CPU downclocking while testing w/ prime95

Kevin Shelbrock

Reputable
Feb 26, 2015
84
0
4,660
So, im running an AMD FX-6300 3.5GHz(3.8 Turbo) that when turbo is running will peak at 4.1ghz and never see's temp issues (running a Coolermaster hyper evo 212 and it usually peaks at 52 celcius when under load @ 4.1Ghz. However, when i run Prime95 to do a legit stress test it fluctuates between 3Ghz and 3.5GHz. If i turn the turbo off it sits steady at 3.5GHz where its supposed. Why will it not peak to its turbo speed when running prime but does so under any other process?
 
Solution
Try running an older version of p95, like 26.6 The newer versions use abnormally high degrees of AVX instructions among others and this could lead to overly high temps, resulting in the cpu throttling when all 6 cores are maxed.

One thing to remember is that turbo is not the same on all cores. When you see 4.1GHz, thats cores 1-2 only. It'll be 3.9-4.0 for 3-4 and 3.7-3.8 for 5-6. This is done to prevent core overheat during normal full cores usage. P95 is considered extreme usage at 100% on all 6 cores.

Socket temp is kinda inaccurate. It's not exactly the temp of the socket itself, but the area of the mobo around the socket, which happens to be heated by the cpu, cpu cooler, VRM's etc. It's a good indicator of temps at idle, but...
Between temperature protections and other such things, Turbo Core is probably throttling the CPU to try to maintain better temperatures. You can disable Turbo Core and try t manually overclock the CPU instead. Unfortunately I am not too experienced with AM3 overclocking, but you might even be able to set up different frequencies for different usage scenarios.
 


That would make sense if i had never seen higher clock speeds or higher temps outside of prime, but as i stated it only downclocks under prime. I can game on turbo and hit 55ish without it downclocking below the 3.8 turbo. It is only when i run Prime tests.
 


Prime is a stress test, there is no comparing it to gaming load. You are running that Proc at full bore trying to tear it apart with Prime. You would have likely fried the thing if not for the thermal/power protections taking place. There are several things that could be taking place, the most likely are temps above 61 C or voltage drops. Leave Prime alone and don't lose any sleep over it.

 


HWMonitor has the temps at 52 degrees when running prime while cpu is turbo'd, not showing and flux or drop in voltage during the same time frame. The reason im concerned over it is i want to see how far i can push an overclock temp wise, and i cant do that if when i run prime it wont hit the clock speed i have it set for.
 


I've never used HWMonitor, but I've used enough of the others to be skeptical of the readings. AMD temperature monitoring is based on current over time, there isn't actually a sensor and many third party solutions report incorrect low or high temps, hence the presence of adjustable offsets in several suites. In addition the polling rates can be too low on software based monitors to catch the tiny drops in Voltage (you can change this too but it can impact performance). I would suggest trying AMD overdrive to verify your readings just in case as it seems to get it right most of the time. Also realize that many people have reported cores disabling and other anomalies while using Prime on the FX series. At the end of the day what you are encountering sounds like the intended function, my own experiences were similar. Remember you are loading ALL cores with Prime at 100% this is not a normal scenario so you will see different behavior than you are accustomed to.

(edit to add) Note that overclocking requires the disabling of Cool n Quiet, have you tried this yet?
 


No i have not disabled cool n quiet because im not prepared to OC until i can get accurate temperature readings from stock clock speeds or stock turbo. And if you are telling me the CPU temps are not accurate due to a lack of sensor (would the socket temp be a sufficient way to gauge CPU temp?). And as far as AMD overddrive is concerned it shows me within a 10th of a degree the same (and the same 1/100 variance for voltage as well) as what HWMonitor shows ( ran another short run of test to verify). So what your telling me is with an FX processor there is no real way to gauge how safe your being when OC'ing?
 
Try running an older version of p95, like 26.6 The newer versions use abnormally high degrees of AVX instructions among others and this could lead to overly high temps, resulting in the cpu throttling when all 6 cores are maxed.

One thing to remember is that turbo is not the same on all cores. When you see 4.1GHz, thats cores 1-2 only. It'll be 3.9-4.0 for 3-4 and 3.7-3.8 for 5-6. This is done to prevent core overheat during normal full cores usage. P95 is considered extreme usage at 100% on all 6 cores.

Socket temp is kinda inaccurate. It's not exactly the temp of the socket itself, but the area of the mobo around the socket, which happens to be heated by the cpu, cpu cooler, VRM's etc. It's a good indicator of temps at idle, but once a good load is introduced, gets pretty inaccurate. The best method of lowering socket temps under load is adding a large fan behind the mobo. This uses the solder spikes as miniature heatsinks and provides cooling for the other components on the backside of the board.
 
Solution