3400G OC

[Islam Ghunym]

So I got a 3400G with 2 single rank micron E memory kits and ASUS B450-MA mobo.

I got F/U/Mclock stable at 1733 MHz 1:1:1 and latencies to 14-16-11-32-53 and have tightened other secondary and tertiary timings as tight as possible.

disabled CPU boost and PBO and have that CPU runs at 3700 MHz only at less than 25 watts. the plan was to OC iGP only.

processed to iGP frequency to try increasing it as much as possible, but had problem here so here are details.

1-increased GFX to 1.2V

2- tried +20 MHz OC, but didn't get stable OC so increased LLC from level 1 (regular) to level 2 (high)

This caused another crash and it looked less stable (same for extreme LLC level)

3- reverted frequency back to 1400 MHz (no OC) and kept extreme LLC, but GPU kept crashing even on 1000 MHz which made me wonder why??

4- tried higher SOC to 1.2V which does not look safe especially that it has been reported that actual SOC voltage is higher than what system report by monitoring or input value, but here I still have unstable GPU at any frequency

5- now here I have nothing left to do and after hours of checking memory and CPU stability again and all was fine I found that using LLC is causing GPU to crash on load and finally got the iGPU frequency at 1400 MHz no more than that because memory is already OCed to limits at 1.45 DRAM Voltage

6-tried higher GFX voltages 1.225 to see if it will help and I got 1460 MHz.

the problem is at no LLC SOC voltage drops harshly on load from 1.2 to 1.045 which makes the iGP unstable at higher frequencies so LLC has to help as it gets SOC to 1.14V on load rather than 1.045, but it makes the system crash for unknown reason.

What comes to my head is that there are maybe some limitations on power since higher LLC levels will require more power, but there is no access to any of EDC, TDC or other power limits in this particular motherboard.

Ryzen master also does not have options for changing these limits.

If I check Ryzen master monitoring while on heavy stress Graphic load, I can see that these limits are being passed peacefully so now I just want to now why LLC is causing issues rather than helping.

 

[Jurrunio]

GFX voltage is tied to SOC voltage, and the memory controller doesnt like high SOC voltage. For something as memory bound as Vega 11, it makes no sense to chase GPU clocks at the cost of memory speed.

 

Also 3400G is a Zen+ based part, it should have zero user control over FUMclk.

[Islam Ghunym]

Just now, Jurrunio said:

Also 3400G is a Zen+ based part, it should have zero user control over FUMclk.

yes I just wanted to say that I have them synchronized that's all

 

Just now, Jurrunio said:

GFX voltage is tied to SOC voltage, and the memory controller doesnt like high SOC voltage. For something as memory bound as Vega 11, it makes no sense to chase GPU clocks at the cost of memory speed.

 

Yes Yes. I think this is it. so now I have to drop GFX voltage and start from stock and up.

 I will report soon.

[Islam Ghunym]

1 hour ago, Jurrunio said:

GFX voltage is tied to SOC voltage, and the memory controller doesnt like high SOC voltage. For something as memory bound as Vega 11, it makes no sense to chase GPU clocks at the cost of memory speed.

in my case higher GFX voltage is helping to achieve higher stable clocks and if my memory don't run well on higher SOC voltages I would get different results.

I revised memory tests and it seem I conducted them while SOC voltage is 1.094 and 1.187 so I think in my case memory runs fine without errors over nights on this values.

and t seems that it does not show artifacts on GPU memory test and compute as it shows no errors on the CPU and that was at 1.06 SOC voltage during heavy load tests.

I think memory is fine and probably something else is causing GPU to crash on high and extreme LLC. maybe there is problem with engineering this shit from ASUS on AMD APU.

[Jurrunio]

8 minutes ago, Islam Ghunym said:

in my case higher GFX voltage is helping to achieve higher stable clocks and if my memory don't run well on higher SOC voltages I would get different results.

I revised memory tests and it seem I conducted them while SOC voltage is 1.094 and 1.187 so I think in my case memory runs fine without errors over nights on this values.

and t seems that it does not show artifacts on GPU memory test and compute as it shows no errors on the CPU and that was at 1.06 SOC voltage during heavy load tests.

I think memory is fine and probably something else is causing GPU to crash on high and extreme LLC. maybe there is problem with engineering this shit from ASUS on AMD APU.

Or maybe you just picking the worst board possible for the job. that SOC VRM is the weakest of all 2 phase SOC VRM boards.

[Islam Ghunym]

On 6/2/2020 at 1:58 AM, Jurrunio said:

Or maybe you just picking the worst board possible for the job. that SOC VRM is the weakest of all 2 phase SOC VRM boards.

It's not mine. I just got it to overclock and then will give it back to her owner. I wouldn't buy such a crap to OC on, but that's the cheap AMD lol.