R5 3600 + PBO need opinion on my settings

[bat33]

Hello guys,

So ive got R5 3600 watercooled with 280MM AIO, B450 msi gaming plus max, i was manual OC all core 4.2ghz at 1.2v rock stable for many years, temp never above 50C

i recently bought a 6800XT and im facing cpu bottleneck so im planning now to use PBO.

according to what ive see here this is what ive set in bios :

PPT : 95
TDC: 60
EDC: 90

scalar X3

core boost +200

cpu voltage offset -0.0750

auto OC and core boost ON

not using ryzen master, also set power mangement in windows to ryzen balanced

launched a test on cyberpunk where ive got cpu bottleneck , monitored core frequency and some core boost up to 2225mhz, cpu temp around 50C max and cpu voltage never above 1.4v barely under.

am i doing the things right ? could i improve again ? im not sure since core boost already +200 scalar X3 (not sure if i can increase again)
also not sure if the voltage negative offset can reduce performance.

Let me know guys thanks

 

[damric]

Well if it's stable with the +200, you could try sprinkling a bit if base clock. Try 101, 102, ect. If your board can do fractions that's going to be even better. It won't be able to tolerate much before it becomes unstable, but if you can do like 104MHz that will significantly add to your clock speed. Don't be surprised if it doesn't work at all though. Other than that you need to tune RAM and fabric clock.

 

[freeagent]

If it were mine, I would just focus on max static clocks. Set the multi, voltage, and let er rip.

I ran my 3600XT at 4500MHz 1.3375, and 4400MHz, 1.2875.

 

[damric]

If it were mine, I would just focus on max static clocks. Set the multi, voltage, and let er rip.

I ran my 3600XT at 4500MHz 1.3375, and 4400MHz, 1.2875.

lol that's how I ran mine too. For me it was cooler and less power draw to do it that way than letting it boost.

 

[bat33]

If it were mine, I would just focus on max static clocks. Set the multi, voltage, and let er rip.

I ran my 3600XT at 4500MHz 1.3375, and 4400MHz, 1.2875.

well tbh ive run it oc all core 4.2ghz 1.2V for years. rock stable
i wasnt able to get even 4.3ghz tho, even with 1.4v and a 280mm watercooler, not lucky on the silicon.
and recently bought a 6800xt playing at 1440p 144hz, and im facing cpu bottleneck.
i noticed some weird stutter in cyberpunk for example in some cpu bound scenes,
so i tried with pbo and it was way better, no more stutter way more consistent frametime.
can get 4250mhz at best, look like for my case it do a better efficient utilization of the CPU than the all core OC did.
might give as well a better single core performance so make seense for gaming.

 

[Colddecked]

Well if it's stable with the +200, you could try sprinkling a bit if base clock. Try 101, 102, ect. If your board can do fractions that's going to be even better. It won't be able to tolerate much before it becomes unstable, but if you can do like 104MHz that will significantly add to your clock speed. Don't be surprised if it doesn't work at all though. Other than that you need to tune RAM and fabric clock.

I'd skip the base clock oc entirely and focus on RAM and fabric.

 

[bat33]

I'd skip the base clock oc entirely and focus on RAM and fabric.

max FCLK is 1800mhz anything above is not stable best i can do with some 3200CL14 bdie ram is 3600CL 16 16 16 16 32 48, the ram kit could do 3600CL15 easy but for some reason it never been fully stable,
i suspect the 3600 IMC ive tried with two different motherboard and get same result

 

[Colddecked]

max FCLK is 1800mhz anything above is not stable best i can do with some 3200CL14 bdie ram is 3600CL 16 16 16 16 32 48, the ram kit could do 3600CL15 easy but for some reason it never been fully stable,
i suspect the 3600 IMC ive tried with two different motherboard and get same result

If you've done that tuning already and are not satisfied with the performance, you should think about upgrading the CPU. 5600s are pretty cheap.

 

[bat33]

If you've done that tuning already and are not satisfied with the performance, you should think about upgrading the CPU. 5600s are pretty cheap.

