Overclock.net › Forums › AMD › AMD Motherboards › Asrock X370 Taichi Overclocking Thread
New Posts  All Forums:Forum Nav:

Asrock X370 Taichi Overclocking Thread - Page 211

post #2101 of 2541
With my overclock and stable ram to run 3466 on my rig at 14,13,13,13,28 and other lowered timings, I’m running 1.41 Volts. I’ve literally been testing sub timings for two weeks, basically Prime95 has been at 100% for two weeks with the occasionally a crash when I tighten timings to tight. Are you running Samsung b-die? If so 1.45 is supposedly safe, I’d stay under 1.42 though. My case is very cool, mem is cool to touch. If your running Samsung b-die the above is correct but if not disregard. I’m not sure what max voltage is for Hynix Ram


Quote:
Originally Posted by Korrektor View Post

Eh, I think that I will stay with 3066, at least for now till I'll be in mood to continue. I feel like I'm spending a lots of time to nothing or few percents of performance
Even if I get stable in windows there is occasionally random boot problems (like that b2 code or 3 reboot loops every couple of seconds with resetting the bios to default, etc.)
For instance today I've passed two hours of full AIDA testing (including CPU, FPU, memory) but when I rebooted few times and tried to cold boot I failed to boot or got into reboot loop once and this fast three reboots sequences looks and sounds not pleasant at all (Like your PC is dying or something)

What I've applied in total during this days when doing research at this problem (I mentioned this previously a plenty of times in different posts but just as overall conclusion):

- Reduced the SoC to 1.05 and VDDP to 0.85-0.9 (this values seems to be the optimal according to few sources including this forum)
- Increased the RAM voltage (up to 1.36-1.38, not that much, but however sensor readings show generally higher values up to 1.4. I know that they can be wrong (e.g. that software sensors suck). I just afraid keeping ram at high voltages, but maybe I'm wrong and should push it to 1.4?)
- Changed LLC values a few times (they seem to cause no changes)
- Applied 60-68.8 ohms ProcODT value instead of going auto
- Changed BankGroupSwap / BankGroupSwapAlt values (e.g. left them on auto, then fully disabled, then swapped one to another)

Values that I haven't touched yet but thought of:
- Timings. As I understand choosing more loose timings could help bring stability but It feels like a trade-off and as I'll get bigger latencies
- T2 command rate. Logic behind this is basically the same - as I've seen in different sources it will in fact descrease performance and I'm felling that another 100mhz simply not worth it.
- CLDO VDDP. I found little info about that but as I understand in is somehow involved in memory OC. It was on auto all the time

Some good info I found that can help myself in the future (and probably someone else who will read this thread):
- If your windows is installed in UEFI you can disable CSM mode in bios (some legacy way to boot) and this will greatly cut loading times (I wish I knew this a week ealier, all of this long-lasting reboots every few minutes to change something was painful to watch. I also had the Plextor pci-e drive loading stage that added like 5+ seconds to boot times so despite my system is installed on very fast 960 evo I had to wait for about a minute in total every time

- Disabling AM4 Advance boot Training
Now this is questionable part and probably a game changer, because as I understand this is the main reason of the default double-boot behaviour and triple-rebooting in case of failed memory settings (and probably that b2 error as well)
Disabling this reduced the boot time even more. I wonder if I could just disable this when I got stable in windows and be ok? Or as it failed to calibrate on next cold boot it just meant that my testing was not representative and I could end up with bsods and errors once





work can be achieved without it
Sad that I haven't paid much attention to this as with both of this options enabled I got much bigger boot times and wasn't able to try out more stuff

Edited by sierra248 - 10/18/17 at 5:07pm
post #2102 of 2541
Yep its Samsung b-die. It seems that all of my problems appears because this is dual-rank modules. I've seen a lot of people achieving 3333 or 3466 on similar single-rank memory. The reason I purchased DR is that I've seen some info that they perform better but for now I see that this was half-true because there is some hidden pitfalls and I should just go with SR probably

