Question about Ryzen 3000 Series EDC 1 bug
Question about Ryzen 3000 Series EDC 1 bug
Nice warning, this runs the cpu outside the normal operation intended by AMD. You need water cooling or a decent AIO because your cpu is going to run all cores 4.45-4.65GHz in everything. This is an overclock and no one will guarentee it as safe.
So first the settings.
c20
Single Thread-530 score
Multi Thread -5260 score
Idle temperature CPU-28°C
Max temperature CPU-71°C
Max CPU Core Voltage (SVI2 TFN) reached-1.431v
Settings in BIOS
MCLK/UCLK/FCLK Synced 1:1:1 1933Mhz ,that's DDR4 3866Mhz for those who do not know.
PBO Manual
PPT= 0 actual
TDC=0 actual
EDC=1 actual
Scaler 10X
PBO Override 500Mhz
source
Nice videos of the bug, one ingame and one in c20.
Note that this setup has been run for7 months or so and no reported issues.
Not everyone can expect to get IF 1933, the best I can get is 1917.
EDC 1, 100, or 1000 didn't show any noticeable change. The PBO manual suggests using scalar mode on auto for optimal results. On my system, only a 200Mhz override is feasible; it's better to stick with Auto. I'm applying a negative offset of -0,05v, which reduces idle voltage below 1.4v without affecting benchmark scores. The CPU runs at 5098/511 MHz, but the base remains stable. At 20°C, temperatures stay under 65°C. My RAM performs best at 3600MHz Cl16; it would be slower at 3800MHz due to excessive clock skew. I've tried various settings, but this BIOS lets me set DOCP to 3000 and frequency to 3600MHz with some adjustments. No BCLK changes needed with this MB and BIOS, though scores fluctuate between 99.6 and 99.8. FCLK works well at auto for 1800MHz, but with 3600MHz RAM it's fine at 3600MHz.
It works on my BIOS which is AGESA 1.0.0.3 ABBA. It's exploiting a bug, I think there was a post with it working on the 1.0.0.4 (the original thread is 54 pages long I could be wrong).
Basically I get ~5300 CB20 with my ram overclock. This is approx. 100 points higher than leaving the cores on stock (temperature and room air flow affect the score, highest run is 5211). Time spy cpu hits ~11700 with my ram overclock which is a 4.4GHz all core overclock. This gets you a high all core overclock but keeps the ST boost.
The issue is getting the full single core boost. This is why they try values of EDC from 1-11 and use the highest auto-oc value on their boards. MSI seem to have 500MHz but I have 200MHz.
This is totally use at your own risk but there is one report of this running fine from 8/2019 until now.
Consider factors such as silicone lottery and BIOS/AGESA variations. The last BIOS with 1003abba provided a higher single-core boost of 4.42GHZ but lower performance on other cores. That SC boost was shorter than the current 1004b, which is performing better now. I believe improvements could come if Asus addressed this BIOS issue. 5406 is essentially an updated version of 5220 with AGESA 1004b. Some settings overlap (mainly for PBO), but they don’t seem to have the same impact. BCLK remains fixed and not fully adjustable across all frequencies. DOCP isn’t strictly XMP-compliant, though it does help by lowering Cl from 17 to 16.
The primary issue here is the 110A and 70c restriction for AVX loads, similar to CB r20. This is confirmed by the source thread. It means you can still achieve high performance across all cores, with the same results as a manual overclock at 4.4GHz. Asrock bios designs aren't very appealing in this regard.
The initial post from the forum highlights a specific method to enhance PBO and Turbo Boost performance on certain AMD systems. It outlines adjustments such as manual mode settings, PPT and TDC values, and EDC limits to achieve optimal results. The guide also discusses potential issues with newer BIOS versions and the impact of different processor models, recommending careful testing and consideration of cooling requirements.