Userbenchmark CPU throttling

[mmCharles]

Recently, when I complete a Userbenchmark run there's a notification that my CPU is being throttled. The only thing that's changed is a slightly higher overclock from 3.82GHz to 3.95GHz with a small bump in voltage from 1.35v to 1.38v. I've tested both AMD's and the high performance power plans, but the notification is still there. I've never had this come up on the results page before, any ideas why it could be showing now?

 

Temps:

Idle = 30c

Load = 57c (max)

[Projex]

Maybe undo the overclock, or check out the fan

[Tabs]

11 minutes ago, mmCharles said:

Recently, when I complete a Userbenchmark run there's a notification that my CPU is being throttled. The only thing that's changed is a slightly higher overclock from 3.82GHz to 3.95GHz with a small bump in voltage from 1.35v to 1.38v. I've tested both AMD's and the high performance power plans, but the notification is still there. I've never had this come up on the results page before, any ideas why it could be showing now?

 

 

A small bump in voltage can have a large effect on temperature. Voltage will tend to increase power consumption (and generally, heat) exponentially, whereas frequency changes increase temperatures linearly.

 

Combining a higher overclock with higher voltage compounds the effect; if you were already close to the capacity of your cooler, you could easily be running much higher temperatures now, causing throttling.

 

Basically, revert your oc, run the benchmark and record your temperatures. Then, run the benchmark again with your OC and voltage again, and record the temperatures. You may be surprised by the difference.

[mmCharles]

 

21 minutes ago, Tabs said:

A small bump in voltage can have a large effect on temperature. Voltage will tend to increase power consumption (and generally, heat) exponentially, whereas frequency changes increase temperatures linearly.

 

Combining a higher overclock with higher voltage compounds the effect; if you were already close to the capacity of your cooler, you could easily be running much higher temperatures now, causing throttling.

 

Basically, revert your oc, run the benchmark and record your temperatures. Then, run the benchmark again with your OC and voltage again, and record the temperatures. You may be surprised by the difference.

 

Sorry, I forgot to include temps. Idle is between 29c to 31c, under load caps at 58c

[johndms]

1 hour ago, mmCharles said:

 

 

Sorry, I forgot to include temps. Idle is between 29c to 31c, under load caps at 58c

What are you using to monitor your temperatures? Regardless, try leaving HWMonitor open while you run userbenchmark and look at the Package Temperature reading afterwards.

[mmCharles]

3 hours ago, johndms said:

What are you using to monitor your temperatures? Regardless, try leaving HWMonitor open while you run userbenchmark and look at the Package Temperature reading afterwards.

 

I’m using HWMonitor, those were the readings. Maybe there’s a bug in Userbenchmark? Novabench and Cinebench seem to run fine. 

[johndms]

2 hours ago, mmCharles said:

 

I’m using HWMonitor, those were the readings. Maybe there’s a bug in Userbenchmark? Novabench and Cinebench seem to run fine. 

It's possible. I believe Aida64's System Stability Test displays any throttling that may occur. It'd be my next step in verification. Just run a 10 minute test stressing CPU/FPU/Cache with HWMonitor open keeping an eye on Package Temperature. I can't see it happening at such low temperatures, however. Ryzen is supposed to throttle at 95c.

 

It seems we have nearly the same setup. I'd be interested in seeing a screenshot of your HWMonitor's line showing CPU VDD (accurate reading of vcore) during a stress test. My board has horrible vdroop. At 3.9GHz, I'm forced to raise my core voltage in the bios to 1.4125v to be completely stable. Anything less and Aida64 will report a failure after about 30mins in some tests to an hour or two in others. During load, that 1.4125 becomes 1.325 (-0.0875), which is nearing my own recommended safe core voltage for 24/7 use (1.425v).

[mmCharles]

25 minutes ago, johndms said:

It's possible. I believe Aida64's System Stability Test displays any throttling that may occur. It'd be my next step in verification. Just run a 10 minute test stressing CPU/FPU/Cache with HWMonitor open keeping an eye on Package Temperature. I can't see it happening at such low temperatures, however. Ryzen is supposed to throttle at 95c.

