Asrock X370 Taichi Overclocking Thread

When I find the time I will do it for you...do you have a ryzen 3000 or a 1000/2000? Also which bios do you want...please link.

I need to learn and get better at it ~ can't only count on you

I use the 1700X so far
Gladly CBS is unlocked, but i miss PHY Memory Training and i miss adjusted Spread Spectrum on it
Unsure if 3rd gen users have access to manually Throttle Temp limit overwrite, 1/2nd gen have~
SSpec was there on x370 5.80 but that bios is broken, X470 they removed it for whatever reason
^ Finetuned like down bellow


Hmm what do we miss, Fan control is working as intended on this cross-port
Yea no PBO and no Performance Enhancer/Bias modes at all in CBS
Unsure if 3rd gen users have everything in AMD Overclocking, i can't test it before Sunday ~ need reports
AMD PBS is quite limited, it misses:
- Cool 'n Quiet for P2 , Pstate OC (Zenstates gladly works but we have no PBO)
- MEM VDDIO
- APU VDDP
- Adjust V1.8
- Adjust VTT MEM
- Adjust VRM SVI2 - VDDCR_CPU
Example:
That's about it
PBO , PState OC for Cool 'n Quiet, Spread Spectrum and the remain PBS stuff is gone

Current Bios Mod
Mirror
Mirror 2

PS:
Idk if it's in your possibility and idk why, but ASRock limited us in MemOC for two parts:
tWRWRSC | tWRWRSD
tRDRDSC | RDRDSD
both are limited to minimum 2 , while it does work setting them to 1 on MSI boards
Example:

Hi guys,

Half an year ago I upgraded my PC with R5 2600 + Dark Rock 4 and Vega56 from 1600x and 1070 because there were very nice deals and I decided not to wait for 3xxx and try Zen+.
Obviously my board is Taichi X370(5.60) and my RAM is CMK16GX4M2B3200C16 (Samsung E-Die).
The problem is that I am not able to run it at 3200 with 2600. With 1600 I just activated the XMP and the memory was running at 3200 without problems despite not being in the Memory QVL list.
Didn't have much time before and these days I decided to give it a try to stabilize it at 3200. Also because with BIOS 5.10 and 5.60 the system doesn't have problems to POST.
I've been following 1usmus's guide with no luck so far.
On the first step I tried the timings from the calculator. There were errors, so tried all three suggested procODT values. Then tried the range of voltages for SOC and DRAM. Couldn't find DRAM boot voltage(from step 1).
at 60 Ohms for procODT there was just 1 error on his stress test, Prime95 ran for 17-18 minutes until worker stopped. Then I tried to change timings from Part 2. For example +1 tRCDRD - the situation got worse, reverted it, changed tRP. At one point the system gave 0 errors, then in AIDA there was error. Add values to other timing on the list or add one more to the same and I get more errors and almost instantly errors in Prime/AIDA. Reverting back to the more stable settings, and the system doesn't behave like before and I have many errors. I tried all the timings from the list, tried also to respect FAW (tRRDS *4 <= best value <= tRRDS *6), tRC = tRAS + tRP. I even tried 1.42-1.44V DRAM voltage even though the calculator suggests 1.355 to 1.375V. Didn't help.
Having it working at 3200 with 1xxx CPU, I think it should be possible to stabilize it at 3200. Also with previous BIOS-es(AGESA) it was a lot worse as it rarely managed to POST at XMP settings.

Am I missing something or I will need a lot more tweaking of the timings to make it work? What would you suggest me to do? I will be watching for good memory deal(second hand or new) to replace it with something easier and certified for the board. But in the meantime I will continue trying.

