I have a 3600X coming in tomorrow to replace my awesome but slightly aging 1600AF.
I don't plan on overclocking this go-round as I've read that overclocking the 3xxx series can yield negligible results, and being able to do an all-core clock to 4.4GHz (or even 4.2GHz) can be difficult without pushing 1.4v+. I've also read the millions of posts on people having temp/usage spikes at idle, which I don't really care about to be honest.
I'm mostly just curious if using Ryzen Balanced Plan is still the recommended option for the power plan on 3xxx series? This is a gaming PC first and foremost, so I am fine with temp/usage spikes at idle from HWInfo so I can make sure temps aren't exceeding 80c, which I doubt they will because even overclocked to 4.0GHz my 1600AF under extended P95 load didn't exceed 70-72c, depending on ambient.
I can only say what I have found and what I do.
First, I make sure there is nothing funky going on with the MB. In the case of ASUS, they have "Performance Enhancer" which I set to "default"... It's an okay feature, but as soon as you mix it with PBO temps will raise dramatically.
Second, I use buildzoid's PBO recommendations shown here:
As you can see I DO NOT use the Auto OC.
Third, I use a negative voltage offset for Vcore... this will lower temps, but with the settings above will still boost and boost higher/longer. I will not state a value of this offset, since every setup is different and an unstable undervolt may not show it's face for days or weeks. I have even had to tweak my own offset over the course of a few months when a random crash occurs.
Lastly, I use the latest chipset drivers and the Ryzen Performance Power Plan, since I don't want windows to play with power/voltages. With an undervolt we want to CPU to control the voltages, if Windows starts messing with things we can have drops in voltage that cause random crashes at idle. Even with this precaution, random idle crashes indicate the offset used in step 3 is too low.
I will try these settings after running CBR20 and see if there is a noticeable difference.
I think my gameplan will be as such after CPU installation. I'm going to reset BIOS to default settings to remove any evidence of my 1600AF OC, re-enable XMP, then leave everything else standard.
I'll run 5 rounds of CBR20 all stock, record score/max temps, and save it. Then I'll turn on PBO and repeat, then PBO and AO, then I'll try these custom PBO settings. Whichever one has the highest average clockspeed/score and still safe temps I will see which, if any, of the power plans seem to make a difference in anything. I will post my findings here for fun.
I also did look into undervolting and if temps under stress test get into the mid-high 70c range I plan on dropping back .05v and checking again, though I hear that -.062 to -.075 on some chips is better (but on others of course is unstable).
I kind of wish I never got into working on this crap behind the scenes cause 8 years ago me would have just slammed the CPU in and been happy. Now I feel the need to get every bit of performance out that I can, lol.
So I was going to post the results of all the different modes, but honestly none of them made a huge difference. I didn't go with PBO off, but with it on and set to default, with it on and AO on, and with PBO on with custom settings all yielded results within margin of error, right around 3750 points in CBR20. These all were run using Ryzen Balanced plan but set to 90% minimum instead of 99%. Temps at idle are between 30-40, depending on how active Windows is being. After 15 minutes of CBR20, temps hit a peak of 70c (which is exactly where my 4.0GHz overclocked 1600AF hit), though did tend to stay around 68-69c. Gaming is also pretty reasonable, low-mid 60c range depending on the game, which is also where my 1600AF sat. I did do a voltage offset of -.064 which seems after the 15 rounds of CBR20 and various games and benchmarks to be stable, but time will tell. Plus I still need to run a long-term stress test, but I find that actually running the computer doing daily activities/gaming is more accurate than just pounding the snot out of the CPU with a synthetic test for stability. This was all done with an Enermax 240MM AIO and ambient temps likely in the 72-76F range. I did notice though that it WILL boost to the stated 4.4GHz, but as soon as CPU hits 67c it drops down to 4.2GHz, then at 69c it drops to 4.19GHz. I wish there was a way to change when it started dropping speed, because I'd be fine with it being 70c at 4.4GHz boost. Maybe I will do some more messing around with the voltage offset and see if I got one of those lucky chips that doesn't mind the -.075 offset. It's also worth noting though that even during the CPU test portion of TimeSpy, temps only hit 67c or so, keeping the clock at 4.2GHz. It's only under synthetic "beat me up" tests that I've hit 70c.
Overall I'm pretty happy with the upgrade, though honestly it wasn't WORTH the money. But -I- wanted the more updated CPU. Next upgrade will for sure be 3600MHz RAM as right now I'm running 3000MHz (plus I need to get that RGB RAM to set to white and match better). For reference, all other things equal, my 1600AF O/C to 4.0GHz hit 3000 in CBR20 (stock 1600AF clock was 2700, and just for fun, a 4GHz O/C 2200G on air hit a WHOPPING 1450 xD). Guess this goes to show that either the 1600AF is a beast, or there is only so much you can do with similar architecture and 6/12 c/th.
Then next logical step will be an X570 motherboard and Ryzen 5xxx, .
Anyway, thank you for your help/advice, I did end up using and sticking with the settings you shared and all seems really happy within the system.
I did forget to mention, the best CBR20 result came after also setting MCLK and FCLK to their manually set 1:1 ratio (in my case 30 and 1500 of course). The change wasn't huge, but hey, even 1% of free improvement is nice.
I just wanted to jump back in quick:
For fun, I wanted to see what kind of manual all-core OC I could muster. I would have been happy with 4.2GHz, as once CPU was hitting 69-70c it was dropping just a hair below that.
At +.204V (which brings VCore to 1.26), I (at least so far) have found stability at 4.4GHz! So unless something goes crazy wrong with stability, my new all-core clock is my single-core boost speed.
Oh, and fun fact, temps during CBR20 without OC spiked to 66-67 immediately, and by 3-4 runs of the test were at 70c (and 4.175 GHz). With the 4.4GHz manual all-core OC, it took until pass 15 of CBR20 to hit even 68c. Best score so far was 3912, average seems to be in the 3880-3900 range with this manual OC.
So it's faster, and cooler, let's hope it stays stable. Though at 68c max temp I have a little wiggle room to add in just a touch more voltage.