Do you mean the FX-9370 is not a "220W processor" because you have managed a successful undervolt? Actually its TDP is 220W, which means it does not necessarily need 220W, but peaks there eventually. I bet it uses even more than 220W if used on a 970 Performance at BIOS defaults, because voltage is so high (1.535V), like overclocked.

 

Apart from this, AM3+ FX CPUs are known for rather ineffectual undervolting results and high overall power consumption. Congratulations you've managed and hopefully it lasts!

 

To keep within TDP, AMD uses its LLC (Load Line Calibration) technology to lower voltage the more load it gets. That is why it is possible to keep it within TDP specs and/or 60° C.

This should be the voltage drop you have experienced. So no Offset should be needed. Was it unstable that way? It definetly shouldn't be. Btw. this board lead me more than once to wrong conclusions.

Another AMD thing is APM (AMD Power Management). You will notice that at full load (like Prime95) the clock rates begin to fluctuate between say 4400 and 3500 MHz. This allows for even lower voltages at full load, while maintaining a certain performance level.

 

If LLC and APM is active, there should be no need of extra cooling components like VRM - other boards prove that. It's an imposition anyway if users have to experiment with undervolting to get it right. What about users that are not capable as you are? A decent case like Fratal Design should be fine without extra VRM cooling to manage FX-9000 chips - mine do. But the real issue here is, as you have discoverd too, wrong default BIOS settings that lead to improper voltages used by the CPU! I don't remember seeing LLC in there... is Offset ASrock's interpretation of LLC? It should be there and it should be set to Regular, not Medium, High, Extreme or Auto to keep everything cool!

 

However, I still can't see the difference between your and my understanding of Offset Voltage. The drop is desired by AMD at full load and if not implemented right, you get a too hot CPU and way too hot VRM.

 

- I can't speak for the FX-9590, but the FX-9370 it's definitely not a 220W processor - except at BIOS defaults it will likely be.

- The +50mv offset is being misinterpreted. The only thing it does is to mitigate the (insane) voltage drop this board experiences when switching from idle to full load. The only reason OP experienced higher temperatures when increasing the offset is that the effective voltage provided to the CPU was increasing (since vdroop was reduced).

- Hence, the true problem is the 1.535 default vcore that leads to the 220W rating. This ensures the FX-9370 will work stably (provided heat is taken care of), but it is way overkill.

Therefore, the solution for those set on using the FX-9370 on the 970 Performance is: undervolt that CPU! Your mileage may vary, but in my experience you can drastically reduce vcore without experiencing stability issues, and highly improving the thermal environment.

In order to do that, you need to remember that vcore or cpu offset themselves don't drive the temperature under load: what really matters is the actual voltage delivered to the CPU by your vocore/offset combination. In my experience, temperatures are a bit better if the effective vcore is stable rather than bouncing around (which, unfortunately, is very common with this board).

Hence, I recommend lowering the vcore and trying which combination of vcore+offset works better for you (as a general rule, you need to combine higher offsets with lower vcores to achieve the same effective vcore during load).

Another thing to take into account: I found that RAM speed is also an important determinant of VRM/CPU socket temperature and vcore stability (I guess it's due to the IMC).

Now, my results so far:

- With +0mv offset, I could get the FX-9370 to work with 1.4125 vcore. Effective vcore was of the order of 1.32v, and not very stable. 1.4v would now be stable.

- With +50mv offset, I could go to 1.3875 vcore. That would translate to a quite stable 1.33-1.34v delivered to the CPU at load. My temps for VRAM (measured with IR gun and a temp sensor attached to the back of the MB) were lowest with this setting, and CPU socket sensor would barely approach 55C.

- It seems I could go lower, although delivered voltage would become less stable, and no temperature improvement was observed. However, at this point I realized RAM was defaulting to 800Mhz. So, time fix that.

- I was using G.skill sticks rated at 1866 MHz, so I aimed for 1866. Temperatures went higher and delivered voltage less stable as I did this, and it would become unstable before 1866. However, I didn't need to adust vcore: increasing CPU NB (i.e, the IMC) voltage one notch to 1.15v was enough to get everything to run stable. Now the effective voltage dances around more, the VRM temperature (back of the MB) is in the low 70s, and the CPU socket temp reaches 66C under load. 

 

ASRock actually did answer to one of my tickets some weeks after my last post here, offering me to send it in, but by then my goal was to get rid of it.

So I've returned the mainboard to my retailer for the second time. Waiting about 7 weeks for a decision of its wholesaler, I've got my money back.

I was mislead (again) by several factors:

 

.) DIMM slots' channel linkage seems to be the other way round than usual

This is what happended:

 

I've switched back to the original two 8GB Radeon DDR3-2400 modules. BIOS is at default values (C6 setting is Disabled, therefore Turbo inactive!). RAM is automatically set to 1866 MHz (CL13) and 1.65V.

 

 

Another strange thing is the DIMM slot linkage and "corresponing" Command Rate timing. Usually channel A (red slots, nearer to CPU) should be filled first, prior channel B (black slots). If channel B is filled first, it reacts like 4 filled slots, which usually leads to a bit slower Command Rate of 2T.

In my case, when using channel A, the Command Rate was 2T automatically and no POST if 1T was selected manually. In the black slots, which are supposed to be channel B, 1T was set automatically and works stably!

 

So far so good!

 

The next phenomenon is the "Overclock Mode". If I change it from Auto to Manual (leaving reference clock at 200), RAM is automatically set to 800 MHz and values above 1600 MHz are unstable or not bootable, as described initially. So I have to leave it at Auto. Although it is not my intention to OC this system, it degrades future possibilities.

 

[EDIT]

 

 

This Crucial kit is rated for 1.5V at 1866 MHz (while the AMD kit needs 1.65V for 2133/2400 MHz) and runs at that setting on the other mainboard.

 

Your words, parsec, about the memory controller's voltage gives me the idea of a possibly wrong voltage setting for the CPU-NB! If it hasn't enough juice it would react like this. I'm going to check this out soon!

 

Thanks for your input!

 

 

Crucial Ballistix DDR3-1866 (2x8GB) with XMP. Independet of activating the XMP profile, it just works at 1333 MHz or below. If I select 1600 or above, the system hangs with a black screen. I have to turn it off and on again, wait some time until the BIOS recovers with default settings.

 

This is annoying, because both, the AMD Radeon and Crucial Ballistix Kits, run flawlessly on my Sabertooth boards. And why shouldn't it work, the memory controller sits within the CPU!

 

I could live with the boisterous voltage settings and undervolt the system, but buying another RAM kit is out of the question.

 

What could I have done wrong? Should I open an official support ticket?