Gigabyte B450 I AORUS PRO WIFI - Manual OC for SFF build - p states?

[testyoldpower]

I posted this over on the Gigabyte forums, but I haven't had any responses, so I thought I would try here.
====================================================================================================

I just recently put together a small form factor build in a Silverstone Mile ML06-E case with a Ryzen 5 3600 on the B450 I AORUS PRO WIFI motherboard.

Components:
- Ryzen 5 3600 @ 1.1V / 4.175GHz
- B450 I AORUS PRO WIFI (F50)
- 16gb (8gb x 2) Corsair DDR4-3200 CL16 (running at 3600MHz, 18-19-19-42)
- Noctua NL-L12S
- 1TB Samsung 970 NVMe / 2TB Samsung 860 QVO
- Gigabyte LP GeForce GTX 1650 4gb

I put together the above system as my personal computer. I do some light gaming, but mainly am using it for standard productivity work/browsing. At stock settings, the system ran pretty hot and the fans were constantly racing whenever I did anything. I addressed the fan racing by adding a custom fan curve.

For the cpu thermals, I was interested in looking at using the Ryzen 3000-series Eco-mode, but I wasn't able to get it to show up in Ryzen Master (tried enabling PBO like mentioned in various posts). I also messed with Vcore offset voltages, but the cpu still seemed to run very hot in benchmark testing. I ended up picking a voltage (1.1V) that gave me good thermal performance and just bumped up the all-core frequency to find the max stable point (4.175GHz @ 1.1V). I am really happy with the final performance and temperatures, but due to the manual overclock I don't seem to be getting the Vcore/frequency drops during idle that I could when running the cpu as stock.

I have read that folks have been able to address this by using p state overclocking. Is this possible to do on the B450 I AORUS PRO WIFI (F50 BIOS)? It isn't obvious to me in the BIOS.

Ideally, I would like end up with a system that runs at a max voltage (like 1.1V) to keep the cpu temperatures down, but be able to run lower voltages during idle. Any help or ideas are appreciated.

How I have the system now is definitely great and will support what I want to do with it...but I am a tinkerer and would like to understand what's possible.

CPU temps Passmark Cinebench r20

 

[trents]

Do you have two 80mm fans installed on the side of the case?

What are your temps actually like at idle and under load?

You might need to lower the CPU core frequency and the voltage with a negative offset.

 

[testyoldpower]

I don't have the 80mm fans installed. They were rubbing against the gpu shroud. I doubt they would really do much good anyhow since they are just blowing into the gpu fans. The PCB for the gpu blocks any air from getting to the cpu (very small case).

Temps at idle are between 33C and 45C usually. The load temps are linked at the bottom of my original post. I am currently running Manual OC @ 1.1V/4.175GHz all-core. I didn't have great results (or didn't know how to use correctly) when I tried using a negative Vcore offset.

 

[trents]

With that case, isn't there a 120mm fan on top? Is it set to exhaust or intake? Could you attach a pic or two of the assembled system, one with the case panels in place and one exposing the interior of the case so that we can see the layout? Without the 80mm fans installed, it is hard to imagine any kind of air flow through the case. Seems like an APU might have been a better choice.

 

[testyoldpower]

The case does come with a 120mm fan in a multi-purpose bracket. You can use that with a sub-45mm cooler (I believe, going off memory) or remove it and fit a cooler up to 70mm. I did that (Noctua NL-L12S), so the fan is gone. The only fan is the cooler fan and it is set up as exhaust. It is the chosen method for this case based on what I have read from others. It really is a balancing act between case dimensions and heat dissipation. I am not able to take pictures of the system at this time, but I will try to get some. It is super tight, but I am able to hide away the cables pretty well.

 

[trents]

With the rig in my sig I had to set the first smart fan bump up level to start at 60c to keep the fans from constantly ramping up and down in light duty computing.

 

[testyoldpower]

Yeah, I am doing something similar now. I think I set it to 55% from 0 to 55C and then linear to 100% by 80C.

 

[trents]

Let me know how it works for you. I spent a lot of time tweaking the fan curve on my machine. The problem is that even with default stock settings in place these Ryzen CPUs spike vcores around 1.5 volts when they turbo. Just momentarily, mind you, but long enough to heat up a core or two such that the temps spike higher than you expect at idle and in light duty computing.

 

[testyoldpower]

Yeah, I know what you mean by the spikes. Currently, I am running a fixed voltage and frequency (4.175GHz @ 1.1V) and the fan curve is great. If I find a setting that allows me to run without fixed voltage and keep the thermals in check, I will give you more feedback on the fan curve.

 

[trents]

Thank you. You might want to look at LLC and experiment with it.

 

[testyoldpower]

Thanks for the suggestion. I am reading through this post and guide now. Do you know any other posts/guides to read up on for LLC that are worthwhile? I understand what the principles are with LLC, but I am not certain how I am suppose to best use it for power savings rather than max frequency. Do most folks use it in conjunction with Vcore offset voltages or with manual OC?

 

[trents]

Honestly, I can't think of anyone using this forum who got satisfactory results using offset voltage. It seems to just complicate matters, even when it works as intended.

As you probably know, the idea behind LLC is to fight voltage droop under load. It helps prevents the vcore from dropping below what is needed to remain stable under load. Without it, you have to enter a higher vcore amount than you need to be stable under idle and casual computing scenarios. Theoretically, the LLC offset only kicks in under load and allows you to have lower vcore during idle times and everyday casual computing.