Attached Thumbnails

 

You can try this 3 presets, RTT_Park of 240 with high Proc0DT is very stressful for the IMC , i think yours are dual ranked too , as they have the identical requirements

This one has to work:

This one would be Fail-Safe, but i forgot the CAD_BUS Timing dividers / you can try&error find them out:

And this would be pushing to 3400:

Start with 1.48v VDIMM & 1.075 SOC (flat loadline)
later you can lower SOC with LLC3 down to 1.05 for 3200MT/s or increase up to 1.125v flat LLC1
Use 20 cycles of 1usmus TestMem_v3 (1:30h for 16GB , 3h for 32GB)
If TM5 errors , please screenshot when and with which error (make it 2-3 errors to know what it actually is)

It depends on your Region, and really the deals
So far i'm quite happy with the perf of the Viper Steel 4000C19-19
the 4133C19-21-21 would be Hynix-C die and the 4400C19-19 are high binned B-Dies
Got the 4000 ones for 130€ , the 4400 where 170€ a bit out of my budget
XCalibur 4000 are known B-dies but expensive
Flare X 3200C14 are hit or miss, always B-dies but rather binning lottery

The remain better ones are Micron-E dies, they have to be good ~ but i miss experience to recommend them

Thank you for the fast response. How did you get these timings? They seem a lot lower than what I am trying to do but maybe the voltages will help.
I was using the TM5 and didn't pay much attention to the type of error. Maybe Test 1 or 2 was at fault when I had only 1 error.
I don't know what means flat loadline? I am using around 1.025V for SOC with LLC3 and 1.365V but HWinfo was showing more aroung 1.385V.

Sadly, I won't be home till Sunday so I will try your recommendations then and will write here if I had luck.

p.s. I live in Bulgaria but buy parts regularly from Germany and even UK. Those Flare X are around 150€ here, the cheaper Viper Steel 4000C19-19 a little bit more. Honestly I don't want to pay that much for RAM and will wait for something on sale or something second hand.

Ah nice to meet you
I have some collegues in the UK, germany the Viper Steels where for 130€ too , even dropped to 110€ 3 days ago ~ because of oversupply
Loadline you see when you hover with your mouse over the CPU Loadline , at the right side is a graph
Flat Loadline means Level 1 , but flat loadline can overshoot - like you saw it using 1.365v
Later when you have b*chy ram (i mean you already do ), or run them very borderline, you'll notice that 0.01v change trows them unstable

How , hmm DRAM calculator is not always 100% accurate, it does help with RTT values, but voltage wise it's only recommending a baseline
Most of the times you want to something further then what the kits are designed for with their rated voltage, soo it can happen that you push 1.42 and higher
* because there is unexplored headroom , you know ^^#
This is some very well made crash-course on memory OC

Hmm, 3200 should run with ease
It's all about setting it right, and too much voltage has negative effects and causes struggle with cooling
Bios 5.6 was buggy but at least it's better then 5.10 on X370
Many here already converted over to X470 to have a bugfree AGESA 1.0.0.3

If you haven't found your max core OC with lowest voltage possible, stay on 1.05v SOC (fixed with LLC1) @ 3000MT/s and lower core voltage as best if possible
(i use OCCT , medium instruction set with AVX2 enabled to speed up testing - 10-15min should be fine to notice if you throttle or even fully error out)
Afterwards add 2 ticks core voltage ontop (if possible or lower core clock) and start to work on ram
TM5, the errors can mean a lot, it depends when it errors and with which number
You'll get a feeling to what means what after time~

@Veii are you using the rear output when you test your IEMs or the front output on your case?
Rear output gives me close to no hiss with 112dB/mW 24ohms IEMs.
From the front output, on the other hand, I can hear it pretty clearly.

Man sorry I am gonna wait until they release the 1.0.0.3 ABBA as it is superior to the ABB version so would be a lot of work for nothing. Send me a PM

Oh, i was away for quite a bit ~ i'm sorry
PM send, we'll figure it out eventually
I played around with cross converting ours to the CH6
Saddly the C6H has the IR35201 in 4+2 mode with the IR3599 using 4+2 doublers
While the Taichi has the IR35201 in 6+2 mode with the IR3598 using 6+2 doublers

Well there is a difference that the CH6 uses own IR3535 Gate Drivers - 8x CPU, 4x SOC
While the Taichi has them integrated in the IR3598 6+2 sSet

To me, pretty much identical ^^#
The rest VRM components it's 1 to 1 the same thing
Saddly not from the start 1 to 1 like the X470 taichi, but i think it may be someday in the future possible to change it a bit , fully converting it to a C6H

Nothing done so far, hanging still on RAMs
Directly from the board - anything in between would be variable