Yeah the next one im looking for is def the 5800x3d. There is some nice price on used market but have to be fast.

 

[Ruru]

I use(d) everything on auto with 3600 PBO. Just put that it uses mainboard's limits and +200 more headroom, -0.06V offset for vcore.

 

[Takla2]

Don't bother with pbo on that cpu. a manually locked 4.2ghz is much better. but you won't get that at 1.2v. thats way too greedy. 1.25v SET is much more realistic.

 

[bat33]

Don't bother with pbo on that cpu. a manually locked 4.2ghz is much better. but you won't get that at 1.2v. thats way too greedy. 1.25v SET is much more realistic.

its better for multicore but not for single core and many games depend of single core perfomance

I use(d) everything on auto with 3600 PBO. Just put that it uses mainboard's limits and +200 more headroom, -0.06V offset for vcore.

yeah thats what im doing actually, with scalar X5 and vcore offset -0.1V, full stable i think i can reduce voltage again, cpu run pretty cool like 50°c max in normal usage.

 

[Toothless]

Work on your ram timings to get latency down. Not just the primary like 16-16-16-32 or whatever you got going on. If it needs some more vDIMM voltage could do like 1.35-1.4v. Sadly your chip is just fairly locked down.

If you can get 4.4ghz at like, 1.375v that's about the limit minus pushing more voltage.

 

[bat33]

Work on your ram timings to get latency down. Not just the primary like 16-16-16-32 or whatever you got going on. If it needs some more vDIMM voltage could do like 1.35-1.4v. Sadly your chip is just fairly locked down.

If you can get 4.4ghz at like, 1.375v that's about the limit minus pushing more voltage.

no, never been stable at 4.3ghz even with 1.4v even with a 280mm AIO. also i set all my ram timing manually, worked with some guru3d guy on it and best i can do is 3600CL16 with some bdie kit 3200CL14

 

[Toothless]

no, never been stable at 4.3ghz even with 1.4v even with a 280mm AIO. also i set all my ram timing manually, worked with some guru3d guy on it and best i can do is 3600CL16 with some bdie kit 3200CL14
View attachment 317938

Are you able to get tCL to 14? Everything else looks pretty good.

 

[bat33]

Are you able to get tCL to 14? Everything else looks pretty good.

yeah i just tried, and pc not even restarted, only way was to clear CMOS

 

[Ruru]

yeah thats what im doing actually, with scalar X5 and vcore offset -0.1V, full stable i think i can reduce voltage again, cpu run pretty cool like 50°c max in normal usage.

I have no idea what that scalar does so I've let it as is on my 2nd PC with a 3600.

 

[Toothless]

I have no idea what that scalar does so I've let it as is on my 2nd PC with a 3600.

From Reddit..

PBO scalar adjusts the FIT/FITness/FailuresInTime limit of your part by that factor. Normally, Ryzen CPUs firmware manages the frequency based on the thermals and voltage and it will reduce your frequency if it sees either of those two going into ranges that reduce the long term reliability of the part. Increasing that limit (at the user’s discretion and own risk, voiding warranty, etc.) can let your system sustain higher frequencies without being throttled.

So by default, the scalar is 1x, factory default. If you increase it to 2x, it will double the limit, etc. up to 10x.

More power, more heat, more chance to kill the chip, possibly more performance.

 

[Ruru]

From Reddit..

PBO scalar adjusts the FIT/FITness/FailuresInTime limit of your part by that factor. Normally, Ryzen CPUs firmware manages the frequency based on the thermals and voltage and it will reduce your frequency if it sees either of those two going into ranges that reduce the long term reliability of the part. Increasing that limit (at the user’s discretion and own risk, voiding warranty, etc.) can let your system sustain higher frequencies without being throttled.

So by default, the scalar is 1x, factory default. If you increase it to 2x, it will double the limit, etc. up to 10x.

More power, more heat, more chance to kill the chip, possibly more performance.

