Need help with getting DDR4 Samsung B-die stable

[Darktalon]

I'm going crazy trying to figure this stuff out. I've got a Samsung b-die kit, g.skill, date is 2020 Jul, I just bought them last week off Newegg. F4-3200c14d-32gvk to be specific. It has xmp set for 3200 c14-14-14-34, it is 2x16. I have a 6700k, and an Asrock Z170 Gaming K6. It absolutely will not boot using xmp, so I had to increase the vccio/sa to 1.25 and 1.3.

Then I tried to see what my absolute limits would be, I put vram to 1.5v, io to 1.25 which is the max I can pick, and sa 1.35, also the max I can pick. No matter how loose the timings, it absolutely would not post beyond 3700.

So I scale it back to 3600, and I get it to take 16-16-16-38, and into windows. Feels pretty stable, I would expect that since I'm dumping a stupid amount of voltage in this thing. Well then I try hci memtest pro, and of course, 1 error pops up after 2 hours. I check the dimms reportred temps, they are like 44c.

Well so I scale back the voltage to dram 1.4 and sa 1.3, and see if that helps any in case I was unstable due to temp, plus I make my fans on cpu and case run faster. Well that took a few C off and running the test to 100%, yay no errors, though I'm sure it would have if I ran overnight. I move to try out prime95, and ran large without making any changes, within 5 minutes it's failed on multiple threads. I also saw elsewhere to try doing 112k no avx to test the memory controller, I do that, INSTANT failure. So I pop the sa voltage back to max and it still instant fails.

At this point I'm really frustrated and dissapointed because it looks like I can't even get 3600cl16 stable, which just feels like a kick in the **** when I have a 3200cl14 kit.

I'm not sure what to do, I'm really lost on if I can do anything with the secondary and tertiary timings, I've left these on auto because they make my head hurt to have 50 settings that aren't explained anywhere. I've tried to read so many things and I just can't find a good video or guide that explains those timings, and idk if it would even help if I'm already unstable using the I assume much looser auto settings.

Am I just absolutely screwed because my memory controller is apparently a dumpster fire?

 

[bigtallanddopey]

Just had a look and there is a note on the motherboard manufacturing site that basically states that 3833 can only be achieved with a single module. The value below that, that is explicitly mentioned is 3600. So anything above 3600 is not guaranteed on that board unless you use a single ram stick.

So it could be the case that only upto 3600 is supported by your motherboard and you will struggle to get much more unless you really really mess about with all the timings. But that can be a little dangerous and corrupt things.

As for not getting 3600, I suppose that technically isn’t guaranteed by the ram. You have bought a 3200 kit. Maybe the ram just isn’t capable. Or maybe it’s a combination of the ram and the motherboard and the cpu just not letting you get what you want.

 

[Darktalon]

I would be accepting of 3600 as a max speed, if I could at least get timings to bring my latency in line with 3200 CL14, but since I already maxed out voltage and it wasn't stable enough at 3600CL16, is there anything else I can try, or am I better off to just go back to 3200 CL14?

 

[Woomack]

On Z170 memory clock was usually not working well past DDR4-3600 and memory capacity was affecting that more than on new series. I would keep it at 3200 CL14 as on these older series there is barely any improvement because of memory clock while it's much easier to stabilize it at 3200 than anything higher.
If you had some top OC motherboard then it would be possible to set up to 4000 but there weren't many motherboards that could do that with 16GB modules.

 

[Darktalon]

Is there a guide or cheat sheet for a way to tighten up my secondaries and tertiaries at 3200 cl14? I will feel dumb if I can't go beyond XMP in any way.

 

[Woomack]

There is no guide that fits every motherboard. Some timings depend on the motherboard or used chipset. This is also why in SPD or XMP are only some timings and everything else is matched by the motherboard.
Some timings are not affecting performance or not significantly.
For sure you can set lower tRFC. Something about 280-340 should work. At auto, it's probably around 540. It helps a bit with latency. TFAW can probably go down to 18-24.

The easiest way to check how low it can go is to run memory at SPD settings to check timings, and later try to drop one by one to the values from SPD but memory clock at 3200.
In the other way:
1. Run memory at auto
2. Check what timings are in BIOS, not sure if your motherboard is showing them as greyed-out or something like that, if not then can use something like ASRock timing configurator - can download it here https://www.asrock.com/mb/Intel/Z170 OC Formula/#Download
3. Use timings from SPD in your manual settings, some may fail so add them by 1-2 at once and run a quick stability test, something like 7Zip benchmark with close to maximum memory capacity is good and quick (~3 passes are enough)

 

[Darktalon]

There is no guide that fits every motherboard. Some timings depend on the motherboard or used chipset. This is also why in SPD or XMP are only some timings and everything else is matched by the motherboard.
Some timings are not affecting performance or not significantly.
For sure you can set lower tRFC. Something about 280-340 should work. At auto, it's probably around 540. It helps a bit with latency. TFAW can probably go down to 18-24.

The easiest way to check how low it can go is to run memory at SPD settings to check timings, and later try to drop one by one to the values from SPD but memory clock at 3200.
In the other way:
1. Run memory at auto
2. Check what timings are in BIOS, not sure if your motherboard is showing them as greyed-out or something like that, if not then can use something like ASRock timing configurator - can download it here https://www.asrock.com/mb/Intel/Z170 OC Formula/#Download
3. Use timings from SPD in your manual settings, some may fail so add them by 1-2 at once and run a quick stability test, something like 7Zip benchmark with close to maximum memory capacity is good and quick (~3 passes are enough)

How does this look so far? It has been stable so far in everything I toss at it. (Karhu Ram Test, MemTestPro, and Prime95)

My XMP had latency at 51.7 and copy at 41737, so at least in AIDA64, I've gotten improvements.

https://imgur.com/a/dzeBjxj

 

[Woomack]

Looks good. You can set Command Rate to 1N/T but there is not much more that will bring a significant improvement, even in synthetic benchmarks.