Please Help - Ryzen 5950X PBO Boost Clock Override

[VortX]

Hi all,

I am having trouble getting Precision Boost - Boost Clock Override to work. I have set the core clock override to +5MHz, +200MHz and even +300MHz but nothing seems to have an effect. I also tried setting the scalar to 10x, EDT/PPT/TDC to 'Motherboard', manually tweaking them and also tried tweaking curve optimiser. No matter what I try, the clock will never boost over the factory 4.9GHz single-thread (testing with Cinebench R23). It also has no effect on multi-thread testing. I also do not appear to go above 5.0GHz, at all, during any test (I know some people do even on stock).

I should note separately that setting PBO to Motherboard does have an effect and multi-threaded clocks jump to around 4.4GHz and Cinebench scores increase substantially.

I should also point out that I have absolutely no trouble exceeding factory clock speeds by manually overclocking, it's just the PBO clock boost that does not appear to work.

I am also running on a custom water cool loop and temperatures are fine. With PBO enabled I am not exceeding 60C single-thread and 77C multi-thread.

I currently have a Gigabyte Aorus Master X570 with F31 BIOS. I am aware that PBO2 is not supported until AGESA 1.1.8.0 but apparently the Master mobo has the necessary code to make it work (PBO2 is indeed listed in settings).

Anyway, why the heck can I not get my CPU to clock anymore than factory when using PBO clock boost? It's driving me nuts...

Thanks,

 

[VortX]

Beginning to think I have a duf chip... I ran Ryzen Master Auto OC and it sets the maximum to 5.15GHz but it just won't go above 5GHz... this is baffling.

 

[Curlyriff]

@pete910 any comments as same board I think right? Is that same behaviour?

As Pete knows I have had issues all over place with my build tbh and have almost given up. I have full watercooled loop and haven't worked out what the issues are.

However my board (Gigabyte Aorus B550m Pro) it is limited by TDC, EDC & PPT at default (95A, 140A & 142w) giving me a peak 3.99Ghz whilst actually pulling 100w CPU power @ 60c. PBO on gives me about 4.25Ghz all core, this seems to be mostly down to much higher limits with 200A, 250A & 250w limits but it hits EDC at 100% and peaks at 85c pulling 167w.

Single core speeds are identical with it default and PBO because I can't see anything limiting it. I get the same 4.94Ghz peak in both default and PBO. So considering the difference in performance to the power/heat I just leave it default and have assumed it just a really duff chip.

 

[VortX]

Dup post.

 

[VortX]

Ok so it looks like I have made some progress. While it would seem that in some respect, I don't have the best of 5950X's when it comes to overclocking, I have found a means to get it over 5GHz. I spent the evening playing with curve optimizer. It's a very tricky thing to truly ensure stability due to the nature of the tests you need to run - stress testing doesn't cut it, you also need to test low-demand workloads.

Anyway, it seems my CCD0 has a lot more headroom than CCD1. I am currently testing CCD0 with a negative offset of -15 and all appears well so far. I will then move to per-core optimisations. CCD1 on the other hand, it is nowhere near as flexible with even the smallest of negative offsets causing issues. This appears to be holding the overall CPU back.

I will say however that although Curve Optimizer has helped me get over 5GHz, I am still not able to get anywhere that would force the boost overdrive to kick in; not for any single core. I have managed to get 5040MHz so far. I don't see myself ever getting to 5100-5200MHz like others unfortunately.

I will report back when I have finished optimizations and see if I can squeeze anymore out.

 

[pete910]

It's down to luck with CPU i feel, I too can only hit 5050 on 2 cores. Multi highest is 4600 ish

I have tried to set things manually and it ends up performing worse. Left all on auto with a CO of -15 seems to work fine for me.

 

[Doug2507]

Nvm.

 

[VortX]

It's down to luck with CPU i feel, I too can only hit 5050 on 2 cores. Multi highest is 4600 ish

I have tried to set things manually and it ends up performing worse. Left all on auto with a CO of -15 seems to work fine for me.

Strange, I am finding a similar ceiling - 15! Still, a long way to go before I find what the max I can push each core is though.

A shame really, invested heavily in a custom loop system and I end up with a ‘duff’ chip. Ah well, it’s still performing exceedingly well and even with my lower clocks I am hitting higher benchmarks than those with higher cores. Guess my low temps are keeping things consistent.

 

[philo-sofa]

This new fangled PBO stuff is a bit opaque; wish I could take time to sit down and understand what's happening. I get all core overclocks of ~4.5 GHz on PBO auto, and fewer threaded apps see up to 4.8-4.9. But I don't really get what's going on or how to optimise it and that's sub-optimal.

It's a bit like being in 2005 with no explanation for what an FSB is, so I empathise VortX but there's not much more I can add at present.

 

[VortX]

Well gents, it seems curve optimiser is the secret sauce. With CCD0 now optimised and stress tested, I am hitting over 5GHz on every core of CCD0 with multi-threaded and single-threaded scores up. I am even hitting over 5.2GHz on some cores (with clock overdrive finally having some headroom). With CCD0 in the bag, I will be moving onto CCD1 over the coming days. I’ll post again with my final results

For those interested, my current settings are as follows for curve optimiser (per-core, negative offset)

C0: -30
C1: -23
C2: -29
C3: -30
C4: -30
C5: -29
C6: -28
C7: -18

Other than that, PBO power limits are set to ‘Motherboard’ and clock overdrive is set to 200MHz. Everything else is left in auto.

