[SOLVED] I don't understand my OC ?

[Guardianc17]

Hello everyone! I built my PC about a year ago and only recently got into learning about overclocking. I was having trouble understanding what's going on as I think that my CPU is not performing as it should. I followed Graphically Challenged's video (AMD R5 3600 Overclocking Guide MSI B450-A PRO MAX).

I have a B450 Tomahawk MAX motherboard, a Corsair RMx 850 watt psu, 32 GB of ram, and a 1660ti. For cooling, I'm using a Cooler Master 212.

	Overclock 1 (somewhat stable)	OC 2 (no post)	Stock BIOS settings (PBO off, High Performance power plan in windows)
Frequency (in BIOS)	4000 MHz	4200 MHz	Auto
BIOS voltage	1.35	1.4125	Auto
Idle Temp	35-40C	N/A	36-42C
Load Temp	69-72C	N/A	64-66C
Cinebench multicore score	8949	N/A	7883
CPU Z measured voltage	1.36	N/A	1.264-1.372
HWinfo measured voltage	1.337-1.35	N/A	1.256-1.472
Ryzen Master measured voltage	1.099-1.1	N/A	1.29-1.39

NOTE: My 4.2 GHz overclock failed. Couldn't get into windows. I didn't know what to put for idle frequencies at stock settings as Ryzen Master and HWInfo kept jumping around and giving me different values.

My Cinebench R23 scores are nowhere near other people's stock scores. I was only able to get to other people's scores when I overclocked. I read that the average is around 8900-9000 for multicore score.

I don't understand why voltages at stock with PBO disabled run so high. I also don't understand why a CPU that says it can reach 4.2 GHz can't reach (and I mean myself) a stable overclock. Is there something wrong with my CPU/motherboard? Could those high stock voltages be ruining my CPU? How is everyone else able to overclock to 4.1 GHz across all cores with much lower voltages? (even when factoring in the silicon lottery)

From what I understood from other videos and forum posts, we want the voltage to be as stable as possible and voltages over 1.3 are high for a CPU. For my first overclock I saw an almost constant voltage and normal idle/load temps. I believe I also had reasonable temps as those temps only occurred under a CPU stress test. While gaming I would see temps around 55-60C.

I was not able to reach any sort of stable overclock under 1.35 volts. Even if I set it to something low like 3.8 GHz across all cores. I feel like I get more stable voltages/temps when I overclocked it than when I leave my setting at stock.

I also found a new problem where after having my PC turned on all day or when I wake it from sleep, it will no longer hold its OC setup and CPU frequencies and voltages are jumping around.

Thank you for taking your time to read this long post and any advice or insight would be appreciated.

 

[iPeekYou]

It could be you just got unlucky with the silicon lottery. Although my static OC was fine with 1.3v (since removed --too high for my liking), tinkering with its voltage by too much or too little results in either straight BSOD on boot, or BIOS refusing to boot. Bad luck with lottery causes the voltage-frequency curve to steepen much faster than on better binned chips. Anecdotal: I can get 4.18 stable on 1.35v, but 4.2 requires 1.4v to be stable --only just.

Not sure if it helps, but on my board setting the voltage offset too low (over 0,06v IIRC) causes BSOD, too high and it does nothing, on the static OC tab too high voltage causes BIOS to reboot. This also another reason why I didn't bother much with Ryzen OC unless you got at least a silver sample.

Also, Ryzen chugging voltages on idle is "normal". Ryzen boosts quite high on Windows background tasks and uses an ungodly amount of voltage to do that. I saw 1.5v peak at times, both with PBO enabled or not. It's AMD's stock boost algorithm that's causing voltage spikes.

PBO doesn't do too much, it only increases the limit of the boost algorithm (the Precision Boost 2). Namely the Package Power Target (PPT), Electrical Design Current (EDC), and Thermal Design Current (TDC). All three values were taken from the chip when PBO is disabled (or auto --same thing), and can be modified through PBO (auto limits means you're using the motherboard's values, but you can put stupid high numbers manually). Basically, if you're not already limited by any of those three (as can be seen in Ryzen Master), you're not benefitting much (if any) from PBO.

If you are doing manual OC for better performance, I'd recommend against it at this point. Static OC kills Zens, and not to much benefit unless you got lucky with your chip. Otherwise, OC your RAM to at least 3600Mhz, and keep the chip as cool as you can. Sometimes even a 5deg C delta will result in bit higher clocks.

 

[drea.drechsler]

Hello everyone! I built my PC about a year ago and only recently got into learning about overclocking. I was having trouble understanding what's going on as I think that my CPU is not performing as it should. I followed Graphically Challenged's video (AMD R5 3600 Overclocking Guide MSI B450-A PRO MAX).

I have a B450 Tomahawk MAX motherboard, a Corsair RMx 850 watt psu, 32 GB of ram, and a 1660ti. For cooling, I'm using a Cooler Master 212.