There was one guy with the same kit few posts later and he said that he was OK with much less changes. I feel like I'm getting something terribly wrong because I did much more and still failed
post #2103 of 2541
Quote:
Originally Posted by Amir007 View Post

Alright I will finally try this soon and let you know. I think I get it now. Thank you for over explaining it to me. I only have CPU fan 1 populated by the water cooler but that records the pump speed as my fan is plugged via Sata power cable NZXT comes with. I can control my radiator fan via CAM software...its pretty nice actually it has all the tools you need. So far my rig is very stable and not crashing anymore but again I'm not sure if its because I removed the AMD power chipset drivers that gives you Ryzen Power plan.

I'll report back. Thanks again sierra.

Cool, I run the newest chipset drivers from dthe AMD site that just came out beginning October I believe. I do run the balanced ryzen plan but I’ve modified it to everything maximum performance, never sleep, never hibernate and such while I’m testing. I think I’ve changed it to same as maximum performance settings, it’s just now called ryzen balanced:





post #2104 of 2541
Reset everything in bios, do your overclock, voltage and set LLC (I use level 1)and geardown enabled. and set DRAM to 1.39. Reboot, then set xmp to 3200.
Worth a shot


quote name="Korrektor" url="/t/1627407/asrock-x370-taichi-overclocking-thread/2100#post_26400572"]Yep its Samsung b-die. It seems that all of my problems appears because this is dual-rank modules. I've seen a lot of people achieving 3333 or 3466 on similar single-rank memory. The reason I purchased DR is that I've seen some info that they perform better but for now I see that this was half-true because there is some hidden pitfalls and I should just go with SR probably

There was one guy with the same kit few posts later and he said that he was OK with much less changes. I feel like I'm getting something terribly wrong because I did much more and still failed[/quote]
post #2105 of 2541
You'll get a memory training fail at 3200 or higher so you set memory training fail count to at least 3, possibly 4 so no fast boot
For Dual Rank Modules BankGroupSwap Enabled, BankGroupSwapAlt Disabled is used
Set DRAM Voltage to 1.4
ProcODT for Dual Rank modules should be higher than normal so go up to like 80 and 96.6 Ohms
Since they're DR B-dies, set CR 1T and Geardown Mode Enabled

Try booting into a lower strap first with these settings like 3066 then try increasing the strap to 3200

I'm using SoC LLC3 and CPU LLC2, they aren't super aggressive tbh

CLDO_VDDP might help if you have a memclkhole at 3200 so when all else fails, this should be the setting you tweak pretty much.

As I understand it, it's about finding correct values for CLDO_VDDP and ProcODT when OCing DR modules on Ryzen
Edited by Brightmist - 10/18/17 at 5:49pm
post #2106 of 2541
Guys, keep in mind when suggesting or changing your voltages that The Load Line Calibration will increase voltage to both the ram and Vcore. This may prevent a machine from stable operation or force someone into higher voltages than preferable.

For example, DRAM set to 3.70 (on my board) will increase to 3.84 with an LLC Stage 1 setting. I've also noticed at times a non-linear progression in voltage but haven't tried to investigate.

Nothing but problems since Bios 3.10 and would strongly advise against copying settings, especially in regards to voltage and LLC. With that being said, both my machines have always ran their best with lower than typical voltage, and additionally, the less you poke at it with the proverbial stick, the easier it is to reconfigure when the S hits the fan.

Avoid changing LLC as it's one of the better settings to set and forget until you need it. Level 5 provides more accurate data sharing in regards to physically set integers within bios. Or atleast that's my thought around it.

Taichi seems to prefer subtle changes rather than several crashes in a row. Find what works and slowly adjust until you're somewhat less unsatisfied.





