G.Skill 3200 C15 not running well on C6H

[Masterchief79]

Hey guys,

not really satisfied with my memory performance so far so I'm asking you for help. It's 2x8GB of this one, I bought it because TridentZ were supposed to run great on Ryzen as they are singlesided and single rank (at least the 3200C14 are, so I figured mine are probably too, right?)

 

So, here are my problems with it:

• It doesn't run 3200 CL15, maxed out at about 3100MHz - more and I get "0d" when booting.
  
• It doesn't really seem to run stable, at 2933MHz CL13 it seems to negatively impact my Firestrike Combined Score
  
• I can't adjust subtimings in the BIOS of my C6H. When I set TRC to 50 for example (which is default by SPD) it still stays at 69, according to CPU-Z and MemTweakIt. I also can't adjust the fan speed in windows and a few other things. Maybe tools not working so well yet with 0902?
  
• Somehow the Command Rate is at 2T?! I thought this was a 1T only platform.
  

 

Things I have tried:

• newest BIOS version 0902.
  
• SOC voltage of 1,2V
  
• memory voltage of 1,35V and more
  
• booting with 3200C16 (still "0d")
  
• swapping modules and slots around
  

 

Hopefully you can help me eliminate these annoying things about this system.

Edited March 15, 2017 by Masterchief79

[Alex@ro]

3200c15 can be hynix or even micron, or even double'sided &e-die, a pic with serial number or naked ic will help. Higher straps on bios 0902 are 2t,this is changed from previous bioses in order to help clocks and stability(according to asus). However since 0902 is the bios that fixes the EC corruption i would recommned you stay on this one for now.

 

Sent from my SM-N910F using Tapatalk

[I.nfraR.ed]

For me 0902 booted 3200C12 from first time, while 8305 was going through F9 sycles before finally booting up.

Try Taiphoon burner, IIRC the free version can read SPD. I'd say it is very likely your memory is not B-die.

Are the primaries flat in the specs or something like 15-18-18?

Also you can't adjust subtimings manually with current AGESA code, you can only change primaries in bios, all others are controlled by AMD and the strap you choose.

elmor said the subtimings are just placeholders at the moment in case AMD unlocks them again in the future.

Edited March 15, 2017 by I.nfraR.ed

[chew*]

The latest microcode that I have hates odd CL. try running even CL.

 

I would physically look at your sticks to confirm single sided. 2T is what my Hynix was defaulting at on my double sided sticks.

 

Double sided is almost the equivalent of running 4 single sided which also default at 2T.

 

Not sure why everyone is obsessed with running SOC @ 1.2 I run all my chips at 1.0 just fine. less volts in less heat output.

Edited March 15, 2017 by chew*

[Masterchief79]

Alright, thanks for advice. My sticks are singlesided, I pulled them out and had a look. Also thanks Infrared, that's some useful information.

It turns out my CPU had this memory "hole" a little earlier than expected (this is described in the Asus C6H OC Guide). For me, 3100-3400MHz are a no-go, but 3500MHz booted flawlessly. *facepalm*

 

I could even get some good latencies out of those. I ran SuperPi 32M at 3520MHz CL13-13-13-32@1,425V 1T if I remember correctly.

 

Another thing I noticed is that with anything above 105MHz BCLK my M2 SSD stops working (960 Evo). So for testing the memory I had to install a fresh Win10 on a SATA SSD. Also, now that I have this sorted out, the M2 performance is in the bin, so the issues don't quite stop for me there ;(

This btw also means that I can't run my memory on its specification because for 24/7 I can't go higher than 105MHz BCLK (because PCI-E 3.0 doesn't work above that) and I can't clock it between 3100 and 3400MHz. So I'll just have to stick with 2933MHz CL13. Ah well.

[I.nfraR.ed]

chew* is right, SOC 1.2V doesn't help with anything, leave it at 1.0V. Even on LN2 I didn't see a difference, but haven't played much with it to be 100% sure.

So your mem is B-die then, good.

[chew*]

Alright, thanks for advice. My sticks are singlesided, I pulled them out and had a look. Also thanks Infrared, that's some useful information.

It turns out my CPU had this memory "hole" a little earlier than expected (this is described in the Asus C6H OC Guide). For me, 3100-3400MHz are a no-go, but 3500MHz booted flawlessly. *facepalm*

 

I could even get some good latencies out of those. I ran SuperPi 32M at 3520MHz CL13-13-13-32@1,425V 1T if I remember correctly.

 

Another thing I noticed is that with anything above 105MHz BCLK my M2 SSD stops working (960 Evo). So for testing the memory I had to install a fresh Win10 on a SATA SSD. Also, now that I have this sorted out, the M2 performance is in the bin, so the issues don't quite stop for me there ;(

This btw also means that I can't run my memory on its specification because for 24/7 I can't go higher than 105MHz BCLK (because PCI-E 3.0 doesn't work above that) and I can't clock it between 3100 and 3400MHz. So I'll just have to stick with 2933MHz CL13. Ah well.

 

Be advised bclk is strapped to PCI. Many devices will not like high PCI, some will corrupt some will break. Including onboard devices hard drives vga.......

 

Best to disable all when "benching at high PCI"

 

SOC may help stabilize IMC with 4 dims, no clue ( don't have 4 dims ) but with 2x8g single sided Samsung B die 1.0 is fine.

Edited March 15, 2017 by chew*

[I.nfraR.ed]

I have 4x8GB B-die and will get a new board today, so I'm going to test 4x8 if my 1800X is still alive.

[Masterchief79]

Be advised bclk is strapped to PCI. Many devices will not like high PCI, some will corrupt some will break. Including onboard devices hard drives vga.......

 

Best to disable all when "benching at high PCI"

 

SOC may help stabilize IMC with 4 dims, no clue ( don't have 4 dims ) but with 2x8g single sided Samsung B die 1.0 is fine.

 

Thanks again.

I know the Refclkis strapped to all the internal devices but the Asus C6H and the Gigabyte X370 K7 IIRC use a second, external chip to take over clocking those devices. So as soon as you clock the refclk higher than 105MHz, for example the PCI-E clock, will get decoupled from the main bus clock and clocked via another chip. I read this somewhere, tried to find the source but no luck. Will look into it further when I have more time. It would make sense though.

Edit: It was this news. It's in German unfortunately but it basically confirms what I said. If somebody's interested, just try google translate I guess.

 

Regarding the SOC voltage: I know this was unnecessary but 1,2V was deemed safe by a lot of sources for benching and I just didn't want to take any chances. It's still useful information, for 24/7 I will wind this down to 1,0V or whatever turns out to be stable.

 

Also, AMD announced new ram ratios in may, don't know if this was publicly known already. Was on their blog.

Edited March 15, 2017 by Masterchief79

[chew*]

All ref clock boards are using an ASMEDIA chip. I could be wrong but this was for fine grain control. The cpu is actually capable without the chip.

 

1.2v was deemed safe in reviewers guide yes.

 

Why anyone would just jack voltages up without verifying that they are actually helping anything is beyond me however.

 

First thing I did was lower all voltages looking for potential overlooked gains.

 

Then I retested raising.

Edited March 15, 2017 by chew*

[seanay00]

Have a look at this if you get stuck.

 

https://community.amd.com/community/gaming/blog/2017/03/13/tips-for-building-a-better-amd-ryzen-system