Cinebench R23 score is higher when cores run at lower speeds...?

[Ron30]

I've been messing around with Overclocks and testing a Per CCD curve optimized setting vs a curve shaped setting.

 

The Optimized setting seems to give me higher max frequencies across almost every core, but nearly 400 less score in the multithreaded test. Why would this be?

 

Optimized:

 

Shaped:

[RONOTHAN##]

First off, Cinebench isn't exactly known to be the most consistent benchmark in the world, so realistically 400 point is borderline margin of error (I've seen over 1000 points from run to run in R23 at some points). 

 

Second, there is something called clock stretching that exists on these modern CPUs. Rather than outright crash when the frequency gets too high for stability, the system will just drop the clock speed and not report it. Unless you're doing full manual overclocking with static multipliers and such, you will run into this phenomenon to some degree, so realistically you just ignore the clock speed and just pay attention to the performance. 

 

Also, HWMonitor that you're using is known to have issues reading the clock speed on modern CPUs, especially Ryzen. It doesn't look like it's freaking out in that screenshot, but it's very common for it to just give nonsense readings and realistically can't be trusted. Use either Ryzen Master or HWInfo instead. 

[Ron30]

8 minutes ago, RONOTHAN## said:

First off, Cinebench isn't exactly known to be the most consistent benchmark in the world, so realistically 400 point is borderline margin of error (I've seen over 1000 points from run to run in R23 at some points). 

 

Second, there is something called clock stretching that exists on these modern CPUs. Rather than outright crash when the frequency gets too high for stability, the system will just drop the clock speed and not report it. Unless you're doing full manual overclocking with static multipliers and such, you will run into this phenomenon to some degree, so realistically you just ignore the clock speed and just pay attention to the performance. 

 

Also, HWMonitor that you're using is known to have issues reading the clock speed on modern CPUs, especially Ryzen. It doesn't look like it's freaking out in that screenshot, but it's very common for it to just give nonsense readings and realistically can't be trusted. Use either Ryzen Master or HWInfo instead. 

So what's tool should I use to gauge performance then?

[RONOTHAN##]

Just now, Ron30 said:

So what's tool should I use to gauge performance then?

Cinebench is usually fine, just not the most precision. There are other CPU benchmarks that exist out there like Blender, Y-Cruncher, etc., pick one that's a bit more in line with what you actually do on the system and tune from there, running it multiple times to get a bit of an average to hopefully factor run to run variance out somewhat.