LLC may also increase SOC voltage with higher settings.
post #2107 of 2541
Quote:
Originally Posted by sierra248 View Post

Cool, I run the newest chipset drivers from dthe AMD site that just came out beginning October I believe. I do run the balanced ryzen plan but I’ve modified it to everything maximum performance, never sleep, never hibernate and such while I’m testing. I think I’ve changed it to same as maximum performance settings, it’s just now called ryzen balanced:






I tried the OC tips you gave me and my PC froze in windows login. I tried 1.4v @4Ghz. I'll try again tomorrow. I cleared the CMOS and loaded defaults and then set my ram to 3200mhz and everything else stock. I'am still experiencing a bug of some sort when I run Cinebench where my score is 2x less than what it should be. I noticed my CPU speed goes down to 550Mhz during 100% cpu load while running the test. Why is this? I think this is the issue. I set my power options to High performance in windows so why is my speed dropping so low while testing? ***?




The scores in red and orange are my scores per each try. I feel like I got a lemon AMD 1800x.
post #2108 of 2541
Thanks for the recent advices. I guess I'll still just stay at 3066 at least for few weeks, I really want to work and play on my new PC in the first place, last week I was too focused on all of this values and spent lot of time just watching the voltages, temps, etc, instead of having fun with new hardware. I will write later then

Speaking of voltages LLC I agree with gimpinator post above
I guess that high LLCs levels are excessive and you should carefully measure all of your main voltages both on idle and loaded states
I noticed that my DRAM voltages is higher that I've manually put in bios.
Edited by Korrektor - 10/19/17 at 12:11am
post #2109 of 2541
Quote:
Originally Posted by gimpinator View Post

Guys, keep in mind when suggesting or changing your voltages that The Load Line Calibration will increase voltage to both the ram and Vcore. This may prevent a machine from stable operation or force someone into higher voltages than preferable.

For example, DRAM set to 3.70 (on my board) will increase to 3.84 with an LLC Stage 1 setting. I've also noticed at times a non-linear progression in voltage but haven't tried to investigate.

Nothing but problems since Bios 3.10 and would strongly advise against copying settings, especially in regards to voltage and LLC. With that being said, both my machines have always ran their best with lower than typical voltage, and additionally, the less you poke at it with the proverbial stick, the easier it is to reconfigure when the S hits the fan.

Avoid changing LLC as it's one of the better settings to set and forget until you need it. Level 5 provides more accurate data sharing in regards to physically set integers within bios. Or atleast that's my thought around it.

Taichi seems to prefer subtle changes rather than several crashes in a row. Find what works and slowly adjust until you're somewhat less unsatisfied.





LLC may also increase SOC voltage with higher settings.

LLC isn't supposed to be increasing DRAM voltage. It's literally called VCORE LLC in BIOS.

RAM also doesn't work with 3.70/3.84V, I'm guessing you misstyped and it's 1.370/1.384. In any case, that's perfectly within sensor fault tolerances so your RAM voltage might just be the same.

LLC is there to counter vdroop but again, watching voltages under load made me decide to increase it to LLC2 from LLC3 since a high enough vdroop will probably destabilize my IMC under load since I'm running my RAM @ 3466C14 timings (1T/GDM disabled).

LLC3 is also fine for both SoC and Vcore and you won't end up with higher voltage than you set under BIOS when under full load or atleast that's what sensor readings are showing.
post #2110 of 2541
amir007

have you tried just using the pstate method

its worked flawlessly for me since bios 1.5 onwards

i run a 1700 at 3.9ghz and 1.325 volts cpu LLC level2

my cinebench is version 15 but it scores 1750+ (which is quite a respectable score) and will do that consistantly

heres is a guide on setting the pstate overclock
(second half of video)

https://www.youtube.com/watch?v=52Tw-wcT7o4

settings for 3.9 ghz and 1.325 volts are
9c
8
24
second box
9c
8
24
third box
auto
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: AMD Motherboards
Overclock.net › Forums › AMD › AMD Motherboards › Asrock X370 Taichi Overclocking Thread