	Overclock 1 (somewhat stable)	OC 2 (no post)	Stock BIOS settings (PBO off, High Performance power plan in windows)
Frequency (in BIOS)	4000 MHz	4200 MHz	Auto
BIOS voltage	1.35	1.4125	Auto
Idle Temp	35-40C	N/A	36-42C
Load Temp	69-72C	N/A	64-66C
Cinebench multicore score	8949	N/A	7883
CPU Z measured voltage	1.36	N/A	1.264-1.372
HWinfo measured voltage	1.337-1.35	N/A	1.256-1.472
Ryzen Master measured voltage	1.099-1.1	N/A	1.29-1.39

NOTE: My 4.2 GHz overclock failed. Couldn't get into windows. I didn't know what to put for idle frequencies at stock settings as Ryzen Master and HWInfo kept jumping around and giving me different values.

My Cinebench R23 scores are nowhere near other people's stock scores. I was only able to get to other people's scores when I overclocked. I read that the average is around 8900-9000 for multicore score.

I don't understand why voltages at stock with PBO disabled run so high. I also don't understand why a CPU that says it can reach 4.2 GHz can't reach (and I mean myself) a stable overclock. Is there something wrong with my CPU/motherboard? Could those high stock voltages be ruining my CPU? How is everyone else able to overclock to 4.1 GHz across all cores with much lower voltages? (even when factoring in the silicon lottery)

From what I understood from other videos and forum posts, we want the voltage to be as stable as possible and voltages over 1.3 are high for a CPU. For my first overclock I saw an almost constant voltage and normal idle/load temps. I believe I also had reasonable temps as those temps only occurred under a CPU stress test. While gaming I would see temps around 55-60C.

I was not able to reach any sort of stable overclock under 1.35 volts. Even if I set it to something low like 3.8 GHz across all cores. I feel like I get more stable voltages/temps when I overclocked it than when I leave my setting at stock.

I also found a new problem where after having my PC turned on all day or when I wake it from sleep, it will no longer hold its OC setup and CPU frequencies and voltages are jumping around.

Thank you for taking your time to read this long post and any advice or insight would be appreciated.

What's the HWINfo (SVI2 TFN) core voltage doing? That's the important one to monitor.

At stock the CPU boosts single cores at a time up to max clocks (4.2G). To do that it might raise voltage up to 1.5V to keep it stable, but it monitors hundreds of sensors across the cores and lowers voltage and clocks when temperature or core current gets to high to keep it safe. You can't go that high or it will burn up, so you try to find the lowest voltage you can to run safely. It also never runs all cores at 4.2Ghz simultaneously, although it may appear so in slow monitoring software.

I disagree about PBO not doing too much, but you absolutely need good cooling to see it's benefit as it does raise temperature a lot. The boost algorithm is still working to protect the processor with PBO so as it gets hot it will still want to lower clocks.

You might also need to 'tweak' the settings a bit to get the most of it. Try setting TDC and EDC to 230, PPT to 330, raise the boost clock to the additional 200Mhz and set the PBO scalar to manual and 5x. It may not look much different if monitoring clocks in light useage but it should hold a higher clock at heavy useage as with CB23 multi-thread.

And to run CB23 and get best and most consistent scores to compare to others: run first thing after fresh restart, with nothing else running especially no monitoring software. When you start it open the task manager, the details tab and set it to real-time priority. Then dismiss TM and start the multithread test in single-time run. When it runs it will appear to lock up the system as it has exclusive use of it with realtime so be patient till it finishes.

And whatever you manage to get for a manual overclock, compare it's performance in CB23's single thread test to what you can get with stock and PBO. Single thread performance is important for gaming as games are still light threaded and many times a manual overclock will hurt ST performance.

 

[iPeekYou]

I disagree about PBO not doing too much, but you absolutely need good cooling to see it's benefit as it does raise temperature a lot. The boost algorithm is still working to protect the processor with PBO so as it gets hot it will still want to lower clocks.

You might also need to 'tweak' the settings a bit to get the most of it. Try setting TDC and EDC to 230, PPT to 330, raise the boost clock to the additional 200Mhz and set the PBO scalar to manual and 5x. It may not look much different if monitoring clocks in light useage but it should hold a higher clock at heavy useage as with CB23 multi-thread.

And to run CB23 and get best and most consistent scores to compare to others: run first thing after fresh restart, with nothing else running especially no monitoring software. When you start it open the task manager, the details tab and set it to real-time priority. Then dismiss TM and start the multithread test in single-time run. When it runs it will appear to lock up the system as it has exclusive use of it with realtime so be patient till it finishes.

And whatever you manage to get for a manual overclock, compare it's performance in CB23's single thread test to what you can get with stock and PBO. Single thread performance is important for gaming as games are still light threaded and many times a manual overclock will hurt ST performance.

PBO does some, just not really as good as AMD markets it. The confusion between PB2 and PBO serves no good for us consumers as well.

With my current setup, PBO disabled nets me a good 4.1Ghz (advertised boost clock) across all cores on anything other than OCCT small data set. PBO maxed nets me an additional 50-125Mhz depending on core quality. Not much, but at a cost of noticeably worse temps. CB20 results are within degree of variance, <5%.

