Question PBO - doing it wrong?

[TM1172]

I recently had to RMA my R9 5900x due to it spontaneously dying (still not sure what happened). The new CPU showed up yesterday and I installed it with a new cooler (Dark Rock 4 Pro).

Now I’m experiencing an interesting thing. I turned on PBO in Ryzen Master and didn’t change any settings. I ran Cinebench and scored around 21570. I ran TimeSpy and got about a 13000 CPU score. My temps were looking sort of high (83-85 in just the one multi core pass of Cinebench) so I ran P95 with HWInfo open and saw temps go almost immediately to 90. Clocks started dropping, down to around 4.15 on all cores.

Adjusting TDC/ppt/edc didn’t help with temps too much but brought clocks back up to about 4.55 on all cores.

Curve optimizer with -21 on all cores didn’t drop temps significantly or at all.

Where it sounds weird to me - Turning PBO off (OC mode: Default in Ryzen master)but keeping core optimizer on dropped temps nearly 20 degrees and resulted in a perf bump. Cinebench was lower by maybe 100 points but TimeSpy increased by 500+ points. My clocks averaged 4.85-4.9 on all cores. P95 now tops out at 88c.

Can anyone tell me what I’m doing wrong here? I though PBO was a “turn it on and go” thing that would boost performance but I’m seeing the opposite. I’ve followed a few guides on how to use it and I’m confused as heck on what is happening with this chip, especially as with PBO on, the cpu was drawing about 180w and outpacing the DR4P cooler, which was supposed to be one of the best air options for this cpu.

 

[drea.drechsler]

...
Curve optimizer with -21 on all cores didn’t drop temps significantly or at all.

...

-21 is too much on some cores...and probably not enough on many others. The gold star cores are usually stable around -10 or -12 while other cores can go all the way to -30. Finding the max per-core setting takes a bit of time but it gives best results. Get a tool like Core Cycler to help with stability testing.

Even a fully optimized CO doesn't always result in lower temps since the processor will still try to push to 90C in heavy workloads, it's the way the algorithm works. But it's doing it on slightly lower voltage at any given clock speed. That's a difficult thing for us to see with the monitoring tools we have available to us but you can see it when the system is mostly idle.

Usually, PPT and TDC around stock settings and EDC slightly reduced from stock settings for best results, that may be why PBO "off" worked for you. Pushing TDC, and especially PPT, to higher settings rapidly increases temps and makes the CPU uncoolable. There's simply not enough surface area to get the heat out of dies so it just pulls clocks back earlier and more. This is where sub-ambient cooling becomes helpful but that's not practical for the average user.

Don't be greedy with additional boost setting. Going for +200MHz and higher EDC setting might give you 5.05Ghz boosts in light bursty workloads on your gold star cores (all cores if you get per-core CO set up right) but they heat up quickly and pulls back on clocks sooner, and more, in heavy (Cinebench) workloads. The same with PBO Scalar settings. For my 5800X I find +25 or 50Mhz and a scalar setting of x2 is about optimum for Cinebench scores. But then a 5900X is dual-die so has somewhat different properties.

Uninstall RyzenMaster and do it through BIOS. It works better. RM is a tool for extreme overclockers who have a difficult time restarting when running on LN2.

 

[Karadjgne]

The only 2 things PBO does is raise the power limits and with such can add upto 200MHz to the boost. Using Curve Optimizer, you lower voltages and power use in general, so if that's lower than the stock settings, PBO isn't going to change anything, even with the 200MHz clock boost, because you aren't hitting that power limit in the first place. In other words, stock socket limits for that cpu is 142w. Enable PBO and that limit is raised to closer to 200w. But if you only use 130w, having a 200w limit or 142w limit makes Zero difference.

It's a Ryzen, not an intel. They do not behave the same way. An intel will boost to the limits you set, regardless of temps and voltages unless that boost hits the thermal limit, only then will it downclock, to save its own skin, regardless of performance.

Ryzens will boost according to workload, power and temps, but start downclocking certain cores by small amounts (20-50Hz generally) in order to keep temps in check and supply best possible performance. A Ryzen will start downclocking @ 80°C slowly, an intel waits until @ 100°C then chops the clocks like an angry butcher.

Your Prime95 was using AVX technologies, so it's hitting @ 130% power usage vs cinebench using closer to 100% average. Which is why Prime95's temps were higher.

I'm thinking there's something not right with the cooler install. A 5900x shouldn't be hitting those temps with a DRP4. Did you paste the cpu correctly? Or just use the 'accepted' Intel pea method?