What’s strange is that the official AGESA code supposedly only supports up to 30 increments but the Aorus Master let’s you exceed that. So, for the cores I am already at -30 I may even try going further. Fun times ahead!

 

[DarkBahamut]

IMO you definitely haven't tested enough for stability yet. When I tested curve optimiser it wasn't uncommon to see random BSODs after several days of uptime/gaming.

Because of how precision boost works it's completely impossible to test with any workload in a short period of time. One day it will decide to boost to a particular clock because of that instant workload/temps/etc and it will be just the wrong side of stable.

You can test all core drops more quickly for additional testing, but dialling in specific values like -28, -29 would require multiple days minimum for each individual core before you could really say it's properly stable.

 

[VortX]

IMO you definitely haven't tested enough for stability yet. When I tested curve optimiser it wasn't uncommon to see random BSODs after several days of uptime/gaming.

Because of how precision boost works it's completely impossible to test with any workload in a short period of time. One day it will decide to boost to a particular clock because of that instant workload/temps/etc and it will be just the wrong side of stable.

You can test all core drops more quickly for additional testing, but dialling in specific values like -28, -29 would require multiple days minimum for each individual core before you could really say it's properly stable.

Quite possibly which is why I am testing various scenarios. So far I have found the most reliable program for uncovering stability issues with curve optimiser is PC Mark, it puts the PC under all variations of day-to-day tasks which is perfect and often caused crashes when nothing else would. For example, Cinebench found C7 to be stable at around -25. PC Mark crashed until I tweaked that to -18.

So for each core I have run:

A per core boost tester.
PC Mark benchmark.
3D Mark benchmark.
Cinebench multi core run (multiple times)
Cinebench single core
OCCT 1 hour stress test
Leave to idle/background tasks for 10-20 mins.

Finally, to test over a longer period I am running prime95 for a minimum of 8 hours to test the CCD as a whole.

If after all of that I run into any weird BSOD or crashes then I will just reduce the curve by 1 and continue testing normal usage.

 

[BigBANGtheory]

I had a play with curve optimiser and PBO last week and my thoughts are, wait for Ryzen ClockTuner 2.1 or at least a decent guide that steps though exactly what to adjust and when.

For me (and it maybe my chip or my lack of understanding/experience with all the bios settings) I get a ton of instability as soon as I negatively offset the curve even by -5. So I can see boost examples of 5GHz when stress testing but that is no use when Windows crashes 5min later when the system is idle.

 

[VortX]

I had a play with curve optimiser and PBO last week and my thoughts are, wait for Ryzen ClockTuner 2.1 or at least a decent guide that steps though exactly what to adjust and when.

For me (and it maybe my chip or my lack of understanding/experience with all the bios settings) I get a ton of instability as soon as I negatively offset the curve even by -5. So I can see boost examples of 5GHz when stress testing but that is no use when Windows crashes 5min later when the system is idle.

Probably good advice for those that don't like to experiment in the wild west. I personally find it a fun learning experience though. I would state based on my experience over the past few days however that a blanket curve optimiser across all cores is not really that helpful. It is better to tweak cores individually as silicon lottery plays a big part. Secondly, yes, do not rely on 'stress testers'. You need tools that test all ranges of workloads across all cores. PC Mark is the best I have found so far.

 

[DarkBahamut]

None of the testers are much good. The problem you get that is in single thread loads the boost clock is held down. So if you see peaks of 5100 or 5200 MHz, you'll find under a sustained heavy single thread load like Cinebench that the effective clock is more like 200 MHz lower. The system will be stable under that held down boost clock, but when idling the system will take the opportunity to do a full 5200 MHz boost for a few milliseconds but the undervolt won't take it and the system will BSOD. Anything that just heavily loads 1 core, or all cores is only testing one point in the entire curve.

Games will work somewhat better as their varied load levels allows the CPU to boost more and use a wider variety of clockspeeds so you get a more complete picture of how stable the system is.

 

[VortX]

None of the testers are much good. The problem you get that is in single thread loads the boost clock is held down. So if you see peaks of 5100 or 5200 MHz, you'll find under a sustained heavy single thread load like Cinebench that the effective clock is more like 200 MHz lower. The system will be stable under that held down boost clock, but when idling the system will take the opportunity to do a full 5200 MHz boost for a few milliseconds but the undervolt won't take it and the system will BSOD. Anything that just heavily loads 1 core, or all cores is only testing one point in the entire curve.

Games will work somewhat better as their varied load levels allows the CPU to boost more and use a wider variety of clockspeeds so you get a more complete picture of how stable the system is.

Agreed. It is very difficult to test. That said, PC Mark has been brilliant for finding problems. Especially the document/spreadsheet tests, they have thrown the most crashes for me. My theory is that if I can run every possible test, including normal workload and idle tests... if it does randomly BSOD at some point I will just up the power increment across the range and see how it behaves from there and just repeat the process until I no longer have any problems. Hopefully someone comes out with an effective ‘stress’ tester though.