Maybe I should rephrase it as Steve Burke said it, PBO doesn't do much if you're not thermally limited. Some gains can be had, sure, but not much over stock behavior. In my case, I am limited by the chip itself with bad silicon lottery and hitting power limits much earlier than thermal.

IMO, the good approach to PBO is the new PBO2 with dynamic undervolt capability on a user configurable curve. Allows the user to trim a bit of temps for longer boost period (increasing distance to thermal and power limits).

I do use the 300/230/230 PBO limits from Buildzoid, and stuck to it. Scalar I left at auto since I find anything over 2x only nets me 25Mhz on all cores. YMMV, some people swear by 5x or 10x.

I do concur that Ryzen OC is finicky and not really worth it. Gains in gaming are very little, and sometimes result in lower 99th percentile/lowest FPS that I argue is better performance metric for processors than comparing average FPS.

 

[drea.drechsler]

....
Maybe I should rephrase it as Steve Burke said it, PBO doesn't do much if you're not thermally limited.
..

I know he's not that much a fan of it, but I got that impression back in the days he was testing Zen 2 with stock coolers because AMD required it. Are you sure he didn't say "...PBO doesn't do much if you are thermally limited" ?

At any rate...I'm running a 3700X with a 240mm AIO... a CM ML-240 so kind of bottom end for 240mm but still pretty good. So running stock, in a super heavy all-thread workload (like P95, small FFT) it will drop to ~3.7Ghz, +/25Mhz as it jumps around. Temps are pretty good..in mid 60's. But just enabling PBO with the settings I mentioned (ppt 330, edc/tdc 230, scalar 5x) it will drop to 4.0-4.05Ghz once it stabilizes in same workload, temps in low 70's now. That's about a 300Mhz improvement by PBO and I'm happy...but there's more!

I mix things up using the EDC=10 bug...set EDC to 10 and disable Global C-States. It gets me an additional 100-150Mhz when stable but temps are now running in the mid 80's. That's obviously a bug (since with a 10A limit on core current it should be throttled) but it's effect is real as it even shows, consistently, with Cinebench 23 and 20 benchmark runs.

So...with or without the bug I am seeing a benefit from PBO. I just think it's how you measure it, and you really do need good cooling to go along with it. And I also think it can depend on motherboard and possibly when the chiplets were diffused. Mine are really early silicon, first on the market, as I got my CPU in August of '19. so the process has matured, newer dies are doubtless much better.

 

[scout_03]

what is your ram name and model also did you activate the axmp for it in bios before you id overclock cpu could you also post a picture of the ram spd page in cpu-z here .

 

[iPeekYou]

I know he's not that much a fan of it, but I got that impression back in the days he was testing Zen 2 with stock coolers because AMD required it. Are you sure he didn't say "...PBO doesn't do much if you are thermally limited" ?

But just enabling PBO with the settings I mentioned (ppt 330, edc/tdc 230, scalar 5x) it will drop to 4.0-4.05Ghz once it stabilizes in same workload, temps in low 70's now. That's about a 300Mhz improvement by PBO and I'm happy...but there's more!

I mix things up using the EDC=10 bug...set EDC to 10 and disable Global C-States. It gets me an additional 100-150Mhz when stable but temps are now running in the mid 80's.

So...with or without the bug I am seeing a benefit from PBO. I just think it's how you measure it, and you really do need good cooling to go along with it. And I also think it can depend on motherboard and possibly when the chiplets were diffused. Mine are really early silicon, first on the market, as I got my CPU in August of '19. so the process has matured, newer dies are doubtless much better.

Snipped a few lines just to make it better for others to read, I get the gist.

I'm pretty sure he said if you're not thermally limited...since his argument was that PBO only adds to power limits basically (unless thermal throttle is manually changed). I was going by AMD's PBO triangle that shows PBO uses some spare power limits instead of gaining from thermal limit (IIRC, both stock and PBO was shown to max on thermal limits).

From what I've seen on my own, I gained extra 100Mhz or so on avg with PBO, same temps (within a degree C) under OCCT just with higher PPT values reported.

It's great that you gained 300Mhz from PBO, I only see the AutoOC benefit really. But I'd chalk it up to the difference between 65W 6c/6t like mine vs 8c/16t since I see 3.9-4.0 all cores on OCCT small with or without PBO. Mine's also an early chip.

Regarding cooling, there's a point where getting better cooler doesn't add much, if any. Tom's own testing shows PBO on stock 3600 cooler actually isn't that far off from PBO with H115i. Then again, I'd argue at least a Hyper 212 is a must these days unless you run 65W chip with Wraith Prism.

Yeah, motherboard and individual chip really do matter. A lot of people got EDC bug working, my board just refuses 10 let alone 1 and kicks me back to default BIOS.

On another note, I think it's better to use amped PBO limits+slight negative offset voltage to get better temps under PBO. My crap bin chip lost 25-50Mhz at most with -0.05v but ran 4-5°C cooler (both under OCCT small and gaming). Any better chip can extract more mileage with that approach.