Ah. So not worth it for a small frequency increase which practically isn't even noticeable in other than synthetic benchmarks.

 

[GerKNG]

its better for multicore but not for single core and many games depend of single core perfomance

that doesn't mean that a game runs off a single core.
and having a 6 core running at 70-90% load in modern games is probably not on the max SC boost anymore.

 

[bat33]

that doesn't mean that a game runs off a single core.
and having a 6 core running at 70-90% load in modern games is probably not on the max SC boost anymore.

as said before, my stable manual OC was 4.2ghz max, with pbo i can reach 4250mhz, single core manual OC 484 in cinebench R20, 492 with PBO.
and gaming benchmark confirm this, i get better score with pbo than manual OC

Ah. So not worth it for a small frequency increase which practically isn't even noticeable in other than synthetic benchmarks.

i set scalar X5 and a 0.1v negative offset on the vcore, i get slighlty better score with scalar X5 than X1, i always monitor my temps, vcore, TDC and EDC on stress test and its all fine, even with scalar X5 it doesnt go above recomendation, but i would always use negative volltage offset when using scalar more than X1

 

[Takla2]

its better for multicore but not for single core and many games depend of single core perfomance

Bullshit. You can get 4.2GHz locked on all cores with 1.3v, which is more single AND multicore performance then what ever the pathetic boost manages.

 

[bat33]

Bullshit. You can get 4.2GHz locked on all cores with 1.3v, which is more single AND multicore performance then what ever the pathetic boost manages.

Your completely wrong, i did the test and get better score in cinebench single core with pbo vs manual oc at same frequency. Plz spread your bullshit somewhere else

 

[photonboy]

I did a lot of testing with my R9-3900x and ended up just setting the manual whatever multiplier (I think) to end up with 4.225MHz and set a small (negative) voltage offset but I forget what it was. I think it was PBO but I forget that too. I used CPU-Z as a guide while monitoring CPU temperatures. So I'd run the CPU stress test, note the CPU temperature and subract the room temperature to go only by the Delta (i.e. 80deg - 20deg = 60deg Delta). I'd then run the BENCHMARK. If the slight undervolt gave the same or better result at the same or better temp I was happy. I hate, hate, hate Ryzen Master so I just changed two settings in my BIOS.

CPU-Z single-thread I think was about 512.

There doesn't seem to be much overclock headroom on a standard R5/R7 Ryzen 3000 CPU. Silicon lottery yes, but generally speaking not much variation. Pushing the voltage can actually hurt clock speed. Considering a 200MHz, sustainable overclock is at BEST another 5% and that you're roughly at MAXIMUM FREQUENCY near 4.2GHz already likely there's no point in experimenting.

I'd probably just concentrate on getting the voltage to the LOWEST that's stable and can maintain 4.2 to 4.25GHz all core. That lower voltage should reduce temps slightly and by extension keep single-core boosting higher.

*If I'm not getting specific and getting terms a bit off then no worries because after many HOURS I figured out it mostly didn't matter as I couldn't affect a basic overclock one way or the other by more than 2% or so. I could get 4.225MHz easily but 4.3GHz (a max 1.78% gain) was incredibly hard to get to with much, much difficulty. And you should NEVER push too close to instability.*

I've got an old Noctua NH-D14 cooler and my CPU fan spins at about 1000RPM with all cores sitting close to 100% usage (usually fan is under 400RPM in idle). Water cooling seems overkill for an R5-3600.
To keep NOISE down, my CPU fan ramp is:
20% to 60degC,
ramp to 40% at 73degC,
ramp to 60% at 80degC,
ramp to 90% at 83degC,

Basically I set an increasingly steeper ramp which keeps the CPU hotter but keeps fan noise down. CPU's can run hot just fine as long as they are within spec.

Part 2:
*I have an R9-3900x + RTX4070... so our CPU bottleneck should be practically IDENTICAL.

