, I need 1.30V on 10600K.
4200 15-15-15 should be max you can get out of this kit and your CPU, mine need 1.54V.

Also can you make a Thaiphoon Burner screenshot?

Oh, to be 24/7 stable you need to add +0.1V to that TM5 30 minutes stability, better yet +0.2V.

[DarkErazor]

Thaiphoon attached.
More than 1h stable @ 1.47V VDIMM.
Will continue to tighten, but not increase the frequency further.

[DarkErazor]

Well it just did not work out @DDR 4000:
Whenever I tried to lower the essential timings of tRDRD_dg and tWRWR_DG to 4, TestMem 5 produced errors.
I might be able to try higher voltages as VCCIO&VCCSA were @ 1.22V max, but for the time being, I found a setting @ DDR 3600 which is stable after 3 runs in TestMem 5.
See attached:





This is with XMP on and "Memory Enhancement" on normal setting.
I will try to max out this as for now.
If anyone has a clue, what I can do, to run these RAM sticks at their specified speed DDR4000 with low subtimings, let me know. It would be appreciated.
For now, Pass Mark says, I am faster than the regular DDR3600 CL14 RAM from G.Skill, so that should be fine for me.

[GrimReaper]

You don't need tRDRD_dg and tWRWR_DG at 4, lowest that is still stable. 4000 will be better in any case than 3600, probably even with those 16 subtimings.
I never bother with subtimings until I'm satisfied with my speed, primary timings and voltage. Subtimings are then not so hard, some just need +0.03V or so to run.

[DarkErazor]

Hello,

Even though I have the RAM sticks running @ DDR3600 Cl14, it still bothers me, they are not running at specified speeds.
While looking into the issue again, I found this post on reddit:
https://www.reddit.com/r/overclockin...eb2x&context=3
So the mainboard layout seems to be the issue above DDR3600 when using only two sticks.
I plan to order another pair of the same sticks and try again. After all, RAM is supposed to increase in price in the future anyways.
I will post here again with my findings.

[DarkErazor]

Okay, so I ordered another 2 sticks of the same type and tried again. This time, it was much more pleasant to adjust the RAM.
From 3600 CL14 I moved to 4000 CL16 with tightened timings. I could not get the RAM to stabilize above 4133 MHz, even though VCCSA tried was up to 1.32V and VCCIO to 1.25 V with VDDR @ 1.5V and timings 18-20-20-40 according to the overclocking guide mentioned in an earlier post.
That is why I am staying @ DDR4000 with the following timings while setting memory enhancement in the gigabyte bios at "enhanced performance".
These result in 4600 points in the memory test from Passmark's Performance Test.
Games are smoother now with higher consistent frames, so I will stick with this, until it is time for a bigger upgrade.