cancel
Showing results for 
Search instead for 
Did you mean: 

Processors

michael3700x
Journeyman III

3700x erroneously under-reports socket power usage and overclocks itself when in stock

I have a Ryzen 3700x (stock) running on Asrock x570 Phantom Gaming-ITX/TB. 3700x is known to have 88W PPT limit at stock settings. When I run the popular Cinebench R20 benchmark multi-core, it runs fine and resulting score is also fine (actually, a little too high for stock 3700x but more on this below). But the problem is Ryzen Master shows CPU socket is consuming only 52% of 88W budget which comes out to about 46W. Hwinfo64 monitor tool also shows a similar number of 48W for "CPU Package Power". This 46-48W number is significantly lower than expected for 3700x running this benchmark because everyone with 3700x reports maxing out (or close to) the PPT limit of 88W budget for this benchmark.

I believe the socket power usage reported by Ryzen Master (and Hwinfo64) is way lower than the actual power usage. In fact, if you look at the Hwinfo64's individual core power numbers, each core is consuming about 12W. Total core power usage would be 12W x 8 cores = 96W. Add SoC power usage and the socket power comes out to somewhat above 120W. I realize Hwinfo64 is not to be trusted blindly as some are known to be inaccurate. But this number jibes much more closely with my wall power meter reading after accounting for all components of my system and PSU efficiency. Unusually high CB benchmark score (4950) also jibes with a scenario where CPU is at higher clock speed than at stock.

If this was a mere telemetry reporting error, I wouldn't have come here to report this. The real issue is the CPU seems to be using this erroneous power number to overclock itself ("Hey, I'm only using 46W, let's boost to 88W!") consuming much more than 88W expected of its stock settings. This affects users' cooling solution and perhaps silicon health and data integrity. Overclocking, intentional or not, is generally frowned upon by people who prioritize power efficiency and data integrity over maximum clock speed.

Other info: I ran this same CPU on couple other motherboards past few weeks but did not have this problem. The reason I came here instead of Asrock is because I think this erroneous socket power number is estimated and used by the CPU rather than the motherboard.

I still enjoy my new AMD CPU and the new build. I hope you guys can take a serious look at this. Please let me know if you have any question or need more data, I will be happy to oblige.

0 Likes
2 Replies
brucer
Forerunner

try using cpuid hwmonitor and see what it says..

0 Likes

Same bad readings with hwmonitor. This number originates from CPU SMU block. It's clear its calculation is broken. Unfortunately it's also affecting core clock speed management. I hope it's not damaging my silicon. But I do see memory stress test runs failing with this buggy stock behavior. Enabling "Eco-Mode" in Ryzen Master brings down CPU to normal stock behavior in terms of wall power usage, peak multi-core clock during CB, CB score, etc. Memory test runs okay with this setup. EDIT: I'm running my DDR4 at XMP 3200 MT/s (stock timings) which is officially supported by both AMD and the memory maker.

0 Likes