The amount of vdroop under load you will experience is dependent on the quality of your motherboard's VRM component and how well it is cooled, as well as the power demands of the CPU you are using. On my system, when I am running a stress test with LLC on Auto, I'm getting a voltage droop of about .125, from 1.45 down ot 1.325. If I set my LLC to High, the voltage droop is only about .075. That is still a lot but it is a cheap B450 itx motherboard and allows me to be stable at higher all core overclocks. Ideally, the LLC should level should be (could be) set to a level to give 0 voltage droop under load.

Your CPU is built on a 7nm fabrication process die so your thermals should be better than mine and your vdroop should be less than mine. My 2700x is built on a 12nm fab process.

 

[testyoldpower]

Thanks, that is precisely what my understanding of LLC was as well. But I guess that I don't really understand how to determine what the optimal LLC to use is. I tried only varying the LLC setting in the BIOS and testing with Cinebench r20 (MT) and monitoring with HWMonitor. I am not seeing any noticeable difference. Perhaps this is because I am running a Manual OC?



I re-ran the 'Normal' LLC again at the end because I believed that all of these results were likely the same and that how long between when I ran the tests/how much the components cooled down was probably causing the small changes in numbers.

 

[trents]

I would suggest using HWInfo64 to monitor your temps and voltages. It's much better suited to the Ryzen architecture than is HWMonitor and gives more details. Having said that, it looks like your motherboard bios may just bypass LLC altogether when you set frequency and vcore manually or at least freezes the LLC value. You are only drooping about .038 volts. Much less than my rig.

Try leaving the vcore on auto while setting frequency manually.

 

[testyoldpower]

Thanks for the tips, will keep messing with it and try HWInfo64 instead. I did try setting a fixed Vcore and leaving the Manual OC setting for frequency at 0. It ran without boost at all (3.6GHz). Ideally, something like that is what I want though, setting a max voltage for the CPU to use and let the frequency land where it may. I have already seen through manual OC testing that I can run at 4.175GHz at 1.1V and 4.2GHz appears stable at 1.15V. I was hoping to just somehow indicate that the max voltage I ever want to use is say 1.2V. And then let AMD boosting algorithms do what they want for frequency with that capped voltage. I haven't really figured out how to do that.

 

[trents]

How granular is your manual voltage setting? I can only adjust in .0125 increments. For instance, the next lowest preset manual voltage from 1.4225 is 1.41. Board manufacturers are having to remove a lot of information from the 16 mb eproms on lower end B450 boards in order to accommodate the AEGESA coding for the 3xxx series CPUs. The new standard is 32 mb eproms.

 

[testyoldpower]

In the Advanced Voltage menu, the adjustments are 0.00625V, so half of what you are seeing. In the Overclocking menu, manual OC, I can just type in a number in mV. I am not sure how granular it will accept though.

Most of my original testing I did through the initial Advanced Frequency/Voltage menu, setting 1.1V and 4.175GHz for the settings that I found to be the best at the time. Since reading the overclocking guide that I listed above, I have moved back to setting the frequency in those initial menus to Auto, and the Voltage to Normal, and am applying numbers in the Overclocking menu under Manual OC.

According to that guide, what I was doing might enable clock stretching rather than giving me the results I was expecting. I don't believe that I saw that, but thought I would rather be safe.

 

[trents]

So I've been experimenting with offset voltage and LLC this morning. Offset seems to do nothing but add or subtract the declared offset value from what bios sets vcore to automatically. I cannot use both override and offset at the same time as when I use offset mode the line item to set the voltage manually (i.e., override is grayed out). As it turns out, however, LLC is the tool in my bios that gives me more granularity in vcore. I have 8 setting levels I can choose from in LLC.

I envy you in being able to run at 4.175 ghz on only 1.1 vcore. I need 1.369 net voltage (after vdroop) to be able to run Cinebench R20 without crashing. Moving from 12nm fab to 7nm fab makes a huge difference. What are your mosfet temps under load?

For my setup, I have concluded that default bios values (except using AXMP for the memory) give me the best combination of performance and temps. At stock I get a boost on three cores to 4.35 ghz under moderate loads. With something like Cinebench R20 that really taxes all the cores, that drops to about 3.9 something on all cores. But for daily driving the 4.3+ boost is kind of nice.

 

[testyoldpower]

What are your mosfet temps under load?

The VRM MOS went to 51C and VSOC MOS went to 48C during the Cinebench run where the CPU hit 75C. Just looked back through a trial HWInfo64 log that I ran when you recommended I switch. I haven't run a log with Prime95.

4.175GHz @ 1.1V is right at the edge. 4.2GHz @ 1.1V rebooted about 7 mins into Prime95. And I tried 4.25GHz @ 1.15V and it crashed out on the 2nd run of Cinebench. I am not really sure how this lines up to other folks 3600s since I am really trying to focus on the best performance at a reasonable power level so my system can handle it. Most of the time, folks that are overclocking have a hardier cooling and airflow system to allow for more brute force overclocking.

 

[trents]

I your situation, lack of air flow will cause a steady buildup of heat in longer stress tests. As long as it's stable in how you use it then that's what counts.

Here's my CB R20 stress test info with everything at stock default and memory at 3200 mhz. One thing to keep in mind is that the "X" Ryzens are binned higher, made possible by a tad better silicon than their non X counterparts but mostly just clocked higher from the factory (with corresponding higher voltages) I think.