The CPU bottleneck will vary a lot by the game. MOST of the time you won't have a CPU bottleneck. In most games you can at least hit 60FPS at reasonable game settings. For me, there's very few games. Some of the ones hard to hit 60FPS would be games like Baldur's Gate 3 (during the larger city).

You can easily determine IF you're CPU bottlenecked, but you really can't do much about it. You aren't going to gain more than 2% by tinkering with your CPU settings (so don't do it as you may make things less stable). Even if you could hit 4% higher CPU then that gets you from 50FPS to 52FPS in CPU bound situations.

How to figure out CPU bottleneck:
1) run game without any FPS cap (no VSYNC, no other FPS cap)
2) have an OSD to monitor the GPU (not CPU) frequency and Usage
3) EXAMPLE: if the GPU usage is 96% and frequency is near max then there's minimal to no CPU bottleneck
4) EXAMPLE: if the GPU usage is 80% and frequency is near max then you could gain UP TO 25% higher FPS with a faster CPU

You'd really need an R7-5800X3D to make any noticeable dent in CPU bound situations. Ray-tracing tends to favor faster CPU's.

If you don't have a Freesync monitor then that's where I'd put any extra funds.

TLDR: messing with your CPU to try to overclock further is pointless

 

[bat33]

I did a lot of testing with my R9-3900x and ended up just setting the manual whatever multiplier (I think) to end up with 4.225MHz and set a small (negative) voltage offset but I forget what it was. I think it was PBO but I forget that too. I used CPU-Z as a guide while monitoring CPU temperatures. So I'd run the CPU stress test, note the CPU temperature and subract the room temperature to go only by the Delta (i.e. 80deg - 20deg = 60deg Delta). I'd then run the BENCHMARK. If the slight undervolt gave the same or better result at the same or better temp I was happy. I hate, hate, hate Ryzen Master so I just changed two settings in my BIOS.

CPU-Z single-thread I think was about 512.

There doesn't seem to be much overclock headroom on a standard R5/R7 Ryzen 3000 CPU. Silicon lottery yes, but generally speaking not much variation. Pushing the voltage can actually hurt clock speed. Considering a 200MHz, sustainable overclock is at BEST another 5% and that you're roughly at MAXIMUM FREQUENCY near 4.2GHz already likely there's no point in experimenting.

I'd probably just concentrate on getting the voltage to the LOWEST that's stable and can maintain 4.2 to 4.25GHz all core. That lower voltage should reduce temps slightly and by extension keep single-core boosting higher.

*If I'm not getting specific and getting terms a bit off then no worries because after many HOURS I figured out it mostly didn't matter as I couldn't affect a basic overclock one way or the other by more than 2% or so. I could get 4.225MHz easily but 4.3GHz (a max 1.78% gain) was incredibly hard to get to with much, much difficulty. And you should NEVER push too close to instability.*

I've got an old Noctua NH-D14 cooler and my CPU fan spins at about 1000RPM with all cores sitting close to 100% usage (usually fan is under 400RPM in idle). Water cooling seems overkill for an R5-3600.
To keep NOISE down, my CPU fan ramp is:
20% to 60degC,
ramp to 40% at 73degC,
ramp to 60% at 80degC,
ramp to 90% at 83degC,

Basically I set an increasingly steeper ramp which keeps the CPU hotter but keeps fan noise down. CPU's can run hot just fine as long as they are within spec.

yes thats what im actually doing, PBO with a negative voltage offset 0.1v, fully stable, with 280MM AIO, im like 50°c max in game and 65°c max in heavy benchmark like P95 or even y cruncher.

i think PBO get the advantage to use the better core. new bios feature called CPPC prefered core, basically it say to windows which core to use for a better efficiency, thats the big advantage vs manual OC,

i strongly believe PBO make a better efficient use of the cores and its not only about max clock.

And yeah im facing CPU bottleneck in some heavy CPU games like cyberpunk, sometimes CPU is 85% used and gpu like 70%, with a 6800xt, at 1440p 144hz monitor, freesync.

and yeah the 5800X3D is def the one im looking for