 

It seems we have nearly the same setup. I'd be interested in seeing a screenshot of your HWMonitor's line showing CPU VDD (accurate reading of vcore) during a stress test. My board has horrible vdroop. At 3.9GHz, I'm forced to raise my core voltage in the bios to 1.4125v to be completely stable. Anything less and Aida64 will report a failure after about 30mins in some tests to an hour or two in others. During load, that 1.4125 becomes 1.325 (-0.0875), which is nearing my own recommended safe core voltage for 24/7 use (1.425v).

 

 

I followed your advice and ran a 10 minute test with Aida64. I normally use Intel Burn Test to test stability of my overclocks, so this was new to me. The 69c max in HWMonitor was a random spike, temps stabilized at 61c. I have a feeling the throttling message is just a bug in Userbenchmark at this point.

[johndms]

18 minutes ago, mmCharles said:

I followed your advice and ran a 10 minute test with Aida64. I normally use Intel Burn Test to test stability of my overclocks, so this was new to me. The 69c max in HWMonitor was a random spike, temps stabilized at 61c. I have a feeling the throttling message is just a bug in Userbenchmark at this point.

Yeah, I ignore the temperature spikes as well. If you click Preferences in Aida64, you will probably have the option to enable "CPU Diode" sensor. That will correspond to Package Temperature in HWMonitor.

 

So this is with 1.38v set in the bios? You're dropping to as low as 1.281, so we both have similar vdroop. After seeing this, I'd be interested in seeing if you can do that same test for more than 1 hour. When I verify my overclock stability, I let that same Aida64 test run overnight (6+ hours). It all depends on how SURE you want to be about it being stable. I've had tests fail at 30 minutes, some at 2 hours, some at 4 hours. I've even heard of some who verify for 24 hours, which is a bit much in my opinion.

 

Your temperatures seem great though and I don't see any throttling. You can do a Google Image Search for "Aida64 throttling" to see what it'll look like. And anything below 80c is perfectly fine for an overclock. My temperatures at 3.9 averaged around 64c. So I'm in agreement with you, it sounds like just a bug with userbenchmark.

[mmCharles]

17 minutes ago, johndms said:

Yeah, I ignore the temperature spikes as well. If you click Preferences in Aida64, you will probably have the option to enable "CPU Diode" sensor. That will correspond to Package Temperature in HWMonitor.

 

So this is with 1.38v set in the bios? You're dropping to as low as 1.281, so we both have similar vdroop. After seeing this, I'd be interested in seeing if you can do that same test for more than 1 hour. When I verify my overclock stability, I let that same Aida64 test run overnight (6+ hours). It all depends on how SURE you want to be about it being stable. I've had tests fail at 30 minutes, some at 2 hours, some at 4 hours. I've even heard of some who verify for 24 hours, which is a bit much in my opinion.

 

Your temperatures seem great though and I don't see any throttling. You can do a Google Image Search for "Aida64 throttling" to see what it'll look like. And anything below 80c is perfectly fine for an overclock. My temperatures at 3.9 averaged around 64c. So I'm in agreement with you, it sounds like just a bug with userbenchmark.

Yeah, everything's set in the BIOS. I was under the impression the voltage drops were some kinda of efficiency measure by AMD? Even at stock speeds the voltage dips a lot. I don't tend to do anything crazy with my system, other than gaming, streaming and editing so I don't normally test for longer than 30 minutes.

[johndms]

3 minutes ago, mmCharles said:

I was under the impression the voltage drops were some kinda of efficiency measure by AMD?

I've never heard of that. I know that most boards have a bios option called Load Line Calibration (LLC) which attempts to counter vdroop under load. With 1.4125 in bios, if I had LLC capabilities, that voltage would only drop to 1.40, for example, instead of 1.325. Some extreme LLC settings will actually push the voltage higher.

[mmCharles]

34 minutes ago, johndms said:

I've never heard of that. I know that most boards have a bios option called Load Line Calibration (LLC) which attempts to counter vdroop under load. With 1.4125 in bios, if I had LLC capabilities, that voltage would only drop to 1.40, for example, instead of 1.325. Some extreme LLC settings will actually push the voltage higher.

 

I'm probably talking out of my ass in all honesty. I'm fairly new to overclocking. All I've done is set a multiplier on the core and upped the voltage until stable under load for 30 minutes. I briefly looked into LLC, but our boards don't have any options to change it.