Question 3950x - Is Ryzen Master or CPU-Z Misreporting Idle Clocks?

[Ian_UK1]

New build - Fractal Define R6 USB-C case, Ryzen 3950x, Be Quiet! Dark Rock Pro4 cooler, Gigabyte Aorus x570 Elite with latest BIOS, 128GB Corsair Vengeance 3000, RTX 2060 Super, Samsung SSD. O/S is Win10 Pro; fresh install with the latest drivers from AMD and using Ryzen Balanced Power Plan. BIOS is at stock settings apart from XMP 'Profile 1' selected for memory.

The machine will be used for graphics work and rendering, won't be overclocked.

The 'problem'? At idle, both Task Manager and CPU-Z report clocks for all 16 cores at 3400MHz, whereas Ryzen Master shows all cores in sleep mode at about 100MHz! One (or more) of them must be wrong! Idle temps are about 35 deg C. with the Dark Rock's fans barely turning, so I'm guessing Ryzen Master is the tool reporting the clocks correctly? I'd be really grateful if anyone can offer any insights as to what might cause such a big difference between what Ryzen Master is reporting and what Task Manager / CPU-Z are reporting.

Thanks in advance for any suggestions you may have.

 

[Hitman928]

Trust Ryzen Master for idle conditions.

 

[IEC]

The processor has advanced hardware power management features and will sleep/idle at very low power states. In addition, it changes states so quickly that monitoring tools aren't that great outside of load situations.

If you do benchmarks and the performance is as expected AND you are getting proper idle in Ryzen master with cores at sleep, you have a properly configured system.

My 3950X at idle (ECO mode enabled):

Spoiler

 

[Ian_UK1]

That looks eerily familiar...!

 

[DrMrLordX]

When in doubt, use a tool like HWiNFO64 to check power draw at idle to see what it's really doing

 

[Ian_UK1]

Thanks for the suggestion. I'll install HWiNFO64 and see what it tells me. Ryzen Master does have a CPU Power Meter though - 4th one along in the screenshot above - that is showing a 1.3 - 1.4 W CPU power draw at idle, in my machine, with all cores in 'sleep' state. Interestingly, the SOC power draw (5th meter along) is 5x that of the CPU at idle. I'm guessing 'SOC' in the case of the Ryzen, means the I/O chiplet?

 

[DrMrLordX]

Thanks for the suggestion. I'll install HWiNFO64 and see what it tells me. Ryzen Master does have a CPU Power Meter though - 4th one along in the screenshot above - that is showing a 1.3 - 1.4 W CPU power draw at idle, in my machine, with all cores in 'sleep' state. Interestingly, the SOC power draw (5th meter along) is 5x that of the CPU at idle. I'm guessing 'SOC' in the case of the Ryzen, means the I/O chiplet?

Yeah all the Ryzen 3000 CPUs with an I/O die will burn some power @ idle which is why Renoir doesn't use an I/O die like that. And yeah Ryzen Master also has a power meter. I just find reading the percentages to be a bit annoying.

 

[Ian_UK1]

I've now had a chance to check this using HWINFO64 as suggested earlier - same result. When Ryzen Master shows all cores in sleep mode at 100Mhz (or thereabouts) HWINFO64, together with CPU-Z and Windows Task Manager, reports all cores running at Base Clock.

I've also tested this on another Ryzen 3000 build I've just completed - this time a 3600X installed in an Asrock B550M Pro4 with latest BIOS. All BIOS settings are currently at default, except for XMP being enabled. Again, i'm getting exactly the same results - when the system is idling, Ryzen Master reports all cores asleep at circa. 100MHz, whilst Task Manager, CPU-Z and HWINFO64 report all 6 cores at Base Clock (3800MHz or thereabouts). Idle temp (using stock Wraith Spire cooler) is only 33C, so it appears, once again, that Ryzen Master is reporting the clocks correctly.

Has anyone else encountered this misreporting of Ryzen clock speeds by Windows Tack Manager and others? Is it just 'normal' for Ryzen-based computers, or have I maybe missed a setting somewhere? Apologies for all the apparently dumb questions, but I'm new to building Ryzen-based systems!

 

[DrMrLordX]

HWiNFO64 reports a constant 3750 MHz on my 3900X.

 

[AmericanLocomotive]

You need to look at the HWiNFO64 "Effective" clock speed on the sensor's page.

The default AMD Zen 2 power-plan keeps the CPU at near base clocks at all times. The CPU then shuts-down unused cores, and causes partially-loaded cores to skip clock-cycles. So while the CPU is still internally running at whatever frequency, the cores are effectively running at a much lower speed.

 

[DrMrLordX]

You need to look at the HWiNFO64 "Effective" clock speed on the sensor's page.

I think that's a feature on the newer versions. But yes it is there. It actually provides data on a thread-by-thread basis (T0, T1 per core).

 

[ondma]

When in doubt, use a tool like HWiNFO64 to check power draw at idle to see what it's really doing

This is what I was about to suggest. Power draw should clearly tell you if the cpu is powering down or not.

 

[Ian_UK1]

You need to look at the HWiNFO64 "Effective" clock speed on the sensor's page.

The default AMD Zen 2 power-plan keeps the CPU at near base clocks at all times. The CPU then shuts-down unused cores, and causes partially-loaded cores to skip clock-cycles. So while the CPU is still internally running at whatever frequency, the cores are effectively running at a much lower speed.

This is really interesting. I'd thought AMD must be using different 'mechanisms' to control effective clock speeds compared to Intel. I guess this means that the 'real' clock speeds are actually being reported correctly by both CPU-Z and Task Manager, but as both are unaware of clock cycles being skipped, neither are capable of reporting the effective speed?

I hadn't seen the HWiNFO64 'Effective Clock Speed' as I'd only had a small amount of time to familiarise myself with the software. Will check it out when I've a spare minute or two.

 

[Martimus]

This is really interesting. I'd thought AMD must be using different 'mechanisms' to control effective clock speeds compared to Intel. I guess this means that the 'real' clock speeds are actually being reported correctly by both CPU-Z and Task Manager, but as both are unaware of clock cycles being skipped, neither are capable of reporting the effective speed?

I hadn't seen the HWiNFO64 'Effective Clock Speed' as I'd only had a small amount of time to familiarise myself with the software. Will check it out when I've a spare minute or two.

I remember hearing that AMD goes into and out of power states much faster than Intel. Orders of magnitude faster on Zen 2. None of the tools were able to track this, and many actually woke up the cores when they polled them.

Of course this all came out when the chips were released last year, so I can't remember the specifics. I would trust Ryzen Master over older tools, since it is designed for these chips.